United States
Environmental Protection
Agency
Office of Water
(4305)
823B16002
December 2016
EPA Guidance for Conducting
Fish Consumption Surveys
-------
Guidance for Conducting Fish
Consumption Surveys
Office of Science and Technology
Standards and Health Protection Division
Office of Water
United States Environmental Protection Agency
Washington, DC
December 2016
-------
ACRONYMS AND ABBREVIATIONS
ABS
Address-Based Sampling
AI/AN
American Indian and Alaska Native
ASA24
Automated Self-Administered 24-hour Recall
AWQC
Ambient Water Quality Criteria
BRFSS
Behavioral Risk Factor Surveillance Survey
CATI
Computer-Assisted Telephone Interviewing
CAPI
Computer-Assisted Personal Interview
CBPR
Community-Based Participatory Research
CDC
Centers for Disease Control and Prevention
CDS
Computerized Delivery Sequence
CERCLA
Comprehensive Environmental Response, Compensation, and Liability Act
CFR
Code of Federal Regulations
CRITFC
Columbia River Inter-Tribal Fish Commission
CTUIR
Confederated Tribes of the Umatilla Indian Reservation
cv
Coefficient of Variation
CWA
Clean Water Act
Deff
Design Effect
DDT
Dichlorodiphenyltrichloroethane
DFW
Department of Fish and Wildlife
DHQ
Diet History Questionnaire
DNR
Department of Natural Resource
DQO
Data Quality Objectives
DSF
U.S. Postal Service Delivery Sequence File
EPA
U.S. Environmental Protection Agency
FCR
Fish Consumption Rate
FCST
Fish Consumption Survey Tool
FDA
U.S. Food and Drug Administration
FFQ
Food Frequency Questionnaire
FHCRC
Fred Hutchinson Cancer Research Center
FNDDS
Food and Nutrient Database for Dietary Studies
FWC
Fish and Wildlife Conservation Commission
g
Gram
H
Desired Half Width
ICR
Information Collection Request
IHS
Indian Health Service
IRB
Institutional Review Board
Kcal
Kilocalorie
L
Confidence Interval Length
MSM
Multiple Source Method
N
Number of Completed Responses
Minimum Sample Size
NAAL
National Assessment of Adult Literacy
NCC
University of Minnesota Nutrition Coordinating Center
NCES
National Center for Education Statistics
NCI
National Cancer Institute
i
-------
NDS-R
Nutrition Data System for Research
NEJAC
National Environmental Justice Advisory Council
NHANES
National Health and Nutrition Examination Survey
OHRP
Office for Human Research Protections
OMB
Office of Management and Budget
oz
ounce
P
Probability
PBT
Persistent Bioaccumulative Toxics
PCB
Polychlorinated Biphenyl
PII
Personal Identifying Information
PIN
Personal Identification Number
PIPS
Post-Interviewing Processing System
PSU
Primary Sampling Unit
PWD
Parks and Wildlife Department
RDD
Random Digit Dialing
RFE
Regulatory Fish Encyclopedia
SE
Standard Error
SIDE
Software for Intake Distribution Estimation
SPADE
Statistical Program to Assess Dietary Exposure
ssu
Secondary Sampling Unit
TEK
Traditional Environmental Knowledge
TWRA
Tennessee Wildlife Resources Agency
UFCR
Usual Fish Consumption Rate
USD A
U.S. Department of Agriculture
11
-------
GLOSSARY
Access Point Survey: A survey that is administered at locations where fishers gain entry to fishing
or hunting areas. Examples include boat ramps, docks, and wildlife refuge check stations.
Accuracy: A measure of agreement, expressed numerically as a percentage, between a measured
value and an accepted or true value.
Address-Based Sampling (ABS): A type of sample that uses addresses selected from the U.S.
Postal Service Delivery Sequence File (DSF). In address-based samples, the address is the unit
selected.
Anadromous: Migrating from salt water to spawn in fresh water
Angler: A person who fishes with hook and line and net.
Area Probability Sample: A type of sample which usually includes clustered or a multi-stage
sample selection within defined geographic areas.
Bias: Property of a statistical estimation process that consistently overestimates or underestimates a
population parameter. The degree of bias is represented by the discrepancy between the expected
value of an estimate and the true or actual value of the population parameter being estimated.
CATI: Computer-Assisted Telephone Interviewing, a method of telephone interviewing in which a
structured questionnaire is programmed into a computer. The interviewer enters the respondent's
replies directly into the computer program.
Census: A complete enumeration of a population.
Commercially Caught Fish: Fish caught or harvested for commercial profit.
Confidence Interval: The range of values within which it is estimated a population parameter lies
with a defined level of confidence, based on sample data.
Confidence Level: The probability that a population parameter lies within a given range.
Consumer-Only FCR: Estimates of fish consumption rates that are for the subset of the study
population (and the larger population they were selected to represent) that consume fish. They are
calculated utilizing data from those study respondents that reported fish consumption, excluding
those who did not. The time period for defining consumer-only can vary — 24-hour recall, 30-day
consumption, any consumption in a year, etc., and this choice, along with the statistical
methodology, affects the estimate of FCR. (See Per Capita.)
Creel Survey: A survey generally employed by fishery managers to collect data from anglers on the
types of fish caught, the size of the fish, time spent fishing, and other data related to both the fishing
trip and the catch.
111
-------
Descriptive Statistics: The branch of statistics that involves summarizing, tabulating, organizing,
and graphing data to describe a sample of objects or individuals that have been measured or
observed. Descriptive statistics typically include means, medians, standard deviations, percentages,
and other statistics that illustrate magnitudes, diversity, and frequency.
Frequency Distribution: A tabular or graphical presentation of the number of times each value
occurs in a data set.
Fish Consumption Rate (FCR): The quantity of fish consumed per unit of time, e.g., grams per
day (g/day).
General Population: All individuals in a geographic area, without reference to any specific
characteristic.
Heritage Rate: A heritage rate is the amount of fish consumed prior to non-indigenous or modern
sources of contamination and interference with the natural lifecycle of fish, in addition to changes in
human society. While it is often thought of as a historic rate, it can also be a current unsuppressed
rate.
Inferential Statistics: The branch of statistics that involves making inferences about the value of
one or more population parameters, on the basis of sample statistics. The most common
applications of inferential statistical procedures are estimation and hypothesis testing.
Intercept Survey: A survey conducted in-person at a specified location of interest. The location
may be specific to the population of interest. Selected participants are interviewed at the selected
locations.
Iowa State University Method: A statistical program (Software for Intake Distribution Estimation
-SIDE) to estimate the distributions of usual intake of nutrients, foods consumed almost daily, and
other dietary components produced by researchers in the Department of Statistics at Iowa State
University in 2001. (http:/ /www.side.stat.iastate.edu/)
List Sample: A type of sample drawn from a list frame. List sample frames consist of households or
persons who possess a characteristic unique to the list. Examples of list frames include persons with
fishing licenses, tribal registries, etc.
Measurement Error: The difference between a measured value of a quantity and its true value.
Measures of Central Tendency: Descriptive statistics that identify the center or middle of a
distribution. Common measures are the mean and median and, less commonly, the mode.
Measures of Dispersion: Descriptive statistics that identify the spread of values of numerical data.
Common measures are the standard deviation, variance, and range.
Multimode Data Collection: A data collection protocol that utilizes more than one method for
contacting and interviewing respondents. Multimode data collection methods are used to maximize
survey response while controlling costs.
IV
-------
Multiple Source Method (MSM): A statistical program to calculate usual dietary intake from 24hr
recall data and supporting data developed by the Department of Epidemiology of the German
Institute of Human Nutrition Potsdam-Rehbriicke and is available for on-line analysis.
(https:/ /msm.dife.de /)
Multivariate Analysis: The analysis of data consisting of multiple variables and examination of
associations among variables (e.g., regression and correlation analysis, analysis of variance and
covariance).
NCI Method: A statistical method to estimate usual dietary intakes of foods and nutrients
developed by the National Cancer Institute, National Institutes of Health, USA, and researchers at
other institutions. This method can be used to estimate the distribution of usual intake for a
population or subpopulation; assess the effects of individual covariates on consumption; and predict
individual intake for use in a model to assess the relationship between diet and disease or other
variables, (http: / /appliedresearch.cancer.gov/diet/usualintakes /method.html)
Oversampling: A process by which a specific group or target population is selected at a higher rate
than other population members, in order to ensure estimates can be calculated for the specific
group.
Per Capita FCR: Estimates of fish consumption rates that are for the entire study population (and
the larger population they were selected to represent). These estimates include those respondents
who rarely, or perhaps never, consume fish. (See Consumer-Only FCR.)
Precision: A measure of uncertainty in an estimate, such as a standard error, coefficient of variation
(CV), or the width or half-width of a confidence interval. A specified value of the precision is
commonly used as a target in designing a survey, including specifying the survey's sample size.
Following completion of the survey, each major estimate derived from the survey is usually
presented along with a measure of its precision.
Population Parameter: A population parameter is a summary number such as a mean or percentile
that describes the entire population. Examples of population parameters are: the mean fish
consumption rate of the population is 22.4 grams per day or 45 percent of the population is 15 years
of age or below.
Random Digit Dialing (RDD): A method used to select samples for telephone surveys by random
selection of telephone numbers within working exchanges. This method permits coverage of both
listed and non-listed telephone numbers.
Random Error: A source of error that contributes variability (reduces precision) but does not
influence the sample mean or median. It is a type of measurement error.
Response Rate: A rate indicating the proportion of cases completing specific components of a
survey. In simplest terms, the number of completed cases divided by the total sample.
Recall Bias: A response error resulting from a respondent's inaccurate recollection of particular
events.
v
-------
Rolling Cohort Method: A survey method that involves randomly placing survey participants into
groups (cohorts), which are then sequentially surveyed over equally spaced intervals (e.g., 2 or more
weeks). Each cohort is asked to provide recall data for a period of time equal to the interval spacing
between cohort surveys. This method is typically used to provide coverage over an entire year while
avoiding the problems associated with long recall periods.
Roving Creel Survey: A creel survey that is conducted by having the interviewer move through the
survey area in a random or defined pattern to contact fishers.
Sample Size: The number of cases (usually households or persons) selected from a frame to be
interviewed or asked to complete the survey. This is usually larger than the number of completed
cases used in the analysis due to nonresponse and missing data.
Sampling Frame: A group of households or persons from which a survey sample is drawn.
Satisficing: Selecting the first acceptable or easiest option or response.
Screening: The process of using a brief series of questions early in the survey process to identify
whether sampled cases belong to the target population or possess a characteristic of interest.
Self-Caught Fish: Fish caught or harvested for personal consumption, or to share with family and
friends. This is typically phrased as "fish caught by you or someone you know" in fish consumption
surveys.
Serving Size: Amount of fish consumed at an individual meal or at a single point in time.
Statistical Program to Assess Dietary Exposure (SPADE): A statistical methodology that
estimates the habitual intake distribution for daily and episodically consumed foods or dietary
components based on information of intake, measured on a limited number of days (e.g., 24-hour
recalls) and sometimes in combination with a questionnaire (e.g., food frequency questionnaires),
produced by National Institute for Public Health and the Environment, Bilthoven, The Netherlands.
(http://www.rivm.nl/en/Topics/S/SPADE/About SPADE)
Sport Fishers: Fishers who fish for pleasure or competition.
Stratified Sample Design: A sampling design that separates population attributes into non-
overlapping groups (strata) from which samples are to be selected. The establishment of strata
occurs prior to sampling.
Subsistence: Relying on natural resources (e.g., fishing, hunting, farming) to provide for basic
needs.
Suppression: The reduction in desired intake or consumption due to environmental or other
factors beyond the control of a population (e.g., fears of chemical contamination in fish, fish
populations of inadequate size to support consumption, loss of access to fisheries resources, loss of
access to fishing equipment, changes in social structure affecting harvest). (See also "Heritage rates")
VI
-------
Systematic Error: A source of error in which measurements consistently depart from the true value
in the same direction; affects the sample mean or median and can result in incorrect estimates and
conclusions; another term for bias.
Usual Fish Consumption Rate (UFCR): Term used to denote an estimate that corresponds to a
long-term average daily consumption rate.
Weights: Sampling weights are needed when sampled units are selected using unequal probability
sampling. The use of weights requires specialized statistical procedures during analysis and allows for
the results to be generalized to the population from which the sample was drawn. Weights are used
to adjust for nonresponse and to assign greater relative importance to some sampled elements than
to others. Weights are calculated as the inverse of the probability of selection, and then adjusted for
other factors to improve accuracy.
Vll
-------
DISCLAIMER
This guidance document provides advice on how to conduct surveys to estimate fish consumption
rates. This guidance does not impose legally binding requirements on the U.S. Environmental
Protection Agency (EPA), states, tribes, other regulatory authorities, or the regulated community,
and may not apply to a particular situation based upon the circumstances. EPA, state, tribal, and
other decision makers retain the discretion to adopt approaches on a case-by-case basis that may
differ from those in the guidance, where appropriate. EPA may update this guidance in the future as
new or additional information becomes available.
The Office of Science and Technology, Office of Water, U.S. Environmental Protection Agency has
approved this guidance for publication. Mention of trade names, products, or services does not
convey and should not be interpreted as conveying official EPA approval, endorsement, or
recommendation for their use.
The suggested citation for this document is:
U.S. EPA. (2016). Guidance for Conducting Fish Consumption Surveys. EPA 823-B-l 6-002. Washington,
DC: U.S. Environmental Protection Agency, Office of Water.
Vlll
-------
SUMMARY OF INTENT
The U.S. Environmental Protection Agency (EPA) has prepared this guidance document in an effort
to assist tribes, states, local governments, and others in designing and conducting statistically valid
fish consumption surveys with valid analytic results. EPA recognizes that studies of fmfish and
shellfish (hereafter, referred to as "fish") consumption patterns are essential in developing water
quality standards and assessing the health risks posed by contaminants in fish. These studies also
play an integral role in developing advisories and bans to ameliorate these risks, and to measure the
level of fishing activity or harvest patterns associated with a particular body of water. Studies are
often focused on geographical or cultural populations potentially at high risk. Surveys of selected
populations are used to estimate how much fish tissue is consumed and the frequency at which it is
consumed. Data on exposure and determination of the distribution of average daily intake are
necessary to assess risks posed to consumers of fish and shellfish (U.S. EPA, 2000b). Consumption
patterns, including the types and amounts of fish consumed, parts of fish consumed, seasonality of
fish consumption, frequencies of meals eaten from these organisms, and the preparation methods
used, can also vary greatly within populations because of differences in age, gender, cultural practices
and/or socioeconomic status. Information obtained from fish surveys can be used to: (1) determine
whether the amounts of fish being eaten are safe in relation to possible chemical contamination,
(2) estimate risks to persons who could consume fish that might contain bioaccumulative and
potentially dangerous levels of toxicants, (3) develop water quality standards to protect human
health, and (4)assess the effectiveness of fish advisories. The design of surveys is intimately linked
with survey objectives. Care must be taken in utilizing the results of surveys designed for one
purpose for other objectives.
The purpose of this document is to provide guidance for the design, conduct, and analysis of
surveys focused on characterizing contemporary ingestion of fish. The methodologies are also
applicable to consumption of other aquatic organisms, such as marine mammals, that may be
consumed by populations of interest. Building on the previous EPA guidance document (U.S. EPA,
1998), the discussion of survey methodologies has been updated to reflect more recent
developments in the area of survey research, including use of cellular telephones, the web, mobile
devices, and use of multi-mode data collection designs. To supplement and provide context to the
described approaches, this guidance document also covers a broad overview of the numerous and
complex issues surrounding the development of a study approach, identification of survey
objectives, sampling options, mode selection, questionnaire development, and operational and
analytic considerations. New sections on the topics of consumption suppression and the role of
heritage rates, especially among tribal populations, have been added. Also, in recognition of the fact
that resources for fish consumption surveys can typically be limited, this document addresses survey
design options within the context of budgetary resources to help the researcher make choices that
best fit the situation. This document does not provide direct guidance focused on how to collect and
analyze fish tissue for contaminants or how to conduct surveys to assess understanding of and
compliance with regulations or voluntary programs, even if the surveys are relevant to water bodies
with consumption advisories.
This guidance document is written for users with at least a basic familiarity of statistical and survey
research techniques. References are provided throughout the document to help readers understand
the important elements of topics that may not be their primary area of expertise. However, it should
be noted that more complex survey designs and analyses required to address more challenging
IX
-------
research questions or advanced analytical goals may require the expertise of a survey and/or
statistical professional. Consultants in both areas are widely available and may indeed be found in
other government departments, academia, or in survey research organizations.
There are many considerations when designing and implementing a survey. In particular, planning
for monitoring response rates and handling survey nonresponse are important issues. Nonresponse
can have a varying degree of biasing effect on survey results (as discussed later in this document).
Topics that are not discussed in this guidance document include: detailed methods for achieving
higher response rates; various approaches for using survey incentives; procedures for hiring and
training qualified interviewers; monitoring study progress; and other procedures that affect survey
quality. These are important topics that deserve careful consideration. For example, there are many
methodological approaches to a well-designed incentive program, although the specific
circumstances of the survey must be taken into account. In some cases, a larger cash incentive may
be appropriate, while in other cases, a smaller incentive might be appropriate. Ultimately, the goal is
to achieve acceptable survey response rates to allow the study goals to be met. A survey operations
specialist could provide assistance to ensure that all appropriate steps are being taken to design and
implement a high quality data collection effort that will satisfy the stated goals and objectives of the
planned research.
x
-------
TABLE OF CONTENTS
Chapter Page
ACRONYMS AND ABBREVIATIONS 1
GLOSSARY 111
DISCLAIMER vm
SUMMARY OF INTENT ix
1 BACKGROUND 1
1.1 Purpose and Objective 1
1.2 Relationship of This Document to Other Guidance
Documents 2
1.3 Organization of This Document 3
1.4 Research Design Issues for Fish Consumption Surveys 4
1.4.1 Development of Research Objectives 4
1.4.2 Budget and Time Considerations 5
1.4.3 Fish Consumption Rate (FCR) vs. Usual Fish
Consumption Rate (UFCR) 6
1.4.4 Data Quality Objectives Process 8
2 SURVEY OBJECTIVES AND INFORMATION NEEDS 10
2.1 Overview 10
2.2 Defining the Survey Objectives 10
2.2.1 Defining the Population of Interest 12
2.2.2 Community/Tribal Input 13
2.2.3 Geographic Area(s) of Interest 14
2.3 Information Needs 14
2.3.1 Physical and Demographic Characteristics of the
Population 14
2.3.2 Hard-to-Survey Populations 15
2.3.3 Fishing Activities and Behavior 16
2.3.4 Preparation and Consumption Patterns 17
2.3.5 Suppression and Heritage Rate Issues 17
2.4 Environmental Justice 18
XI
-------
CONTENTS (continued)
Chapter Page
3 SURVEY APPROACHES AND SAMPLE SELECTION 21
3.1 Introduction 21
3.2 Types of Surveys 21
3.2.1 In-Person Interview 21
3.2.2 Creel Survey 22
3.2.3 Telephone Survey 22
3.2.4 Mail Survey 23
3.2.5 Web or Mobile Survey 23
3.2.6 Mixed Mode Survey 23
3.3 Types of Samples 24
3.3.1 Choosing a Sampling Frame 25
3.3.2 Challenges Associated with Some Sampling
Frames 30
3.4 Benchmarking 32
3.5 Determining Required Sample Sizes 33
3.5.1 Design Effect 35
3.5.2 Sample Sizes for Multiple Contact Surveys 35
3.5.3 Sample Sizes for Single Contact Surveys 39
3.6 Effect of Response Rates 40
3.7 Accuracy 41
3.8 Sample Stratification and Oversampling 42
3.9 Response Bias/Avidity 42
3.10 Choosing an Approach 43
4 QUESTIONNAIRES FOR COLLECTING FISH
CONSUMPTION DATA 58
4.1 Background 58
4.2 Questionnaires for Fish Consumption Surveys 61
4.3 Use of Visual Aids 62
4.4 Respondent Body Weight 63
4.5 Portion Sizes 63
4.6 Collecting Data by Fish Species 64
4.7 Fishing Practices and Locations 64
4.8 Data Capture and Management Methods 65
xii
-------
CONTENTS (continued)
Chapter Page
4.9 Interview/Questionnaire Length 66
4.10 Qualitative Methods to Support Survey Instrument
Development 67
4.11 Pilot Testing 68
5 SURVEY IMPLEMENTATION AND OPERATIONAL
CONSIDERATIONS 70
5.1 Introduction 70
5.2 Defining Eligible Respondents 70
5.2.1 Household Reporters 70
5.2.2 Children 70
5.2.3 Proxy Respondents 71
5.3 Pre-contact Procedures and Introduction of the Study 71
5.4 Informed Consent for Interview 72
5.5 Impact of Seasonality on Data Collection Schedule 74
5.6 Language and Literacy Issues 74
5.7 Survey Nonresponse 75
5.8 Methods for Respondent Retention 77
5.9 Confidentiality 77
5.10 Data Availability 78
6 SUPPRESSION 79
6.1 Estimating Heritage FCR 79
6.2 Alternative Methodologies to Measure Suppression 81
7 ANALYTIC APPROACHES 85
7.1 Introduction to Analysis 85
7.1.1 Weighting 85
7.1.2 Estimating Usual Fish Consumption Rates 85
7.1.3 Reporting of Usual Fish Consumption Rates 87
7.2 Uncertainty 88
7.2.1 Seasonality 89
7.2.2 Bias in the Reported Fish Consumption 89
xm
-------
CONTENTS (continued)
Chapter Page
7.2.3 Use of Standard Recipes 89
7.2.4 Estimation of Usual Fish Consumption 89
7.3 Analyzing Recall Data 91
7.3.1 Preparing the Data 91
7.3.2 Application of the NCI Method 93
7.4 Analyzing Frequency Data 94
8 SUMMARY AND APPENDIX OVERVIEW 97
9 REFERENCES 99
Appendices
A Additional Instrument Development Considerations A-l
References A-4
B Background on Dietary Assessment Methods B-l
References B-5
C Application of the NCI Method Macros C-l
References C-4
D FCST Hard-Copy Questionnaire and Documentation D-l
Tables
3-1 Estimated percentages of the U.S. population who consumed fish
on a given day, NHANES 2007-2012 36
3-2 Minimum sample size versus the expected probability of
consuming fish in any recall (P) and the design effect 37
3-3 Values for A when estimating the mean, the median (50th
percentile), or the 90th percentile of usual fish consumption 38
xiv
-------
CONTENTS (continued)
Tables
3-4 Sample size as a function of the desired confidence interval width
for proportions close to 50 percent and 90 percent 40
3-5 Level 1 framework for choosing a sample design and data
collection mode based on independent factors 47
3-6 Level 2 framework for choosing a sample design and data
collection mode based on independent factors 50
3-7 Level 3 framework for choosing a sample design and data
collection mode based on independent factors 54
4-1 Data needs, advantages, and limitations of various analytical
methodologies for estimating UFCR 60
7-1 Example of 24-hour fish recall data, ounces 91
7-2 Example of summary of 24-hour fish recall data to derive analysis
variables, ounces 92
7-3 Example format of analytic data set 92
7-4 Example of conversions to consistent time periods 95
7-5 Example of conversions to consistent unit of weight 95
7-6 Example of calculating UFCR for each individual 96
7-7 Example of estimating annual UFCR with seasonal data 96
Figures
6-1 Food wheel showing the relative proportion of dietary staples in
an example subsistent diet 80
xv
-------
i
BACKGROUND
Purpose and Objective
Concern over potential human health risks associated with chemically contaminated fish (Wathen et
al., 2015; Stahl et al., 2014; Stahl et al., 2013; Thompson & Boekelheide, 2013; U.S. EPA, 2009;
National Research Council, 2000; Ahmed et al., 1993) has led many tribal, regional, state, and local
governments to study fish consumption patterns among their citizens. Surveys are useful for
deriving information about present, recent past, and near future fish consumption rates. The
suitability of using surveys to gather information about past consumption rates depends on how
long the baseline fish consumption rates have been altered or suppressed. For example, a recent
contamination event that alters fish harvest and consumption patterns is amenable to before-and-
after event comparisons. However, if baseline fish consumption patterns have been affected by
prolonged contamination or suppression, a survey of contemporary people will likely not be able to
define baseline (or heritage) consumption rates. See Chapter 6 for a discussion of non-survey
approaches for assessing heritage rates and estimating suppression of fish consumption.
The processes and procedures by which government agencies develop consumption surveys vary
widely. Additionally, the survey results are used for a variety of purposes, such as a basis for setting
water quality standards, evaluating health risks posed by contaminants in fish, evaluating harvest
rates or patterns, or assessing the effectiveness of consumption advisories or bans. In an effort to
assist tribes, states, local governments, and others in designing and conducting statistically valid fish
consumption surveys with valid analytic results, the U.S. Environmental Protection Agency (EPA)
has prepared this guidance document for conducting fish consumption surveys.
The purpose of this document is to provide guidance for the design, conduct, and analysis of
surveys focused on characterizing contemporary ingestion of finfish and shellfish. Hereafter, the
term "fish" is used to represent both finfish and shellfish, unless otherwise stated. These surveys
may be intended to characterize the fish consumption rates of the full population of the locality or
of targeted groups such as recreational or subsistence fishers, high-consuming individuals, pregnant
or lactating women, women of childbearing age, or disadvantaged economic groups. For the
purposes of calculating fish consumption rates, this guidance provides information on several
methods, but provides more detailed information on the National Cancer Institute (NCI) method.
The methodologies are applicable to consumption of other aquatic organisms that may be
consumed, such as marine mammals, by populations of interest. This document does not provide
direct guidance focused on collection and analysis of fish tissue for contaminants or conducting
surveys to assess understanding of and compliance with regulations or voluntary programs even if
the surveys are relevant to water bodies with consumption advisories.
Fish consumption rate data are essential in developing water quality standards, and they also play an
integral role in developing consumption advisories and bans and in legal proceedings. More broadly,
data on contaminant exposure and determination of the distribution of long-term average intake are
necessary to assess risks posed to consumers of fish (U.S. EPA, 2000b). Consumption patterns,
including the types, parts, and amounts of fish, seasonality of fish consumption, frequencies of
meals eaten, and the preparation methods used, can vary greatly within populations. These variations
may be related to demographic characteristics, such as age, gender, race/ethnicity, socioeconomic
1
-------
status, and geographic area (Birch et al., 2014; Razzaghi & Tinker, 2014; Soon et al., 2014; U.S. EPA,
2014; Connelly et al., 2012; Mahaffey et al., 2009). Consumption patterns can also differ between
populations because of differences in cultural practices (Ellis et al., 2014; Judd et al., 2004).
The survey methods presented in this document may be used to obtain information on the
consumption of fish for purposes of estimating long-term average intake for various fish types of
interest. Examples include all fish, locally caught fish, anadromous vs. resident fish, fish from a
particular body of water, fmfish vs. shellfish, and fish from various habitat types (e.g., freshwater,
estuarine, near coastal marine, marine). This information can then be used to (1) determine whether
the amounts of fish being eaten are safe in relation to assessed chemical contamination, (2) estimate
risks to persons who could consume fish that contain unhealthy levels of bio-accumulative toxic
compounds, and (3) set water quality standards.
1.2 Relationship of This Document to Other Guidance
Documents
The EPA has developed a series of five documents designed to provide guidance to tribal, state,
local, and regional environmental health officials who are responsible for issuing consumption
advisories for non-commercially caught fish. The documents are meant to provide guidance and do
not constitute a regulatory requirement.
The first four documents are as follows (all are available at https://www.epa.gov/fish-tech/epa-
guidance-developing-fish-advisories):
Guidance for Assessing Chemical Contaminant Data for Use in Fish Advisories,
Volume 1: Fish Sampling and Analysis (U.S. EPA, 2000a)
Volume 2: Risk Assessment and Fish Consumption Limits (U.S. EPA, 2000b)
Volume 3: Risk Management (U.S. EPA, 2000c)
Volume 4: Risk Communication (U.S. EPA, 1995)
In 1998, EPA developed a fifth document, "Guidance for Conducting Fish and Wildlife
Consumption Surveys," providing guidance on methods for obtaining consumption rate data for use
in characterizing exposure in a population when estimating potential risks, determining whether a
consumption advisory is warranted, and developing or modifying water quality standards (U.S. EPA,
1998).
Additionally, in 2000, EPA developed a document (U.S. EPA, 2000d), "Methodology for Deriving
Ambient Water Quality Criteria for the Protection of Human Health (2000)." This document
provides a methodology for developing human health ambient water quality criteria as required
under the Clean Water Act. The methodology requires having an acceptable estimate of the rate of
consumption of fish from fresh, estuarine, and near coastal marine waters for the population for
whom the criteria are being developed to protect. In some cases, the national default rate may be
appropriate. In other situations, tribal, state-wide, or local rates are necessary. This document
2
-------
consists of three volumes and is available at https:/ /www.epa.g~ov/wqc/human-health-water-
quality-criteria.
This current guidance document updates the 1998 guidance document concerning methodologies
for conducting fish consumption surveys by reflecting updated survey research tools and concepts,
and providing updates to the standard analytic methods for fish consumption surveys. It provides
guidance that can be used to produce fish consumption rates for deriving ambient water quality
criteria, among other uses.
1.3 Organization of This Document
There are five basic steps in the design and development of a survey for estimating fish
consumption rates. Each of these steps will be addressed in this document.
1. Clear definition of survey objectives (including defining the population of interest)
2. Identification of specific information needs
3. Decisions about a survey approach and sample design
4. Development of an appropriate survey questionnaire
5. Consideration of implementation and operational issues
Chapter 2 presents a discussion of design considerations that the researcher must examine. These
include articulating the overall objectives of the study, as well as understanding the population to be
studied and the geographic area of interest. Outlining the information needs of the research is an
important step, and this is also discussed in Chapter 2. Fish consumption behaviors, fish preparation
methods, portion sizes, and seasonal variation in consumption patterns are all factors to be
considered during the design phase. Issues surrounding suppression and heritage rates are also
discussed as they pertain to comparing current levels of fish consumption with historical levels.
Chapter 3 presents a discussion of types of surveys, sample designs, and sample size requirements so
the researcher will be aware of the numerous options that can be utilized, depending on specific
survey requirements. Other issues surrounding sample design, including the target population,
eligible participants, precision and accuracy requirements, stratification, and response rates, are also
covered. This chapter also discusses challenges associated with some types of sampling frames. The
chapter concludes with a series of tables indicating which approaches are appropriate/not
appropriate/or may be appropriate—based on various survey objectives and types of populations.
These options are presented within a budgetary framework designed to aid the researcher in
determining which approaches will best fit their specific needs from both a scientific and a resource
perspective.
Chapter 4 provides guidance on methodologies for collecting dietary information using the
approaches described in Chapter 3. Specifically, the reader is provided with a description of an
existing instrument that can be used to collect data on fish consumption. Various factors that are
relevant for instrument development, such as respondent recall periods, use of visual aids, portion
sizes, etc., are also discussed. The use of qualitative methods and other pilot testing that can be used
to finalize a survey instrument are described.
3
-------
Chapter 5 presents survey implementation and operational considerations. During the development
process, it is important to define eligible respondents, implement pre-contact procedures to
maximize response rates, ensure the proper implementation of informed consent, and consider how
other factors, such as language and literacy issues, will affect response rates and data quality.
Response rates are discussed and some methods for respondent retention (for surveys with multiple
observation points) are provided.
Chapter 6 provides an overview of methodologies for measuring suppression and estimating
heritage rates.
Chapter 7 discusses analytic considerations. Along with Appendix C, Chapter 7 provides a walk-
through for using the National Cancer Institute Method (NCI Method) to analyze fish consumption
data to obtain estimates of usual fish consumption rates.
Appendix A includes a discussion of additional considerations for instrument development to help
the researcher develop questions that are methodologically sound, and provides example questions
for use in existing surveys. Appendix B presents background on a variety of generally accepted
dietary assessment methods. Appendix C provides details for the application of the NCI Method.
Appendix D provides documentation of a publicly available, automated survey instrument along
with a hard-copy version.
The design of a study needs to reflect the specific needs of the situation. For example, one study
may be needed to determine an appropriate fish consumption rate (FCR) for updating a state's water
quality criteria as required under Section 303 of the Clean Water Act (CWA). While EPA provides a
national default FCR, because fish consumption varies by geographical location, racial/ethnic group,
age, income, and possibly other factors, EPA suggests a "four preference hierarchy" for states and
authorized tribes to follow in selecting a FCR to be used in the development of water quality criteria.
The four preference hierarchy is as follows.
1. Use of local data
2. Use of data reflecting similar geography/population groups
3. Use of data from national surveys
4. Use of EPA's default intake rates
Local data may include data from a variety of contexts, including consumption by the general
population state-wide, by a specific subpopulation within the state or region, consumption of fish
taken from a specific water body or within a specific community, or a traditional baseline heritage
rate. Depending on the data used, it may be appropriate to adjust the contemporary rate to account
for suppression effects by documenting a heritage or unsuppressed rate with additional literature-
based research (for tribes, for instance), or by evaluating recent past rates through a survey (see
1.4
Research Design Issues for Fish Consumption
Surveys
1.4.1
Development of Research Objectives
4
-------
Chapter 6). Some government agencies may be interested in determining a rate to protect the most
sensitive population(s) in a particular area, e.g., women of childbearing age. The EPA Office of
Water's national default intake is 22 g/day, which is the 90th percentile of freshwater and estuarine
fish of the general adult U.S. population (U.S. EPA, 2014). The Exposure Factors Handbook (U.S.
EPA, 2011) and EPA's Guidelines for Exposure Assessment (U.S. EPA, 1992) both recommend
site-specific consideration. In some circumstances a health-based rate may be desired.
Each of the following factors will influence whether a survey is the most appropriate method: the
survey design, including the sample selection to best represent the target population, the data
collection mode (in-person, telephone, web, etc.), the survey instrument, and the analysis method.
While the purposes and uses of the data are established during survey design and questionnaire
development, the manner in which the data are to be analyzed is not always pre-determined. To the
extent possible, the details of analysis and interpretation methods should be defined early in the
survey design process because they may have a significant bearing on the form and content of the
questions to be asked. For example, using some analytical methodologies requires data gained
through multiple contacts with participants and may require a minimum number of respondents
with certain characteristics. Addressing these issues during initial planning and questionnaire design
minimizes difficulties that may arise during data analysis and interpretation.
1.4.2 Budget and Time Considerations
Surveys come in all shapes, sizes, complexities, and costs. Budget and time constraints can play a
significant role in the ultimate decisions about how to design and conduct any survey. Sampling
frame development, quality control, data entry, length and type of data collection, and various types
of analyses can be costly and resource intensive. It is important to note, however, that low-cost
resources available to conduct any given survey may include use of already acquired survey software,
in-kind participation of government personnel (perhaps beyond the division sponsoring the survey),
volunteer labor, and/or accessing existing government resources such as fish license registries.
Alternatively, it may be possible to add a limited number of questions to an existing survey
conducted by the state such as the Behavioral Risk Factor Surveillance Survey (BRFSS) (see Section
3.10). Finally, the services of an outside consultant or consulting firm can be contracted. Costs of
implementation may vary in different regions of the country.
For all these reasons, it is very difficult to provide estimates of the total cost of conducting the
survey designs described in this document. However, within certain budget categories or ranges, this
document provides guidance about a variety of methods that can be employed to ensure that quality
fish consumption surveys can be performed given varying levels of available resources. Matching
available resources and time constraints with achievable survey objectives is perhaps the single most
critical factor in the survey development/design process. Section 3.12 provides tables of survey
approaches that are appropriate for various levels of funding.
5
-------
1.4.3
Fish Consumption Rate (FCR) vs. Usual Fish Consumption
Rate (UFCR)
An individual's FCR is intended to estimate the expected quantity of fish consumed per unit of time.
For a population, there are a variety of FCR distributions depending upon which fish species are
being evaluated; for any particular distribution, some individuals consume more fish per unit of time
and some consume less. Different time units can be used to express the rate (e.g., per day or per
week). The FCR also depends on the window of time used in a questionnaire for querying fish
consumption. The window might be one day or a year. The FCR is often estimated using statistical
analysis. It may change over time. For example, it may be higher in the summer than the winter.
Thus, the distribution of the FCR depends on the time frame (e.g., summer, winter, annual) and the
length of the time window used in asking about consumption.
There are three main types of FCR: (1) current, (2) baseline, and (3) heritage:
¦ A current FCR is the amount of fish currently consumed per unit of time by the target
population. These rates are generally used in water quality criteria programs where the
waters are not contaminated, and fish consumption is not thought to be suppressed and
for immediate interventions such as fish advisories. EPA develops ambient water quality
criteria (i.e., recommendations) based on the latest scientific knowledge which states and
tribal governments may use to determine when water is unsafe for humans, aquatic life,
and wildlife. For more information: https: / /www.epa.gov/wqc /basic-information-
water-quality-criteria.
¦ A baseline FCR is the amount of fish consumed prior to an event (e.g., a pre-release
baseline or pre-construction of an upriver dam) that has resulted in decreased fish
consumption, for example an environmental spill such as the Deepwater Horizon event.
These rates are most generally used in clean-up situations such as at Superfund sites
where current rates are suppressed due to contamination of surface water and the site
needs to be cleaned so that consumption can safely return to the baseline rate.
¦ A heritage FCR is the amount of fish consumed prior to non-indigenous or modern
sources of contamination and interference with the natural lifecycle of fish, in addition
to changes in human society. While it is often thought of as a historic rate, it can also be
a current unsuppressed rate.
The FCR depends on the timeframe in which fish consumption is estimated. If a person eats 7
ounces of fish for dinner every Friday and none at other times, the daily FCR-depends on which day
the interview takes place. If the interview asks for fish consumption on Friday, the daily FCR is 7
ounces per day. The daily FCR is 0 ounces for every other day of the week. If the interview collects
fish consumption over a week, the weekly FCR expressed as a daily rate is 1 ounce per day,
regardless of which day the interview takes place. If a 7-ounce fish meal is consumed on average
once every 7 days, but sometimes 3 days in a week and other times not for several weeks, the weekly
FCR can vary from 0 ounces per day to 3 ounces per day, even though the true long-term FCR is
constant and is the same as in the first example. It is important to note that in some instances even a
time interval of a year may not be a truly representative long-term average due to variation caused by
the year-to-year changes in the size of anadromous fish returns. However, as the time interval
covered by the data gets longer, the distribution of the associated FCRs becomes less variable and
the mean FCR becomes a relatively precise estimate of the true long-term consumption. Adding the
6
-------
term "usual" to "fish consumption rate" (UFCR) implies that the resulting estimates are those that
correspond to long-term averages, rather than short-term estimates, and avoids a distinction
between the true rate and the estimated rate. The UFCR estimate is the amount consumed per day
over a long, although unspecified, period of time. For further discussion about usual intake, please
see the National Cancer Institute's (NCI's) Applied Research website
(http: / / appliedresearch.cancer.gov/diet/usualintakes /).
Assuming the long-term FCR, is constant over time, and given that short-term dietary recall data are
more accurate than long term recall data (Kipnis et al., 2003), methodologies can be designed to
estimate the distribution of the true, long-term FCR even though the data are collected over a
limited time frame.
In the mid-2000s, the NCI developed a statistical methodology to estimate usual intake of
episodically consumed foods (Tooze et al., 2010; Tooze et al., 2006). This method, known as the
"NCI Method,1" has been published, and statistical programs to implement it are available on NCI's
website at https: / / epi.grants.cancer.gov/diet /usualintakes / method.html. There are also other
commonly accepted analytical methods to estimate UFCR. These include use of a food frequency
questionnaire (FFQ), either total diet or food-specific, and other statistical methodologies such as
the Multiple Source Method (MSM) (Harttig et al., 2011), the Iowa State University Method (Nusser
et al., 1996), and the Statistical Program for Age-Adjusted Dietary Assessment (SPADE) (Dekkers et
al., 2014). These statistical methods, along with the NCI Method, typically have a general
requirement that dietary data are collected through two or more contacts with participants utilizing a
24-hour recall to collect dietary data. The time between recalls should be long enough to assume that
the consumption reported in the first recall is not related to (or correlated with) the consumption
reported on the following 24-hour recall.
The use of a FFQ to estimate UFCR is a lower cost option, as it can be conducted with only one
contact with respondents. However, many details of dietary intake may not be captured on typical
total diet FFQs. Each question in a typical total diet FFQ usually represents several foods or
includes all possible preparation methods. The portion sizes queried about in typical total diet FFQs
are generally categorized into three levels such as <2 oz., 2 to 6 oz., and >6 oz. Portion size
quantification in FFQs should consider amounts of different fish preparations that are commonly
consumed by the population. They should be culturally relevant. Models should consider
preparation type and might correspond to numbers of organisms or volumes consumed. While both
24-hour recalls and FFQs are known to have measurement error, typical total diet FFQs can be
more prone to systematic error (Freedman et al., 2004; Kipnis et al., 2003; Subar et al., 2003; Kroke
et al., 1999). Note that research concerning bias in FFQs compared 24-hour recalls have been
conducted on total diet FFQs and not food-specific FFQs.
For the purposes of fish consumption surveys, the term FFQ generally refers to a fish-specific FFQ
and not a total diet FFQ. Reducing the number of foods that are asked about allows for the fish-
specific FFQ to cover all fish species of interest in separate questions and to inquire about
preparation or other factors of interest, such as source of the fish, without becoming too long.
Follow-up questions for each species of interest include an estimate of the respondent's usual
1 Note: EPA has developed an alternative approach which approximates the NCI Method. See EPA, 2014.
7
-------
portion size and should consider amounts by different fish preparations (culturally relevant to the
population of interest) that are commonly consumed by the population. This additional information
may reduce the bias that is observed with the total diet FFQs. Throughout the remainder of this
document, the term FFQ refers to a fish-specific FFQ with portion size estimates, unless otherwise
stated.
Surveys to estimate fish consumption rates have utilized FFQs as well as a combination of a FFQ
and a 24-hour fish recall. Statistical experts suggest the addition of the 24-hour recall may be used to
increase the accuracy of the portion size estimate; however, more research in this area is needed.
There are circumstances for which the dietary data collection requirements for using the NCI
Method (or MSM, the Iowa State University Method, or SPADE) are impractical. For example, if
the desired fish consumption rate estimate is for a small population, it may be difficult to collect
enough data to use the NCI Method. Or, if a survey is intended to estimate consumption of a single
species or infrequently consumed group of species, it may be cost-prohibitive to attempt to collect
the necessary data to use the NCI Method. If fish consumption rates are estimated from FFQ data,
it is important to be aware of and adjust for the possible bias (see Section 7.2). Further discussion of
these methods and a table depicting data needs and advantages and limitations can be found in
Chapters 4 and 7.
An additional consideration is the distinction between consumer-only FCR and per capita FCR. On
a theoretical level, a consumer-only FCR is the FCR for the population that consumes fish,
excluding those who do not, while the per-capita FCR is the FCR for the entire population, both
consumers and nonconsumers. In practice, it is difficult to determine who is a fish consumer and
who is not. Is a fish consumer a respondent that reported fish consumption on the 24-hour recall? Is
a fish consumer a respondent who reported they consume fish once a week, once a month, once a
year? If a study is interested in the long-term average fish consumption (UFCR), then even those
who rarely consume fish, say one time per year, should be considered fish consumers. The upper
end of the resulting distribution of UFCR would represent the frequent consumers while the lower
end of the distribution would represent the infrequent consumers. The NCI Method and other
approximations assume all respondents included in the analysis to be fish consumers, and each is
assigned a non-zero probability of fish consumption based on the factors included in the model
(e.g., gender, age, race/ethnicity, reported frequency of consumption). Some respondents have very
low probabilities of consumption, and thus their resulting usual intakes are low. If a study is truly
only interested in consumer-only FCR, then it must determine how it will define a consumer and
include the appropriate question on the survey to exclude nonconsumers from the analysis.
1.4.4 Data Quality Objectives Process
The Data Quality Objectives (DQO) process provides a systematic procedure for defining the
criteria that a data collection design should meet. It can assure that the type, quantity, and quality of
data used in decision-making will be appropriate for the intended purpose. Following this process
will require the survey team to focus on the purpose of the research and use of the findings before
assuming that a survey will meet their needs as well as through the planning and designing of a
survey. The DQO process has seven steps.
8
-------
1. State the problem — describe the problem to be studied and review relevant literature and
information
2. Identify the decision — identify what questions the study will attempt to resolve and what
actions may result
3. Identify the inputs to the decision — identify the information that needs to be obtained and
the measurements that need to be taken/data that needs to be collected to resolve the
decision statement
4. Define the boundaries of the study — specify the time periods and spatial area to which
decisions will apply and determine when and where the data should be collected
5. Develop a decision rule — define the statistical parameter of interest, specify the action level,
and integrate the previous DQO outputs into a single statement that describes the logical
basis for choosing among alternative actions
6. Specify tolerable limits on decision errors — define the decision maker's tolerable decision
error rates based on a considerations of the consequences of making an incorrect decision
7. Optimize the design for obtaining data — evaluate information from the previous steps,
generate alternative data collection designs, choose the most resource-effective design that
meets all DQOs
For more information, see Guidance on Systematic Planning Using the Data Quality Objectives Process (U.S.
EPA, 2006), available at https://www.epa.gov/quality.
Important considerations when selecting survey design, questionnaire, and analytical approaches
include:
¦ Precision of the estimates: what is the magnitude of the unexplained errors that
contribute uncertainty to the estimate?
¦ Possible bias of the estimates: might the estimates be consistently higher or low?
¦ Representativeness: do the respondents represent the population to be studied?
¦ Comparability with other studies: if important, can the estimates be compared to similar
estimates from other studies?
¦ Completeness: might missing values or non-response adversely affect the estimates and
conclusions?
¦ Sensitivity: do the survey questions and collection procedures provide data to support
the desired analyses, including possibly discriminating among demographic groups or
types of fish consumption?
As part of the planning process, indicators of data quality can be established, such as minimum
response rates and minimum completion rates for survey questionnaires. These data quality
indicators can be monitored during the data collection to document quality and provide a basis for
adjusting survey procedures during the data collection.
9
-------
2
SURVEY OBJECTIVES AND INFORMATION NEEDS
2.1
Overview
As listed in Chapter 1, there are five basic steps in the design and development of a survey for
estimating fish consumption rates. Each of these steps should be addressed within the researcher's
budgetary and resource limitations.
This chapter addresses the first two of these steps — (1) Clear definition of survey objectives
(including defining the population of interest) and (2) Identification of specific information needs.
A clear definition of the survey objectives is an integral step in the survey development process. Fish
consumption estimates obtained through a survey for one set of objectives may not be appropriate
to use for another purpose. For example, current FCR estimated for use in setting advisories may
not be appropriate for setting water quality standards if fish consumption is not at baseline levels. If
used for this purpose, water quality standards would be set where unsuppressed consumption may
result in unhealthy exposure. As each survey approach has unique advantages and disadvantages, the
survey objectives should guide the conduct and design of the sample, survey questionnaire, and
analysis plan to obtain specific results (e.g., estimated UFCR, estimated age distribution of
consumers). The information collected must be targeted to address the objectives, and thus serve as
a planning tool to ensure that the required information is collected in a manner consistent with
planned analysis techniques. Through defining the objectives, the team may find that they require
heritage rates, thus a survey is likely not the best approach (see Chapter 6).
Key considerations include:
¦ The planned use of the data (e.g., setting ambient water quality criteria, setting cleanup
levels, assessing human health risk, assessing advisory effectiveness, risk management)
¦ The target population (e.g., general population of state, county, other location of
concern, local fishers, women of childbearing age, tribal members, consumer-only vs.
per capita, their history and lifestyle)
¦ The type of fish of interest (e.g., all fish consumed from all sources, estuarine and
freshwater fish only, a specific fish species or group, locally caught or commercially
available)
¦ The geographic area of concern (e.g., tribal lands, entire state, one particular lake or
river)
¦ The condition of waters within the geographic area of concern (e.g., existing fish
consumption advisories, watershed contamination, environmentally sensitive area)
¦ The timing of the survey period
¦ The targeted number of data collections from each participant
2.2
Defining the Survey Objectives
10
-------
In state or tribal ambient water quality criteria programs, local fish consumption data are preferred
over national default rates. Thus a state or tribal program may define the survey objectives as
follows:
¦ Estimate the distributions of fish consumption for various fish types by habitat among
the population of interest
¦ Estimate the proportion of fish consumption by fish habitat (i.e., marine, near coastal
marine, estuarine, and freshwater and anadromous vs. resident) or by watershed
¦ Estimate the distribution of consumption of locally caught fish among high consumers
¦ Estimate the proportion of total fish intake that is locally caught vs. commercially
available
Given these objectives, the questionnaire must collect information on the species consumed in order
to obtain rates for estuarine, freshwater, near coastal marine, and marine fish separately from total
fish and source of the fish. Additionally, the study needs to be representative of the population of
interest and needs to ensure enough frequent consumers are represented in the sample. Weighting of
the sample might be employed to adjust the representation of frequent consumers to reflect that of
the population. It may not be possible for a state survey to accurately characterize the consumption
levels of specific groups within the general population given the low fraction of survey results that
are reflective of these groups. It will be particularly difficult to develop upper percentile estimates of
fish consumption for such groups. To the degree possible, a state survey should try to include
enough frequent consumers so that their contribution to the general population FCR is reflected.
Groups that are highly segregated from the general population or distrustful of governmental
entities may not respond to a survey administered by a regulatory agency at the local, state, or federal
level. In these instances, it may be necessary for the state to utilize the results of additional surveys
that are not managed by the state to characterize FCR for specific high fish consuming populations.
The state can establish a relationship with the groups of interest, developing lines of communication
and coordinating efforts regarding survey design and implementation to help to ensure that adequate
and appropriate data from these groups are available to inform a state's regulatory efforts. Federal
agencies with trust responsibility to tribes may assist in facilitating communication between states
and tribes. See Section 2.2.2 for more information.
In some cases, consumption data are desired to evaluate effectiveness of advisories, i.e., the success
of existing advisory messages recommending certain consumption behavior. Determination of actual
consumption levels can serve to improve the accuracy of the advisory. For example, an advisory
program assessing its message to limit consumption of largemouth bass caught at a specific lake to
two times a month and advising the fat be trimmed from the fish and internal organs not be
consumed, may define the survey objectives as follows.
1. Identify people who consume fish from the specific lake
2. Estimate the frequency of their consumption of largemouth bass caught at the lake
3. Estimate the distribution of long-term average intake of largemouth bass caught at the
lake
4. Determine the parts of the fish that are consumed
Given these objectives, it is clear that a state-wide general population survey would not provide the
data required, and information about preparation of the fish would need to be included in the survey
11
-------
instrument. Both the target population and the water body are relatively small, so the survey would
best be conducted at access sites rather than through mail or telephone. Unless addresses and/or
telephone numbers of the target populations are known, such approaches would be unlikely to
capture enough respondents in the specific population for a statistically valid estimate.
2.2.1 Defining the Population of Interest
Identification of the population of interest should be articulated by the study objectives. Defining
the population is of key importance in developing a sample that directly represents the population or
which can be made to represent the population using strata and pre-considered weighting factors.
Surveys can be designed to target groups that might be at greater risk of exposure to contaminants
in fish due to higher consumption rates, such as subsistence fishers. Obtaining accurate exposure
information (e.g., fish consumption data), is a critical aspect of this responsibility. Surveys can be
designed to focus on populations that may be more susceptible to the health effects of contaminant
exposure such as children, pregnant women, people with pre-existing health problems, the elderly,
etc. If the target population is relatively small and it is desirable and feasible to survey the entire
population (i.e., take a census), then the results obtained will be observations of the population
parameters. More likely, however, is a situation in which a subset of the target population needs to
be sampled; the results obtained are sample statistics which, if obtained correctly, are expected to be
good approximations of the population parameters. Census estimates have less risk of error than
sample estimates because they are subject only to the reliability, validity, and measurement error
involved in the survey response (see discussion of Accuracy in Section 3.7). Sample estimates are
subject to the same types of error risk as a census, but also to sample selection bias and sampling
error risks.
A major consideration in the survey design is the size of the population of interest (e.g., number of
pregnant women, or high fish consumers) and the prevalence of the behavior(s) of interest (e.g.,
consumption of locally caught fish) within the larger population. If the subgroup of interest or the
behavior to be studied is relatively rare — or hard to find — then some sample designs should be
eliminated from consideration. For example, a random digit dialing survey, address-based sample, or
area probability sample may require prohibitively large screening efforts to identify a sufficient
number of respondents meeting the criteria for the study. It is also important to learn as much as
possible about the characteristics of the population of interest. The survey design, including
sampling methods and mode(s) of administration, must carefully consider who the potential
respondents will be. Questions such as the following should be addressed during the design phase.
¦ What is the population (or sub-population) of interest?
¦ Where do members of the population live, or where can they be found?
¦ What are the literacy and web penetration rates among the population of interest and
what are the common methods for finding and contacting the members of the
population?
¦ How many of them have telephone numbers (cell and/or landline) and/or fixed
physical or mailing addresses?
¦ How many of them use land lines versus cell phones?
¦ Do the survey objectives require identifying populations of relatively low prevalence in
the general population (e.g., subsistence anglers or pregnant women)?
12
-------
¦ What response rates are needed in order to provide a large enough final expected
sample size for certain subgroups to answer the study's main questions with sufficient
precision?
If key characteristics of the population to be surveyed are known, survey designers should take these
into account and rule out any sample designs or modes that are known to be unsuitable or
inapplicable for a substantial proportion of the target population. For example, one should not plan
to do a telephone survey on a group that does not include a high percentage of individuals with
available telephone numbers. Similarly, a self-administered mail survey would not be suitable for a
group with known low literacy rates or lack of fixed physical or mailing addresses.
States or others planning fish consumption surveys that may inform environmental regulatory
actions, such as establishing water quality standards, for geographic areas that include tribal lands,
rights, or populations should consider the potential relevance of tribes' treaty and/or other reserved
rights to such surveys to ensure that their actions are protective of tribal fishers exercising those
rights, as applicable.
2.2.2 Community/Tribal Input
For all surveys, procedures should be implemented for identifying, contacting, and coordinating with
the appropriate community/tribal leaders early in the study design process. For state surveys that are
also incorporating tribal populations or other culturally diverse sub-populations, this is an important
step that can engage the tribe or local community and obtain their input into the research process.
Tribal officials, local officials, and community partners can help frame the purpose of the study, and
formulate objectives that will result in meaningful information for use by tribal members and other
local residents or fishers. The tribal members can also be helpful in providing local, or traditional,
names for fish, parts of fish consumed, unique fish preparations or dishes, fishing locations, feasts
or communal meals where fish are shared, and perspectives on current vs. historic consumption.
Local context is especially important among tribal populations. If the survey is not being led by the
tribe, care should be exercised to ensure that tribal input and involvement is solicited at each step of
the development process.
Community engagement efforts can serve several purposes. First and foremost, communicating with
community leaders can help obtain cooperation from members of the community and ensure that
the research meets the needs of the community. Engaging with community leaders can also help
identify cultural norms that may be important for cooperation or collecting accurate data (Donatuto
& Harper, 2008). For study designs that target specific populations such as particular ethnic groups
(e.g., Hmong Americans) or American Indian and Alaskan Native tribes, where there may be
inadequate sampling frames, input from community leaders can help to identify strategies for hard-
to-reach community members.
As noted in Section 2.2, it may be preferable for both the tribe and the state, to have tribes (or other
groups) conduct their own separate surveys to better accommodate collection of the detailed
information that is associated with a smaller group for which fish consumption is an important
cultural and/or spiritual activity. Characterization of the cultural, spiritual, and/or historic aspects of
fish consumption, as well as the relationship between public health and fish consumption, are survey
objectives that may be of greater importance to tribes than states. Additionally, survey modalities
13
-------
suited to the general population (e.g., mail, phone, internet) may not be suitable for tribal
populations. Regulatory perspectives of tribal populations may differ from those of government
entities and intended uses of the data should be clarified up front to avoid any misperceptions.
The survey team may be interested in using community-based participatory research (CBPR). CBPR
is a partnership approach to research that involves community members, practitioners, and academic
researchers in all aspects of the research process. It enables all partners to contribute their expertise
and share responsibility and ownership (Israel et al., 1998). See Israel et al., 2012; Minkler and
Wallerstein, 2003; and Israel et al., 1998, for more information.
2.2.3 Geographic Area(s) of Interest
If fish consumption rates are desired for a specific geographic area of interest, the area in which the
survey is to be conducted should be carefully defined. It will be a factor in how the sampling frame
is developed. Is there an interest in surveying only persons living within a certain area regardless of
where they catch their fish, or should the sample include persons who harvest and eat fish caught in
a certain area? Defining the geographic area can be done on a watershed basis, a certain radius
around a Superfund site, by travel distance to fishing locations, or by jurisdictional boundary. It may
be important to review the history of the watershed, looking for past or current contamination,
dams, spawning areas, environmentally sensitive areas, etc.
2.3 Information Needs
In addition to the overall purpose and objectives of a consumption study, the need for information
about specific aspects of consumption or characteristics of fish consumers should be considered.
The extent to which these factors are important and the type of information needed to meet the
objectives of the study will influence which survey approach is selected.
2.3.1 Physical and Demographic Characteristics of the Population
Collecting physical and socio demographic characteristics of fish consumers is important for a variety
of reasons. These data allow for presentation of results by population characteristics, and they allow
for analysis of nonresponse bias. Characterizing fish consumption rates by factors known to be
related to fish consumption in the presentation of results can increase the usefulness of the final
UFCR estimates. For example, fish consumption is related to age, racial/ethnic background, gender,
and income (Birch et al., 2014; Razzaghi & Tinker, 2014; Soon et al., 2014; U.S. EPA, 2014;
Connelly et al., 2012; Holloman & Newman, 2010; Mahaffey et al., 2009; Burger et al., 1999).
Holloman and Newman, 2010, found that low-income African American women in Newport News,
Virginia, consume seafood at a subsistence fisher rate, even though they were not subsistence
fishers. Body weight may be important data to collect in order to estimate FCR normalized to body
weight (g fish consumed/kg body weight). Physical and sociodemographic characteristics of fishers
that may be important (depending on the study objectives) include: race, ethnicity, gender, date of
14
-------
birth, height, weight, pregnancy/lactation status for women, physical disabilities or medical
conditions, number of household members, occupation/employment status, income and education
level, language spoken at home, and urban/rural residence. The extent to which potential predictors
of consumption can be incorporated into the survey and used in the presentation of results depends
on the number of questions that can be accommodated in the chosen survey design and the budget.
2.3.2 HarcUo-Survey Populations
Populations may be hard to survey because, among other reasons, members of the population may
be: (1) hard to define or identify and thus to sample; (2) difficult to find or contact; and/or
(3) unlikely to cooperate with surveys. Survey errors that may occur with hard-to-survey populations
may be the result of sampling challenges (which may require special sample frames), noncoverage or
undercoverage (the sampling frame does not include all members of the population), nonresponse
(failure to obtain complete data from all selected individuals), and measurement errors (the
difference between the "true" value and the reported or measured value) (Smith, T., in Tourangeau
et al., 2014).
Populations or sub-populations are often hard to identify when there are no reliable lists of
population members and/or when there are few individuals among the larger population who meet
the criteria of interest for the survey. They may be hard to find if they have high rates of mobility or
have barriers that impede their accessibility. For example, barriers may include the need for
interviewers to travel long distances to reach the groups of interest, the prevalence of gated
communities within urban areas, and the use of caller ID (and refusal to answer unknown callers) by
many members of the population. Examples of groups unlikely to cooperate include those with little
or no interest in the subject matter, those who may be suspicious of the survey organization or
whomever it represents, or those with low investment in the community. Hard-to-survey
populations are often defined by multiple factors such as these (see Tourangeau et al., 2014).
Throughout this document, we have emphasized the importance of being as knowledgeable as
possible about the population of interest. This will enable the survey design to incorporate
techniques to minimize the impact that hard-to-survey groups may have on survey results. One
important method to identify and reach hard-to-survey populations includes outreach to
organizations working within the communities and with tribal, state, and federal partners that are
already in the community.
For fish consumption surveys, special consideration of sampling challenges should be given when
planning surveys of remote, possibly isolated populations. For example, American Indian/Alaska
Native populations have historically been undercounted (Tourangeau et al., 2014). Hard-to-reach
areas have often turned into unsampled areas due to their remoteness. Some locations in Alaska, for
example, are only serviced by float plane (Smith, T., in Tourangeau et al., 2014). Telephone surveys
may not be an option due to lack of coverage in these rural or frontier areas. For on-the-ground
interviewers, difficulty in locating households may be encountered due to streets with no names or
house numbers. This makes it difficult to verify sampled housing units. In addition, residents of
these areas may have high rates of mobility, on a regular or semi-regular basis, or tenuous
attachment to a household. Not only is this problematic for identifying household members, but
also for conducting follow-up interviews to obtain fish consumption data on multiple occasions, as
required by some survey methodologies. For household surveys that require an enumeration of all
15
-------
household members, rules for identifying household members must be clearly defined and conveyed
to field interviewers so they will know who is to be counted as a household member, whether or not
the individual is physically present at the time of screening.
2.3.3 Fishing Activities and Behavior
Seasonal variability in the amount and/or type of fish consumed by the population of interest
should be considered during the design process. This variability may be due to the availability of
fish, fish spawning patterns, weather, or local customs or traditions associated with different
holidays or times of the year. Seasonal variation may be a factor not only in the amount of fish
consumed, but also in the type(s) of fish eaten, and/or in how the fish is prepared. For example,
during the summer months, fish may be eaten fresh most of the time, but during the winter months
it may only be eaten in dried form.
The direct approach for capturing seasonal variations and accounting for them in estimates of fish
consumption is for data collection to span a full year. With direct estimation methods, a survey
conducted during months of high availability of fish would be expected to overestimate annual
consumption, and a survey conducted during months of low availability of fish would be expected
to underestimate consumption. Ideally, equal proportions of the sample would be interviewed over
fixed time periods until the entire sample had been interviewed. Unequal numbers of interviews per
time period can be accommodated by weighting approaches. Another approach might be to conduct
interviews during two periods, one of low consumption and one of high consumption. If it is not
possible for the data collection period to span an entire year, indirect methods (which model the
seasonal variation) may be used, but these approaches rely heavily on assumptions about the nature
of seasonal variation (see discussion of usual fish consumption rates, or UFCR, in Section 1.2.3.)
One approach that has been used to accommodate seasonal variation is to allow the respondent to
designate two "seasons"—a high and a low consumption season—and report on typical
consumption per species during each "season," along with the duration of the season (adding to 12
months). See, for example, the Suquamish Tribe, 2000.
Examples of other possible fishing-related activities and information that may be needed to meet
survey objectives include the following:
¦ Location(s) of fishing activities (specific sites, type of water body)
¦ Distance(s) of fishing activities from principal residence
¦ Seasonal and temporal distribution of fishing activities (total number of days per season,
which months of the year, overall or for each location)
¦ Fishing effort (hours/outing, hours/day, outings/month, days/month)
¦ Purpose for fishing (consumption, sport only: catch and release, etc.)
¦ Mode of fishing (e.g., nets, traps, hook and line, pier, shore, private boat, charter boat,
scuba)
¦ Type of fish caught
¦ Number of fish caught per outing
¦ Size of fish caught
¦ How long involved in fishing activities and consuming self-caught fish (e.g., new to
sport or years?)
16
-------
¦ What was the fate of the caught fish (e.g., released, consumed by household, sold, given
away?)
2.3.4 Preparation and Consumption Patterns
Fish preparation techniques may have a substantial impact on the levels of certain types of
contaminant exposure (e.g., PCBs, dioxins, and DDT) and, therefore, can be used as an additional
predictor of exposure levels (Forsberg et al., 2012; Zabik & Zabik, 1999; Salama et al., 1998; Zabik
et al., 1996; Zabik, et al., 1995; Voiland et al., 1991; Skea, et al., 1979). Preparation methods include
how the fish was cleaned or trimmed (e.g., was the fat trimmed away?), as well as how—or if—it
was cooked (e.g., baked vs. fried). Response categories to questions about preparation methods
should match local culinary practices and the types of preparation methods of interest for the study
objectives. The change in weight of the fish due to cooking (moisture loss, fat gain, etc.) varies by
cooking method. Thus, in order to convert as-consumed weights to raw weight, cooking methods
need to be collected.
Examples of possible preparation and consumption pattern information that may be needed to meet
survey objectives include the following:
¦ Source of fish consumed (e.g., locally caught, commercially obtained)
¦ Amounts of fish eaten per meal/day/week/month for each person in household (visual
cues are helpful to improve the accuracy of portion size estimates)
¦ Geographic and seasonal variations in consumption
¦ Parts of fish consumed (may vary with the species)
¦ Parts of fish used for cooking but not ingested (e.g., boiling of bones or fish heads)
¦ How the fish were prepared for cooking/eating (e.g., skinned, fillet, steak, shucked)
¦ How the fish were cooked if they were not consumed raw (e.g., baked, fried, steamed,
smoked)
¦ Fish consumed in mixed dishes such as fish stews
¦ Special cultural/ethnic practices in fish consumption and preservation (e.g., smoking,
canning, drying)
¦ Consumption of fish obtained from restaurants, supermarkets, fish markets, roadside
stands, from family or friends, or obtained in other ways such as tribal distribution
programs
¦ Consumption of shared fish (e.g., shared in general or at communal meals, feasts,
ceremonies, or other gatherings)
¦ Whether fish were frozen or preserved and eaten throughout the year, or eaten only
when fresh
2.3.5 Suppression and Heritage Rate Issues
A "suppression effect" occurs when a fish consumption rate (FCR) for a given population, group, or
tribe reflects a current level of consumption that is diminished from the level of consumption that
17
-------
the population, group, or tribe would actually consume in the absence of factors such as the
following, some of which may be relevant and important to consider in the design of surveys:
¦ Health-based suppression of consumption
— Reduction or avoidance due to actual or perceived presence of chemical
contaminants in the fish
— Reduction or avoidance due to fish advisories
¦ Reduced fish populations due to environmental changes (e.g., changes in water
conditions, alteration in fish habitat by dams and shoreline development)
¦ Reduced access to fish (due to fishing regulations such as creel limits or barriers to
access such as land ownership and shoreline development)
¦ Changes in social structure; reallocation of time from traditional lifeways to other
pursuits
¦ Imposition of laws or regulations reducing fish consumption
Suppressed fish consumption is of particular concern among some tribal populations since issues of
physical, spiritual, or social health have been linked to fish consumption (O'Neill, 2013; Donatuto &
Harper, 2008). Suppression may affect national fish consumption rates as well as rates among
smaller populations, especially tribal groups. Among all populations, there may be the desire to eat
more fish than is available or safe to eat, for example the joint FDA-EPA fish advice advises people
to consume 8 to 12 oz. of fish per week. EPA's "Human Health Ambient Water Quality Criteria and
Fish Consumption Rates: Frequently Asked Questions" (U.S. EPA, 2013), states that when setting
water quality criteria, "It is also important to avoid any suppression effect that may occur when a
fish consumption rate for a given subpopulation reflects an artificially diminished level of
consumption from an appropriate baseline level of consumption for that subpopulation because of a
perception that fish are contaminated with pollutants." Environmental standards utilizing
suppressed rates may contribute to a scenario in which future aquatic environments will support no
better than suppressed rates.
A "heritage rate" is the amount of fish that was traditionally consumed prior to non-indigenous or
modern sources of contamination and interference with the natural lifecycle of fish. A heritage rate
of consumption is generally extrapolated or reconstructed from information available in
anthropological or historical literature. In many cases, heritage rates may be the only practical way to
estimate unsuppressed rates — that is, free from the biasing influence of suppression effects, and may
be useful in establishing a baseline for legally protected fishing rights for fishing tribes. As discussed
in Section 2.1.1, if the resulting FCR estimates from the survey will be used for regulatory action
under the Clean Water Act, consultation with EPA is recommended.
Methodologies to assess if there is a suppression effect, to estimate the size of the effect, and to
estimate heritage rates are presented in Chapter 6.
2.4 Environmental Justice
The U.S. EPA defines environmental justice as "the fair treatment and meaningful involvement of
all people regardless of race, color, national origin, or income with respect to the development,
implementation, and enforcement of environmental laws, regulations, and policies. EPA has this
goal for all communities and persons across this Nation. It will be achieved when everyone enjoys
18
-------
the same degree of protection from environmental and health hazards and equal access to the
decision-making process to have a healthy environment in which to live, learn, and work."
In 2002, the National Environmental Justice Advisory Council (NEJAC) published a document that
discusses recommendations to EPA regarding environmental justice and fish consumption (NEJAC,
2002). The document and recommendations provided within address the question, "How should
EPA improve the quality, quantity, and integrity of our Nation's aquatic ecosystems in order to
protect the health and safety of people consuming or using fish, aquatic plants, and wildlife?" and
should be considered carefully by the survey team. The six recommendations provided in the
document are:
1. Require states, territories, and authorized tribes to consider specific uses, including the use
of the waterbody or waterbody segment for subsistence fishing, when designating uses for a
waterbody, and to set water quality criteria that support the specific designated use; provided
that where human health criteria are established based upon consumption of toxic chemicals
that bioaccumulate in fish, regulators should employ appropriate human fish consumption
rates and bioaccumulation factors, including cultural practices (e.g., species, fish parts used,
and manner of cooking and preparation) of tribes and other indigenous and environmental
justice communities using the waterbody; provided further that EPA should encourage and
provide financial and technical support for states, territories, and authorized tribes to control
effectively all sources, including both point sources and nonpoint sources, to achieve the
criteria
2. Work expeditiously to prevent and reduce the generation and release of those
contaminants to the Nation's waters and air that pose the greatest risk of harm to human
health and aquatic resources, including but not limited to persistent bioaccumulative toxics
(PBTs) (e.g., mercury, dioxins, and polychlorinated biphenyls (PCBs)) and other toxic
chemicals, and to clean up and restore aquatic ecosystems contaminated by pollutants
3. Protect the health of populations with high exposure to hazards from contaminated
fish, aquatic organisms and plants, and wildlife, including communities of color, low income
communities, tribes, and other indigenous peoples, by making full use of authorities under
the federal environmental laws and accounting for the cultural, traditional, religious,
historical, economic, and legal contexts in which these affected groups consume and use
aquatic and terrestrial resources
4. Ensure that fish and other aquatic organism consumption advisories are used by
regulators as a short-term, temporary strategy for informing those who consume and use
fish, aquatic organisms and plants, and wildlife of risks while water quality standards are
being attained and while prioritizing and pursuing the cleanup of contamination by
appropriate parties; agencies must evaluate and address such risks, and require risk-producers
to prevent, reduce, and clean up contamination of waters and aquatic ecosystems
5. Because many American Indian and Alaska Native (AI/AN) communities are particularly
prone to environmental harm due to their dependence on subsistence fishing, hunting, and
gathering, conduct environmental research, fish consumption surveys, and
monitoring, in consultation with federally recognized tribes and with the involvement of
concerned tribal organizations, to determine the effects on, and ways to mitigate adverse
effects on the health of AI/AN communities resulting from contaminated water sources
and/or the food chain
6. Consistent with the 1988 EPA Indian Policy for the Administration of Environmental
Programs on Indian Reservations, the federal trust responsibility to federally recognized
tribes, and federal policies recognizing tribal sovereignty and promoting self-determination
19
-------
and self-sufficiency, provide equitable funding and technical support for tribal programs
to protect AI/AN communities and tribal resources from harm caused by contaminated
water and aquatic resources and, until tribes are able to assume responsibility for such
programs, implement and require compliance with the federal environmental laws within
Indian country; provided that, in consultation with tribes, EPA should promptly develop
effective and appropriate regulatory strategies for setting, implementing, and attaining water
quality standards within Indian country; and provided further that, EPA should work with
Alaska Native villages to address the special circumstances that exist in Alaska and to protect
the health of Alaska Natives from environmental threats associated with their extensive
subsistence lifeways.
Resources can be found at https:/ /www.epa.gov/environmentaljustice and
https:/ /www.epa.gov/ejscreen. Additional guidance provided by EPA regarding working with tribal
nations can be found at https://www.epa.gov/tribal.
20
-------
3
SURVEY APPROACHES AND SAMPLE SELECTION
3.1
Introduction
This chapter presents key issues to consider when selecting a survey approach and sample design for
a fish consumption survey — Decisions about a survey approach and sample design. This is the third
of five basic steps in the design and development of a survey for estimating fish consumption rates
(as listed in Chapter 1). There is a need for budget-friendly, yet scientifically sound, approaches to
estimate consumption rates and calculate risk associated with the consumption of chemically
contaminated fish tissue (U.S. EPA, 1989). As states have increased their focus on this type of risk
assessment, the need for site-specific fish consumption surveys has become more evident. In this
chapter, we describe modes of data collection, sampling methods, sampling frames, and other
considerations when selecting an approach to allow calculation of consumption rates for the
population of interest. The discussions in this chapter assume the reader is familiar with elementary
concepts in survey sampling; for an overview of survey sampling, see Lohr, 2010.
It should be noted that the decision about a survey approach must be based on an overall
assessment of a number of factors including, but not limited to, survey objectives, characteristics of
the study population, the desired analytic approach, and budgetary factors. Sections 3.2 and 3.3
describe various types of surveys and sampling strategies that may be considered. At the end of this
chapter, a series of tables organized by budget/funding levels (Tables 3-5 through 3-7) provide
guidance as to when each approach may be the appropriate choice based on multiple independent
factors such as the need for a representative sample, required response rates, characteristics of the
population, prevalence of the behavior of interest, and others. These tables allow readers to take
various study objectives and population characteristics into account within a budgetary framework.
If tribal populations are involved, consultation with the appropriate tribal governments is
recommended in advance of the survey (see Section 3.3.2).
Approaches to conducting fish consumption surveys include single mode (e.g., telephone, web, mail,
in-person interview) and mixed-mode (e.g., mail to phone, mail to web) designs. A discussion of
each of these approaches is presented in the following subsections. Most of these survey approaches
can be used with any type of sampling frame, although some particular combinations may not be
feasible due to budgetary constraints or population characteristics (see Tables 3-5 through 3-7).
In-person interviews can be conducted at known fishing locations, at the respondent's home, or at
some other location (see, for example, CRITFC, 1994 and Michigan Department of Community
Health, 2007). An in-person survey typically results in relatively high response rates because of the
personal contacts made by interviewers, which may include some initial efforts to build the trust of
3.2
Types of Surveys
3.2.1
In-Person Interview
21
-------
the respondent prior to the actual interview. For example, advance letters can be mailed to
prospective survey participants to introduce the study and help establish credibility. In-person
interviews allow for the collection of more complicated information because the interviewer can
work with the respondent to obtain relevant information through a series of probes. Sometimes
visual aids are used (e.g., maps, show cards with response categories, or pictures of fish species,
portion sizes, etc.). Answers to survey questions can be recorded on hard-copy questionnaires or
entered directly into a computer database on a laptop or tablet computer used by the interviewer.
Interviewers are typically recruited for their ability to engage study participants, gain cooperation,
and accurately record responses. They must be trained in the standardized data collection procedures
developed for the survey, and informed about the overall goals of the research. Because of the
opportunity for the interviewer to directly interact with the respondent, administer appropriate
probes, establish time and location suitable for the respondent, and ensure that all questions are
answered, in-person surveys typically result in the highest response rates and obtain high-quality
data. They can, however, be quite expensive.
A creel survey is a specialized form of personal interview to obtain information about fish that have
been caught (see Zale et al., 2013). Creel surveys can be roving (meeting the fishers where they are
fishing), or they can be stationary (meeting fishers where they return to the dock). In addition to
asking a specific set of questions about fishing activity and fish consumption behavior, the
researcher can try to identify and/or measure fish in the fisher's possession (the "creel"). The creel
survey can be conducted at access points (e.g., boat ramps, docks), along the shoreline, or on the
water from a boat. Fish consumption information obtained from the fishers is hypothetical since
actual consumption has not yet occurred; however additional questions can be incorporated into a
creel survey such as a 24-hour fish consumption recall.
A creel survey can also be used to develop a contact list of fishers who can later be sent a mail
survey or called for a telephone survey. This method provides confidence that actual fishers are
being reached, rather than using random digit dial screening of households to locate fishers.
Telephone surveys contact selected respondents by telephone, using either their landline phone or
their cell phone (for example, see Imm et al., 2005). Depending on the survey design, a brief
telephone interview may be conducted to screen participants before inviting them to complete a
more extensive telephone interview. Respondent answers may be entered either onto a hard-copy
questionnaire or directly into a computer database by interviewers trained on the study content and
procedures. To successfully reach a representative sample, a telephone survey should include cell
telephone numbers as well as landline numbers.
3.2.2
Creel Survey
3.2.3
Telephone Survey
22
-------
3.2.4
Mail Survey
Mail surveys consist of a self-administered questionnaire that is mailed to respondents (see
Minnesota Department of Health, 2012). Mail surveys are an inexpensive means for covering a
larger or widely dispersed population of interest. The development of address-based samples has
also allowed mail surveys to serve as effective screening tools for identifying subpopulations (Brick
et al., 2012). Address-based samples can also be used to survey very small, targeted geographic areas.
Because they are self-administered, mail surveys lack the advantages offered by computerization of
the instrument (available with in-person, telephone, or web/mobile modes), which can accurately
navigate the respondent or the interviewer through the survey instrument and conduct real-time
editing of entered data.
Web surveys, like mail surveys, are self-administered survey tools that can collect recent and past
fishing or consumption activities. However, web surveys offer the advantage of computerization,
which allows for a more complex survey instrument to be administered. Web surveys may also take
advantage of graphics such as pictures of different species of fish. Selected individuals are generally
mailed information for accessing the web-based questionnaire (including a personal identification
number, or PIN), or sent an email with a URL linking to the survey instrument. A limitation is that
not all members of the target population may have web access.
Mobile surveys are similar to web surveys in that they are usually administered over the Internet, but
can also utilize custom applications (or apps) that can be downloaded to a mobile device. Web
surveys are often optimized for use with mobile devices, since for some populations a mobile device
may be their primary access to the Internet (Pew Internet and American Life Project, 2012). Mobile
surveys can be designed to collect recent and past consumption or fishing activities, but can also be
used by in-person interviewers to capture other information—for example, photographs of fish, or
to map locations where fish were caught.
Mixed mode surveys take advantage of multiple survey modes in order to maximize response to the
survey and, in many cases, minimize costs. There are a number of approaches to combining
different data collection modes. One example that minimizes costs and maximizes response is to
conduct a mail survey to collect fish consumption data of one (or more) household members. After
the mail survey protocol has been completed (i.e., all nonresponse mailings have been completed),
nonrespondents to the mail survey are contacted by interviewers to collect consumption information
through a telephone interview. Another example that can minimize costs is to begin with a
telephone mode. After the telephone contact protocol has been fulfilled (i.e., a specified number of
contact attempts across times and days), nonrespondents are visited by an interviewer to collect fish
consumption information for a selected person or household. A common mixed mode survey for
dietary data collection with the goal of collecting multiple 24-hour recalls is to do the first 24-hour
3.2.5
Web or Mobile Survey
3.2.6
Mixed Mode Survey
23
-------
recall in person and the follow-up 24-hour recalls by telephone. For more on mixed-mode survey
designs, the reader is directed to de Leeuw, 2005.
When utilizing mixed mode protocols, it is common practice to begin with the most inexpensive
modes, proceeding to the more expensive modes in a sequential order. The researcher should be
cautioned against offering respondents a choice of survey modes simultaneously (concurrent mixed
mode protocols). Research has shown that this can lead to reductions in overall response compared
to single mode designs (see Millar & Dillman, 2011, for an example with web and mail modes).
3.3 Types of Samples
A sample is typically defined as a subset of a larger population or group and through careful design,
selection, and analysis, survey results from the sample can be representative of the population of
interest. General population surveys, or probability surveys, are distinguished from other types of
surveys by the fact that each unit (e.g., individual, household, etc.) in the population is given a
known probability of selection that allows statistical inference to be used to generalize the survey
results to the larger population — something that cannot be done with nonprobability samples. The
first step in defining a sample is to decide whether the sampling unit will be households, individuals,
events, or something else. For fish consumption surveys, the sampling unit is typically the individual
consumer. When sampling (rather than taking a census of) the population, it would be inappropriate
to consider all members of a household in a particular subgroup (e.g., children) as independent
observations of the population because of obvious "household effects." If each individual in every
household were considered an independent case, the consumption estimates for the population
would be skewed toward those of larger families. If the individual is the sampling unit, an
appropriate design might be to randomly select a household and then randomly select a household
member within the target population.
With some sample designs (e.g., address-based sampling, telephone sampling, and area probability
sampling), the ultimate unit within the sampling frame is the household rather than the individual. If
the unit of analysis for a particular fish consumption survey is the individual, and one of these
sampling frames is to be used, an additional stage of sampling is required. That is, after the sampled
household has been identified, individuals must be sampled from among the eligible persons in the
household. There are a number of options for choosing a method for within-household selection.
Further details may be found in Gaziano, 2005, which provides comparisons of various approaches
for selecting survey respondents from among all household members.
As mentioned previously, the type of sample must be determined early in the design process, as
many aspects of the sample design depend on the specification of the target population. For general
population surveys (i.e., those in which the sample population is to be representative of—or
generalizable to—for example, all adults, or all persons age 5 and older), the considerations and
design choices are different from surveys that focus only on specific population subgroups (e.g.,
high consumers, women of child-bearing age, recreational fishers, etc.). General population surveys,
or probability surveys, are distinguished from other types of surveys by the fact that each unit (e.g.,
individual, household, etc.) in the population is given a known probability of selection that allows
statistical inference to be used to generalize the survey results to the larger population—something
that cannot be done with nonprobability samples.
24
-------
In general population surveys, nearly every household is expected to contain at least one member of
the group of interest. Although there may also be the need to develop subgroup estimates (e.g.,
estimates by gender or age group), these goals should be balanced against the goal of obtaining
estimates for the population as a whole. In contrast, when the survey aims to study only targeted
populations (e.g., women of childbearing age), some proportion of households will not contain a
member of this target population. These surveys typically involve a significant screening effort to
determine whether households contain members of the target population. Screening for subgroups
increases cost and requires a longer period for data collection.
In addition to defining the scope of the target population, another important aspect that
characterizes the target population is the geographic area of interest (e.g., regional vs. a particular
state vs. coastal counties in a particular state). The geographic area of interest may be defined, for
example, by a need to assess consumption of fish caught in certain bodies of water.
3.3.1 Choosing a Sampling Frame
Depending on the target population, as well as overall survey objectives, there are a variety of
possible sampling frames to choose from, each with their own strengths and weaknesses. Each of
these is discussed in this subsection. In many cases, the choice of data collection mode will influence
the type of sampling frame that is used. The quality of the sampling frame will depend upon the
statistical methods used to select the sample, and the coverage, i.e., the degree to which the target
population is covered by, or included in, the sampling frame. Researchers with limited resources
should ensure that they choose the highest quality sample to meet their goals. It is important to
choose a sampling frame that allows for a sample that is representative, while also employing a
survey methodology that minimizes potential biases.
Procedures for obtaining and evaluating sample frames should be developed as a part of the study
design. For telephone samples, address-based samples, or area probability samples, this will generally
be an empirical question. One important aspect of the quality of these frames is coverage; i.e., the
degree to which the target population is covered by, or included in, the sampling frame. All of the
sampling frames described in this subsection, if used properly, can be used to generate probability
samples. With appropriate use of sampling weights during the analysis, they can be representative of
a larger population, thus providing results that are generalizable to that population (see discussion of
weights in Chapter 7).
"Dual-frame" samples are developed by combining two of the sample types described in this
section. A dual-frame sample is used when one type of sampling frame offers superior coverage of
the target population but for which sampling and/or data collection are expensive, while another
type of frame offers inferior coverage of the target population but at a lesser cost. Dual-frame
sample designs, and the computation of estimates from these, can be complex, and requires careful
consideration of the following:
¦ Allocation of the sample between the two frames
25
-------
¦ De-duplication between the two frames (i.e., persons included on both frames), either
prior to sampling, through identification of persons in the "overlap," or by screening
out persons in the overlap sampled through one of the frames
¦ Any differences in data collection methods/procedures for the two frames
¦ Combining estimates from the two frames (which involves consideration of differences
in response rates and coverage between the two frames)
Despite these complexities, dual-frame samples may serve particular populations well. For general
population fish consumption surveys, for example, considerable efficiencies may be gained by using
one sampling frame for a specific subpopulation (e.g., lists of licensed fishers if lists are available that
truly represent the subpopulation of interest), combined with a broader sampling frame (e.g.,
address-based sampling) for the remainder of the population. This type of sample will likely require
the expertise of a trained sampling statistician to successfully implement.
Random Digit Dialing (RDD) Sample and Dual-Frame (RDD with Cell Phone Sample)
A random digit dialing (RDD) sample is a sample of telephone numbers. Historically, RDD samples
have been samples of listed and non-listed landline telephone numbers. Every working number in
the population has an equal probability of selection. However, as the proportion of households
having only cell phones has increased (estimated to be 43 percent during the first half of 2014; see
Blumberg and Luke, 2014), survey researchers have increasingly turned to methods that include cell
phone numbers as well as landlines in the sample.
Using both landline and cell phone numbers is a common example of a "dual-frame" sample (as
generally described on the previous page). While the dual-frame telephone approach addresses the
undercoverage associated with landline RDD sampling frames, it may exacerbate the problem of
declining response rates to telephone surveys, as response rates for cell-phone components are
generally lower than their landline counterparts (Brick et al., 2007; Link et al., 2007).
If a dual-frame RDD sample is desired, households with both landline and cell service may be
sampled from either frame. There are two general approaches for accounting for a household's
frame membership in dual-frame telephone surveys: (1) a screening approach in which landline
households are screened out of the cell sample so that each household can, effectively, be included
in the sample through only one frame; and (2) an overlap approach in which households are retained
in the sample regardless of frame membership and dual-frame estimation methods are used. Unless
a general population survey is required, the extensive screening required to identify target
populations (e.g., high consumers of fish or pregnant or lactating women) is not feasible due to the
high costs. Instead, the overlap approach can be used.
A key consideration is how to allocate the sample between the two frames. State- and county-level
telephone service estimates are now available through at least one vendor, and these may be used to
determine an efficient allocation among landline and cell numbers. State- and county-level telephone
service estimates may be available from other vendors that provide telephone samples to survey
organizations, and these may also be used.
Another consideration with dual-frame sampling is that while vendors maintain information that
allows telephone exchanges (i.e., the first six digits of a telephone number) to be linked to
26
-------
geographic areas, these linkages are known to involve error, so for surveys with geographically
restricted target populations, it is necessary to confirm place of residence prior to completing the
interview. To minimize bias from excluding people who have moved into an area while retaining
their previous phone number, it is important to work with vendors that can identify some of these
"foreign" numbers for inclusion in the sampling frame.
Pros/Cons: An advantage of a telephone sample is that it can cover a large geographic area and can
be representative of the population (with telephones). However, it may be biased if a significant
percentage of the population does not have a telephone. Additionally, visual aids cannot be used
unless the address is gathered at time of the screening call and the visual aids are mailed before the
interview. Due to the limitations of high cost (to identify sufficient numbers of subgroup members)
and expected low response rates (due to telephone contacts) in general for fish consumption
surveys, the use of RDD samples may be appropriate only when leveraging existing survey efforts
such as the Behavioral Risk Factor Surveillance System (BRFSS) (see discussion of leveraging in
Section 3.10).
Intercept Sample
If properly implemented, an intercept sample (such as an in-person creel survey or an on-the-bank
survey) can produce a probability sample when the survey population is restricted to people who
must visit one of a limited set of locations. Such a design is venue- and time-based, and could be
used to sample persons known to frequent certain types of locations; for example, to sample fishers
who fish in a specific river at certain points in time. To ensure good coverage of the target
population, intercept surveys should be conducted on different dates, at different times of day, and
at different locations. Locations (often, open-access points) are identified, and for each location to
be visited, survey days/times are randomly assigned and data collection (in the form of interviews
with fishers) occurs during those days/times. At a given location during the sampled data collection
period, there may or may not be sampling of persons. Often, a "take all" approach is used; in other
words, every fisher accessing the body of water at the given location is sampled. Intercept surveys
may also contain a separate data collection effort that counts fishers using the body of water. These
counts are used in estimation to attribute proper coverage to the locations and the survey
days/times. If subjects are to be interviewed more than once, contact information can be collected
to allow for subsequent interviews to be conducted by mail or telephone.
An intercept survey alone would not be a desirable approach for a general population survey (e.g., a
population that includes fishers and non-fishers) because of its very limited coverage. However, it
may be a useful approach for sampling licensed and unlicensed fishers who fish in a specific body of
water. Additional information to be obtained would include whether or not the fishers and/or
anybody else, such as their family, actually consume the fish caught, or if they practice "catch and
release" only. For further discussion of methods and potential challenges in conducting creel
surveys, see Kinnell et al., 2007; Ray et al., 2007a; and Ray et al., 2007b. Also see Price, Su, & Gray,
1994.
Pros/Cons: The advantages of this design are that response rates are expected to be higher than
telephone or self-administered modes because of the in-person contact. However, if interviewers are
not fluent in languages/customs of ethnic groups likely to be encountered, or if interviewers are
perceived to be associated with fish and wildlife or law enforcement groups, there may be an impact
27
-------
on response rates. Individuals interested in fishing may not be interested in providing information.
Creel surveys allow for a high degree of flexibility in the use of visual aids, and screening costs are
low. Other disadvantages include difficulties in making the results representative of a larger
population, and the high potential cost of covering large geographic areas. Additionally,
consumption rate information gathered from a traditional creel survey may be hypothetical because
consumption of the current catch has not yet occurred, and the fish may be consumed by persons
other than the fisher. Conducting a 24-hour fish consumption recall during the creel survey could
eliminate this issue.
List Sample
In some cases, it may be possible to identify a list that contains a significant portion of the target
population. For example, if recreational fishers are the target population of interest, lists of persons
issued fishing licenses might be expected to cover a significant portion of the population. Similarly, a
list of tribal members, or health registry lists (e.g., birth records), may serve as fairly complete listings
for those subpopulations (Connelly et al., 2013). In these cases, such lists could serve as the
sampling frame alone (with recognition of the implications of undercoverage) or could be used in
combination with another approach that yields greater coverage (e.g., area probability sampling).
However, access to this information may be limited to tribal members. Tribes may not participate in
a survey conducted by an outside entity unless information is protected, and the tribe has assurances
that the objectives of the survey are in their best interest. Tribes may also wish to have control and
input on the custody of data and analysis and interpretation of results. These issues will likely require
tribal government approval of the use of tribal enrollment lists.
The information available on each list may influence or even dictate the data collection approach. It
will be important to review what information is available for the list frame of interest before
deciding on a data collection approach. Pilot testing (e.g., checking the completeness and/or validity
of the list) may be helpful prior to using the list for sampling purposes. For example, if a particular
list frame includes addresses for the entire frame, but telephone numbers for only a portion of the
frame, then data collection methods utilizing telephone contact may not be easy to implement.
Examples of useful information are:
¦ Name
¦ Address
¦ Telephone number
¦ License type (fish/water type; lifetime/temporary/seasonal/yearly; other)
¦ Age .
¦ Date license was obtained (can help determine whether information is current)
¦ For tribal enrollment lists, date each person was added to the list or when the list was
last updated, as this is a possible indicator of whether information is current
With list sampling frames, there are list maintenance issues to consider. A common issue involves
duplicates. For example, if a list of persons issued fishing licenses is used but a survey will be done at
the household level, it may be necessary to remove duplicates. Another example of a situation in
which de-duplication is necessary is one in which a license frame includes duplicate listings for
persons issued more than one license (e.g., if licenses are issued for different bodies of water, and a
single fisher holds multiple licenses). Also, if the list is actually a compilation of lists from different
28
-------
sources, it may be necessary to both remove duplicates and reformat the lists to achieve a consistent
format. In addition, steps should be taken to ensure that the most up-to-date lists are used for the
sampling frame.
Pros/Cons: The main advantage of a list sample is the potential for easy targeting of the population
of interest. Disadvantages include the possibility of undercoverage because some people may fish
without a license and the difficulty of developing high quality, unduplicated, and up-to-date lists.
Area Probability Sample
An area probability sample uses multi-stage sampling, with geographically based frames at each
stage. The first stage involves the creation of primary sampling units (PSUs); for national surveys,
PSUs are often counties or groups of geographically contiguous counties. The second stage involves
the creation of secondary sampling units (SSUs) within sampled PSUs; for national surveys, SSUs are
often census tracts or groups of census blocks. Within sampled SSUs, lists of households are
compiled, and households are sampled in the third stage. Typically, the use of an area probability
sample requires "boots on the ground" so that data collectors (i.e., listers) can compile lists of
dwelling units (households) within a selected area. Quality concerns arise when listers miss dwelling
units. This can be due to several factors, such as new areas of construction, hidden dwelling units
(e.g., behind another dwelling unit, or above a business), or unidentified multi-unit structures (e.g., a
house where basement residents are a separate household unit). This may be followed by sampling
persons within households, which usually requires an in-person visit to the household. Although the
stages discussed here describe a "typical" area probability sample, more or fewer stages of selection
may be warranted, depending on the target population.
Because it results in a geographically clustered sample, area probability sampling is generally used
when data collection is planned to be in-person. For example, if the plan is to administer the
interview in-person or to use in-person interviewing to follow up nonrespondents, having a
geographically clustered sample may be necessary from a cost and field management perspective.
Pros/Cons: The advantages of using an area probability sample are that it has the highest coverage
of the population; thus the survey could be designed to be representative of the area sampled, and it
has the highest expected response rate. Disadvantages include the high cost and the probable
complexity of sample clustering. Area probability sampling is relatively expensive, due to the need
for in-person data collection, including household screening for the target population.
Address-Based Sample (ABS)
An address-based sample (ABS) is selected from sampling frames maintained by vendors that
originate from the U.S. Postal Service's Delivery Sequence File (DSF) or Computerized Delivery
Sequence (CDS) file. ABS frames may be constructed within areas as the last (or penultimate) stage
of selection in an area probability sample (discussed in the previous subsection), or they may stand
alone as a frame for a single-stage sample. The former approach may be used if it is necessary to
cluster the sample within relatively small geographic areas; the latter approach may be sufficient if
such clustering is not necessary.
29
-------
Address-based samples rely on lists of addresses maintained by vendors. Providers of these lists
generally strive for high-quality lists, but coverage issues can arise in certain areas. In the case of mail
survey designs, households that rely on P.O. boxes for mail will be missed. If address-based samples
are used for telephone approaches that match a telephone number to the sampling frame, the
number matched may not always be to the address sampled. It will be necessary to verify the
respondent's address to ensure the sampled address was contacted.
Because ABS sampling frames consist of mailing addresses, this type of frame lends itself well to a
data collection approach that uses mail for the initial (or sole) contact with the household. However,
if there are concerns about literacy among the target population, they can be addressed in the ABS
design by using a multi-mode data collection approach or hybrid approaches (e.g., a design that
involves a very simple mail screener followed by an in-person interview). With these variations, mail
can be used to screen and identify the target population, or the screening could be done by
telephone (using telephone numbers matched to addresses, which can generally be obtained for
about 40 to 60 percent of addresses) or in-person. The screening survey may be completed by any
adult household member, reducing the impact of low literacy levels. The substantive survey can then
be administered either by telephone or in-person. A mail screener administered to an ABS sample
will yield greater coverage than a telephone screener administered to an ABS sample (due to the
inability to obtain accurate telephone numbers for some proportion of sampled addresses), at much
lower cost than in-person screening. However, it may still be necessary to include telephone and/or
in-person screening for nonresponse follow-up to the mail screener.
Pros/Cons: Advantages of an ABS are high coverage, moderate to high response rates (depending
on mode of collection), saliency of the topic for those surveyed, and the ability to cover a large
geographic area and be representative. Disadvantages may include high screening costs to reach a
potentially hard-to-find target population (e.g., high fish consumers) and the potential for
undercoverage in some rural areas. A survey using an ABS frame could be conducted through mail,
a mail-to-web hybrid, or web only (with a mail invitation). An in-person or mail-to-phone hybrid
ABS survey would be more expensive.
3.3.2 Challenges Associated with Some Sampling Frames
High Fish Consumers
If the goal is to estimate fish consumption rates for high fish consumers, it is necessary to take
advantage of sampling frames that are likely to include high concentrations of this population. State
lists of fishing license holders are common frames for consumption surveys since these are likely to
target high fish consumers (Ashford et al., 2009). For subpopulations or targeted groups, such as
tribes, there may be tribal lists or tribal enrollment records that could be used for sampling (with the
proper approvals). The first step is to determine what lists are available and how access can be
obtained. For some states, lists of license holders are publicly available, while for others, access may
only be allowable through a state agency (e.g., fishery department, or fish and game agency). In some
cases, state legislatures may restrict the release of personal information from license registries.
30
-------
Many states have multiple licenses and requirements for who needs a license. Also, some states have
"sportsmen" licenses that cover both hunting and fishing—further complicating the identification of
fishers because avid hunters may only be occasional (or non-) fishers. There may be separate licenses
for fresh water fishing, saltwater fishing, and/or shellfish. Further, there may be a primary and
secondary license (e.g., conservation and fishing license; or fishing license and gear permit), and it
may not be necessary to be a holder of both. For fish consumption surveys, it may be important to
sample from multiple license lists, as consumption may vary by license type (Katner et al., 2011). In
such instances, procedures to account for duplication across lists should be developed. Finally,
license requirements may vary by residency or age. For example, senior citizens or tribal members
may be exempt from license requirements or may not be required to purchase all types of licenses.
Knowing this will help inform how well the license frame available will represent the population of
interest.
Small, Isolated Populations and/or Targeted Geographic Areas
As discussed previously, surveys of smaller and/or more isolated populations present special
challenges, including those discussed in Section 2.3.2.
If a specific area or body of water is a targeted area of interest, license requirements for that area or
body of water should be reviewed. For some areas or bodies of water, there may not be a
requirement to be a license holder and other frames or approaches may be necessary (e.g., see
Kinnell et al., 2007).
Access to tribal lists or enrollment records may require cooperation and approval from the Tribal
Council for the tribe (or tribes) of interest. If there are multiple tribes, cooperation and approval will
be needed for each participating tribe. Tribes may have study review panels whose approval is
needed before a Council will consider a study. It is important to understand the tribal approval
process for each tribe whose participation is desired and the likely timelines for approval early in the
planning stages of a study.
For larger geographic areas, approaches not requiring face-to-face contact (e.g., phone and mail
surveys, diaries) could be more appropriate. Available staffing and time resources are also important
considerations in selecting the survey approach since, for example, multiple interviewers can cover
larger geographic areas simultaneously.
Characteristics of the Source of the Fish
The decision about which survey approach to use can depend on the source of the fish, certain
characteristics of the fish, and how they are harvested. If the survey is mainly interested in self-
caught fish, three important characteristics of fish sources are: (1) the number of access points;
(2) the fishing pressure (i.e., the amount of fishing activity); and (3) the size of the geographic area.
Access points refer to fishing locations for shore fishers (e.g., beach, riverbank, boat dock, fishing
pier, etc.) and boat ramps for offshore fishers, as well as parking lots or preserve entrances where
fishers might begin their activities. In locations with multiple access points, off-site approaches such
as telephone surveys, mail surveys, and/or diaries may be preferred. An exception to this is the
31
-------
roving creel survey, an on-site approach that can also yield good results in fishing areas with many
access points, although a creel survey may not be representative.
In fishing areas with high fishing pressure, mail surveys, personal interviews, and access-point creel
surveys may be effective because fishers are concentrated in relatively small areas. Roving creel
surveys, in which the interviewer moves from fisher to fisher and sometimes from site to site, are
more applicable to areas with low fishing pressure, where ample time is available for instantaneous
counts and for interviewing all fishers.
If the survey objectives involve all sources of fish (including purchased fish and/or fish consumed in
restaurants), the researcher must take this into consideration in the selection of the sample design
and survey approach.
Seasonality
Ideally, a survey design would include observations throughout the year to capture information on
practices across all seasons. There are several different ways to accomplish this. One way is to design
a longitudinal survey to obtain consumption data at least four times in a year (i.e., once each season).
This can be a difficult design to implement, as maintaining contact with the same respondents
throughout the year could be costly, requiring rigorous case management and tracing and, ultimately,
may result in too much sample loss to meet survey objectives. Another way to get data throughout
the year is to divide the sample into smaller groups and begin the survey at different points
throughout the year, as discussed in Section 2.3.3. This would provide snapshots of consumption
patterns throughout the year. For the analysis, seasonality can be accounted for analytically by fitting
terms for seasonal trends, although care must be taken to ensure sufficient sample sizes to cover
each season.
3.4 Benchmarking
For general population surveys (and surveys of some specific subpopulations), it may be useful to
benchmark the sampled population to external totals (e.g., population totals from the decennial U.S.
Census or the American Community Survey). To the extent that such data are available from other
studies for the subgroups of interest, such benchmarking can shed light on potential biases (due, for
example, to undercoverage or differential nonresponse among subgroups), and may be used in
adjusting for those biases. For example, a state survey might consider low-income women to be of
particular interest. In this case, it would be useful to know how closely the percentage of low-income
respondents matches the state-wide percentage of low-income women in order to be able to assess
representativeness. To facilitate such benchmarking, it is important to plan for it in the survey design
phase, to ensure that appropriate information is captured in the correct format (e.g., in a manner
consistent with the way the information was captured in the external source). Benchmarking is also
referred to as "post-stratification" or "raking." Additional information on benchmarking can be
found in Zhang, 2000; Oh and Scheuren, 1983; and Holt and Smith, 1979.
32
-------
3.5 Determining Required Sample Sizes
The sample size is the number of individuals selected from the sampling frame to be contacted for
the survey. Decisions about sample size ultimately influence the analytical power of the survey and
must take into account some assumptions about expected response rates. The term "respondents"
refers to the people who complete the data collection and for whom data are available for analysis.
The difference between the sample size and the number of respondents is determined by the
response rate and rate of being in-scope (e.g., if the survey is of tribal members, the percentage of
people living in the area that are members of the tribe).
The target sample size depends on many factors including how individuals are selected, how each
person's data are collected and analyzed, what is to be estimated from the data, and the desired
precision of those estimates. Precision requirements are generally specified as the largest acceptable
standard error or coefficient of variation (CV) of the estimate in terms of the ability to detect
differences between subgroups or between estimates over time, or, simply, the ability to report a
single estimated consumption rate with adequate precision. When specifying the precision
requirements, it is important to consider which subgroups (if any) the sample should be designed to
support. That is, do the specified requirements apply only to the sample as a whole, or do they also
apply to particular subgroups? Separate precision requirements may be specified for different
estimates, such as for mean fish consumption and for the 90th percentile of fish consumption, and
for the whole sample or specific subgroups. Then, a different sample size would be calculated for
each precision requirement. Often, desired precision requirements are re-examined (and adjusted)
after preliminary sample sizes have been computed (based on preliminary precision requirements)
and assessed in light of the budget for the survey.
The detailed process of determining how to select individuals for the survey is referred to as the
sampling design. In any survey, nonresponse can be a major issue. Some individuals may not want to
complete the survey or may not be located. When selecting people for the survey, an adjustment
must be made to account for these expected nonrespondents. For example, if the required number
of completed interviews is 1,000, but only 50 percent of those sampled actually complete a survey,
then 2,000 individuals must be sampled and contacted to achieve the target number of completed
surveys. Similarly, if some proportion of the sampled group is expected to be ineligible for the
survey, this must also be taken into account when determining the sample size. For example, if the
survey is interested only in women of a certain age, and this characteristic cannot be determined
ahead of time from the sampling frame, a much larger number of individuals should be sampled to
ensure that a sufficient number of completed interviews for the target group will ultimately be
available for analysis.
If a list of the target individuals is available, individuals from the list can be randomly selected,
contacted, and interviewed. This process is called simple random sampling. However, a list is often
not available. Alternatively, individuals can be selected by first randomly selecting communities or
zip codes, then randomly selecting small areas within communities or zip codes and interviewing all
individuals in the selected small areas. This process does not require a master list and, for in-person
interviews, may be more cost effective. This is an example of cluster sampling (the communities and
small areas represent clusters of people). The estimates from cluster sampling are generally not as
precise as estimates from a simple random sample. The relative precision is represented by a "design
effect." When calculating the target sample size, the design effect is a factor that accounts for how
33
-------
the respondents are selected. A design effect of 2.0 means that, compared to using simple random
sampling, twice as many individuals need to be interviewed to get estimates that have the precision
that would be obtained if simple random sampling had been possible and was used. Simple random
sampling has a design effect of 1.0. Since simple random sampling is rarely possible, the following
discussion assumes a design effect of 2.0 as an approximation of the design effect that might be
found in fish consumption surveys. Section 3.5.1 provides additional guidance for specifying the
design effect. For more detailed information on sample designs and design effects, see Lohr (2010),
or contact a statistician.
Another consideration for specifying the target sample size is how the data, in this case fish
consumption data, are collected and analyzed. Two basic approaches for obtaining usual intake rates
are:
1. Contacting each person two or more times and, on each contact, collecting information
on recent fish consumption
2. Contacting each person once and collecting as much information as possible on long-
term fish consumption
The first approach is preferable from an analysis perspective because it reduces recall bias
(respondents can more accurately remember recent, as opposed to long-term, fish consumption) and
because it allows for correcting the estimates for random differences between interviews. However,
since this approach requires more complicated statistical modeling and more respondent contacts,
the second approach may be more feasible if resources are limited. Different analyses are required
for these two data collection approaches, and the equations used to estimate the target sample size
depend on which approach is used.
Finally, the target sample size depends on what is to be estimated and the desired precision of those
estimates. Precision can be expressed in different ways. A confidence interval and a standard error
are measures of precision. The coefficient of variation (the standard error divided by the estimate) is
a measure of relative precision. In most applications, a 95 percent confidence interval can be
thought of as the shortest interval which, across independent surveys, has a 95% probability of
covering the true value being estimated. In many applications, the 95% confidence interval can be
adequately approximated by the estimate plus and minus two standard errors. Using the formulas in
Sections 3.5.2-3.6 for calculating the required number of completed responses (N) requires
specifying the desired length of the 95 percent confidence interval for the estimate. Once the desired
confidence interval length (L) is specified, the desired half width is H=L/2, the desired standard
error is se = H/2, and the desired CV is desired standard error divided by the expected estimate.
The survey data might be used to estimate the mean of usual fish consumption across the
population or the 90th percentile of usual fish consumption. The 90th percentile is the level of fish
consumption that is greater than 90 percent of the target population and less than 10 percent of the
population. Some parameters can be estimated more precisely than others. In most cases, the mean
can be estimated with better precision than an extreme percentile, such as the 90th percentile. A
confidence interval for mean fish consumption will usually be narrower than a confidence interval
for the 90th percentile. Increasing the sample size increases the precision of the estimates, resulting
in shorter confidence intervals. As a result, estimating the mean to a desired precision may require
fewer respondents than estimating the 90th percentile with the same precision. In general, increasing
34
-------
the sample size by a factor of four reduces the standard error, the coefficient of variation, and the
width of the confidence interval by a factor of two.
3.5.1
Design Effect
The design effect is an adjustment factor in the calculation of the sample size that is primarily
determined by how the respondents are sampled. The design effect for a simple random sample
from a list is 1.0. For other sample designs, the design effect will typically be greater than 1.0
(although stratified samples may yield design effects less than 1.0). Although design effects can be
very large, for most practical applications involving well-designed samples, design effects will
generally be less than 4.0. Detailed information for specifying the design effect is often not available.
As a result, it may be necessary to consult a survey statistician for guidance and recognize that any
estimate of the design effect is imprecise.
The best estimates of the design effect for the planned survey are obtained from historical surveys
of the same (or a similar) population using the same (or very similar) sample design and collecting
similar information. Thus, if a survey is conducted every three years using the same sample design,
the sample size calculations for the second and later surveys can benefit from the estimated design
effect from the earlier surveys. Software specifically designed for the analysis of survey data will
generally provide estimates of the design effects. The design effect can be somewhat different for
each parameter being estimated. Therefore, if a previous survey does not have estimates of the
parameter of interest, a fallback would be to use an average design effect across several related
parameters.
Suppose, based on the design effects from an historical survey, it is decided to double the sample
size while keeping the same sample design. Doubling the sample size while keeping the same design
is equivalent to repeating the survey twice. For random samples from a list, this corresponds to
randomly selecting twice as many list items (perhaps names or addresses) from the list. For cluster or
area samples with PSUs (clusters) and SSUs (within cluster items), this corresponds to selecting
twice as many clusters (PSUs) while retaining the historical within cluster sample selection
procedures.
In general, we recommend contacting a survey statistician to evaluate the design information
available from historical surveys and estimate an appropriate design effect for the planned survey.
The statistician can use information from historical surveys and make adjustments for differences
between these and the planned design to provide guidance on what design effect to use and on other
aspects of the survey design.
When the survey design requires contacting each person two or more times and, on each contact,
collecting information on recent fish consumption, the sample size calculation requires an estimate
of the expected probability of reported fish consumption on any one contact. Call this probability P.
3.5.2
Sample Sizes for Multiple Contact Surveys
35
-------
Higher values of P result in smaller sample sizes. P depends on the sample design. For any particular
type of fish consumption, higher values of P can be obtained by defining the surveyed population in
a way that includes those with fish consumption and excludes, as much as possible, those without
fish consumption. P using a 7-day recall will be higher than when using a 24-hour recall, though a 7-
day recall results in data with more measurement bias.
P can often be estimated from previous surveys of similar populations using similar data collection
protocols. Since risks from fish contamination are particularly important for those with high fish
consumption, using the proportion of the previous surveyed population that consumed fish may
provide a conservative sample size estimate. Given different proportions of fish consumption in
different demographic groups, the proportion for a demographic group with less frequent fish
consumption might be used.
For example, using the 2007-2012 National Health and Nutrition Examination Survey (NHANES)
data, the proportion of adults reporting any fish consumption in a 24-hour recall is 0.18, or 18
percent, on average across the United States. This value varies among demographic groups and is
smaller when looking at specific types of fish. For example, 13 percent of adults reported consuming
freshwater or estuarine fish in the previous 24 hours. Table 3-1 shows the proportion of
respondents in various demographic groups who reported fish consumption on a 24-hour recall
based on the NHANES data. These values can provide a starting point for estimating P when other
relevant data are not available. NHANES does not provide good estimates for finer categorization
of race and ethnicity than those displayed in Table 3-1 (although NHANES 2001-2012 does provide
data for non-Hispanic Asians). If the target population is American Indian/Alaskan Native, for
example, then a more relevant source should be used.
Table 3-1. Estimated percentages of the U.S. population who consumed fish on a given
day, NHANES 2007-2012
Type of fish consumed
Demographic group
All
fish
Freshwater
and
estuarine
Freshwater
Marine
Children < 18 years
8.6%
5.1%
2.0%
7.9%
Adults (18 and over)
18.0%
13.0%
5.6%
16.4%
Mexican American Adults
15.8%
11.8%
5.4%
14.0%
Other Hispanic Adults
16.2%
11.2%
4.0%
15.5%
Non-Hispanic White Adults
16.9%
11.8%
4.9%
15.6%
Non-Hispanic Black Adults
19.0%
14.6%
6.8%
16.3%
Other Race Adults
33.5%
27.0%
12.6%
31.2%
Adults with household income between $20 and $45k
14.9%
10.0%
4.7%
13.5%
Adults with household income between $45 and $75k
17.4%
12.3%
5.2%
16.2%
Adults with household income over $75k
21.5%
16.5%
7.0%
19.7%
36
-------
Analyzing the reported fish consumption data from multiple contacts (or recalls) requires estimating
variance components: the variability in the amount of fish consumed by the same individual on
different days and the additional variation between people (some people consume fish more often or
usually consume more fish than others). Estimating the variance components requires that multiple
respondents report some fish consumption in at least two recalls. The algorithm may not converge
or the estimates of percentiles may be particularly imprecise if there are few respondents who
reported fish consumption on at least two recalls. Let M be the number of respondents reporting
fish consumption in at least two recalls. Based on simulated data, the probability that the NCI
method fails to converge and provide an estimate of UFCR increases when M drops below 20.
Considering this as a minimum criterion, using the expected probability of reported fish
consumption in any contact (P), a minimum sample size (iVMjn) can be calculated as follows:
20
NMin > pj * Deff
where Deff is the design effect.
Using this minimum sample size, the number of respondents with at least two reports of fish
consumption will most likely be greater than 20. If the overall probability of consuming fish in any
one contact is 10 percent, that is, on average fish is consumed once every 10 days using a 24-hour
recall, and a design effect of 2.0 is assumed, then:
20
NMin > q ^2 * — 4,000.
The following table shows the minimum sample size versus the expected probability of consuming
fish in any recall and the design effect. The minimum sample size increases significantly as the
expected proportion of contacts with reported fish consumption decreases. If the minimum sample
size is unacceptably large, consider increasing the length of the recall period or consider revisions to
the sample design to increase the expected P.
Table 3-2. Minimum sample size versus the expected probability of consuming fish in
any recall (P) and the design effect
P
Design
Effect
0.02
0.05
0.10
0.15
0.20
0.30
0.40
0.50
1.0
50,000
8,000
2,000
889
500
223
125
80
1.5
75,000
12,000
3,000
1,334
750
334
188
120
2.0
100,000
16,000
4,000
1,778
1,000
445
250
160
The equations on the previous page specify a minimum sample size for reliably calculating the
survey estimates of the usual fish consumption rate. That minimum sample size may not provide
estimates with the desired precision. The following formula calculates the target sample size (N)
required to obtain the desired precision of the UFCR estimate:
37
-------
^cv ~ ~(jy2 * ®eff
where Deff is the design effect, CV is the desired coefficient of variation of the estimate, and A is a
constant that depends on what parameter is to be estimated and on the expected probability of
reported fish consumption.
The following table shows values for A when estimating the mean, the median (50th percentile), or
the 90th percentile of usual fish consumption.
Table 3-3. Values for A when estimating the mean, the median (50th percentile), or the
90th percentile of usual fish consumption
Expected Probability of Fish
Consumption (P)
Parameter being estimated
90th Percentile
Median
Mean
0.05
6.46
4.35
5.20
0.10
5.77
4.26
4.34
0.20
4.96
3.77
3.31
0.30
4.29
3.44
2.83
0.45
3.76
3.13
2.36
0.60
3.40
2.76
2.04
0.75
3.21
2.56
1.91
0.90
2.98
2.44
1.77
0.95
3.00
2.43
1.76
For estimating the 90th percentile of UFCR when the expected probability of fish consumption is
10 percent, the constant A is 5.77, from the second line in Table 3-3. If the desired precision
specifies a confidence interval of plus or minus 10 percent (corresponding to a CV of 5% or 0.05)
and using a design effect of 2, the desired sample size would be:
5.77
N = _ * 2 = 4,616 respondents
. 05/
The final target analytical sample size (N) is the maximum of Ncv and iVMjn.
N = Max(NMin,Ncv)
The constants in Table 3-3 were determined by simulating fish consumption data with variance
components similar to the NHANES data with two 24-hour dietary recalls, analyzing the data using
the NCI method (see Chapter 7) and calculating the precision of the estimates. The simulations used
various assumptions representing a range of variances that might be appropriate for 24-hour recall
data, whether from NHANES or other sources. The values in the table are the 75th percentile
estimates across the simulated conditions and thus may be conservative (i.e., high on average).
However, the characteristics of other data collection approaches using other populations are
unknown. Thus, these values are provided only as general guidance to sample size determination.
Also, these target sample sizes must be increased to account for nonresponse (individuals who
cannot be located, are not eligible for the survey, or who refuse to complete the survey).
38
-------
If a 7-day fish recall is used for data collection, the probability of reporting fish consumption in any
one contact (i.e., 7-day recall) will be higher than when using a 24-hour recall. The equations can be
used to approximate the target sample size if the probability of reporting fish consumption in any 7-
day period can be estimated. The values of A from Table 3-3 are likely to provide conservative
estimates of sample size when using a 7-day recall period.
3.5.3 Sample Sizes for Single Contact Surveys
If the second data collection approach is used (i.e., a fish-specific dietary assessment instrument, or
fish-specific FFQ, asking individuals to report long-term fish consumption amounts, or values from
which long-term fish consumption can be calculated), another formula can be used to determine the
optimal sample size.
If we have data on usual fish consumption from N individuals, we can calculate the proportion of
the population with usual fish consumption above some specified value, such as 10 grams per day,
and calculate a confidence interval for that proportion. If we specify the desired precision for the
proportion estimate and the expected proportion (p), we can then calculate the sample size using the
following formula:
3.84 p(l — p)Deff ^ Deff
N ~ W2 <~ht~
where H is the desired half width of the 95 percent confidence interval for the proportion. The ratio
Deff/H2 provides an easily calculated, slightly conservative sample size estimate for when the
expected proportion is unknown or close to 0.5. Table 3-4 shows the sample size as a function of
the desired confidence interval width for proportions close to 50 percent and 90 percent. Notice
that the confidence interval gets smaller (the precision increases) as N increases.
Although the confidence interval in the table is not the same as a percentile, Table 3-4 can be used
to calculate approximate sample sizes for percentiles. For example, when estimating the 90th
percentile of reported long-term fish consumption, a sample size of 277 will provide an estimate of
the 90th percentile with a confidence interval length that is roughly the difference between the 85th
and 95th percentiles, with a design effect of two.
39
-------
Table 3-4. Sample size as a function of the desired confidence interval width for
proportions close to 50 percent and 90 percent
p
H
Confidence Interval —
1
Design Effect
2
3
0.5
0.20
.30 to .70
24
48
72
0.5
0.10
.40 to .60
96
192
288
0.5
0.07
.43 to .57
196
392
588
0.5
0.05
.45 to .55
384
768
1152
0.5
0.04
.46 to .54
600
1200
1800
0.5
0.03
.47 to .53
1067
2134
3200
0.5
0.02
.48 to .52
2400
4800
7200
0.9
0.10
.80 to 1.00
35
70
104
0.9
0.07
.83 to .97
71
142
212
0.9
0.05
.85 to .95
139
277
415
0.9
0.04
.86 to .94
216
432
648
0.9
0.03
.87 to .93
384
768
1152
0.9
0.02
.88 to .92
864
1728
2592
3.6 Effect of Response Rates
Response rates affect the precision of the estimates derived from survey data. The potential for high
response rates varies by mode of administration, with the most costly modes generally resulting in
higher expected response rates. This expectation should be carefully considered when determining
an appropriate sample size, particularly if the survey's objectives require the inclusion of a certain
number of responses from subgroups (e.g., fish consumers).
If the response rate is expressed as the fraction of sampled respondents that provide complete data,
the target sample size is:
JV
Sample size = —
Response rate
All other things being equal, a survey with an expected response rate of 30 percent will require twice
the potential sample size of a survey with an expected response rate of 60 percent. Lower response
rates can result in more bias in the study findings as the likelihood of obtaining information on the
target group of interest may be greatly diminished (e.g., high consumers). With very low response
rates, additional biases may result from the few who are willing to participate being quite different
from the population of interest in general. Thus, it is important to develop a realistic response rate
assumption. This may be based on other similar studies, although each survey has unique qualities
that may affect response rates. These include data collection modes used, length of the
questionnaire, type of respondent, type of questions asked, saliency of the subject matter to the
respondents, experience of the interviewers, etc. (see Section 5.7.)
40
-------
Within the current overall environment of declining response rates to surveys in general, the use of a
well-designed monetary (or non-monetary) incentive program can be beneficial and help boost
response rates if administered properly. The timing (e.g., incremental incentives or one-time),
amount, and type of incentive should be considered with an understanding of the target population,
the perceived burden of participation in the survey, and the study design. For further information on
incentives and their effect on response rates, see Mercer et al., 2015. Additionally, when working
with tribal communities, it would be very beneficial for those conducting the surveys to be familiar
with the cultural protocols related to approaching an individual to share information. These
protocols will vary by tribe, but if respected will result in improved response rates. Other methods
to possibly enhance response rates include the use of media to publicize surveys and
communications from leaders of relevant communities. Note that this publicity may also increase
the potential for bias due to frequent fish consumers being more interested in participating in a fish
consumption survey.
3.7 Accuracy
The required accuracy of the consumption rates is an important topic to be considered when
establishing the survey objectives. All approaches discussed in this document can provide estimates
that meet or exceed the accuracy and quality objectives if resources are sufficient, valid survey
designs including adequate numbers of respondents are used, and the design takes into account the
characteristics of the subject matter and the target population.
There are several components to accuracy, including reliability (i.e., the variability of repeatability of
the response); validity (the ability of the respondent to provide the correct answer); measurement
errors (associated with the interviewer, the respondent, the questionnaire, and the data collection
mode); bias (the overestimation or underestimation due to sampling or non-sampling errors); and
random errors.
Other factors influencing the accuracy of the survey responses include whether the respondent
views the survey as non-threatening or sensitive; whether respondents remain anonymous; the
length of the recall period (recall bias); whether some subgroups of the targeted population were
excluded from being part of the survey (coverage bias); the tendency for respondents to provide
responses that conform to ideal norms or what they believe the interviewer wants to hear, or that
might enhance their self-image (prestige bias); the clarity of questions (question misinterpretation);
the familiarity of the respondent with the subject matter; and the amount of specificity in the
questions (e.g., exact numbers vs. approximations or ranges) (Wentland & Smith, 1993).
For fish consumption surveys focused on a limited subset of species and capture locations, the
potential for misidentification of fish consumed is affected by recall bias, prestige bias, and the
familiarity of the respondent with the subject matter. Some ethnic groups may have their own names
for different species, may group some together under one common name, or there may be local
nicknames for species or groups of species. This could be addressed by having the community be
part of the survey design, and through pre-testing. The potential for misidentification is lowest for
creel surveys and on-site in-person interviews (e.g., on-the-bank surveys) because the interviewer can
both directly observe fish catch and/or allow respondents to visually select the species consumed
from displays of fish species. Survey participation affects the accuracy of consumption estimates by
41
-------
affecting the number and characteristics of respondents who are included in the survey. Also, with
creel surveys, typically only the catch between the start of angling and the interviewer/respondent
interaction is recorded, though as mentioned previously, a 24-hour fish consumption recall could be
added to the traditional creel survey.
In general, surveys that allow all of the targeted population a chance to be included in the sample
(e.g., do not exclude consumers living further away from a water body), include a larger number of
respondents, and have a high response rate, provide a more accurate representative estimate of
behaviors among the target population. An understanding of, and sensitivity to, the characteristics of
the target population of concern can help to minimize nonresponse bias due to culture, religion,
language, and attitudes toward government and authority. See Yaske et al., 1996, and Tarrant &
Manfredo, 1993, for additional information.
Stratification of the sample is generally used for one (or more) of the following purposes: (1) to
improve the precision of overall estimates; (2) to ensure adequate subgroup sample sizes; and (3) to
improve the precision of subgroup estimates. Stratification involves partitioning the population into
mutually exclusive, exhaustive groups (strata). Samples are generally selected independently within
each stratum. See Lohr (2010) for further discussion of how strata should be formed and how they
may be used to effectively achieve those objectives.
Stratification with proportional allocation (i.e., allocating the sample to the strata, proportional to the
population in each stratum) may be used to achieve the first and second objectives. Achieving the
third objective generally involves oversampling the rarer subgroups.
For fish consumption surveys, it may be desirable to stratify, for example, on the following:
1. Region/geographic area (e.g., proximity to a particular body of water), to support the
production of estimates by area
2. Age and gender, to support consumption estimates for women of child-bearing age while
limiting the sample sizes for other age/gender subgroups
3. Fishers vs. non-fishers, to ensure adequate sample sizes in each group for comparisons
between these two groups
4. High consumers vs. others
There are several types of response bias that are specific to fish consumption surveys. First, if fish
consumption advisories have been issued, affected groups may be inclined to provide socially or
culturally desirable responses to questions about whether or not they consumed fish (or certain types
of fish). That is, respondents may be hesitant or may refuse to report that they consumed fish
covered by an advisory. If there are restrictions on catching certain types of fish, respondents may
be reluctant to admit they did so, especially if the interviewer has a connection to some state or local
3.8
Sample Stratification and Oversampling
3.9
Response Bias/Avidity
42
-------
authority. Survey designers must take care to introduce questions about fishing practices and/or
consumption with an objective statement reassuring the respondent that there are no right or wrong
answers to the questions. An assurance of confidentiality must also be provided. See Chapter 6 for a
discussion of methods to avoid response bias for measuring suppression.
Second, avidity bias is systematic error in estimates that results from avid fishers (who presumably
are high consumers) having higher propensities to be selected and/or to respond to the survey (see
Andrews et al., 2010, for discussion.). In order to estimate and correct for avidity bias, it is necessary
to obtain angling frequency information as part of the data collection effort. If those who catch (and
consume) more fish have different behaviors/characteristics than those who fish (and consume)
less, it may be necessary to apply avidity weights during data analysis. Depending on the survey
objectives, this is of particular concern in creel and on-the-bank surveys.
3.10 Choosing an Approach
Selecting a survey mode and sampling frame from among the options described in this chapter
requires consideration of multiple factors, including the realities of possible budget/resource
limitations. In this section, we present survey design options within the context of three different
levels of funding/resources. These range from relatively low-cost approaches to more resource-
intensive efforts. The more resource-intensive efforts are generally accepted as standard
methodological approaches where the budget is not as tightly constrained. Each of the cost
categories (presented in Tables 3-5, 3-6, and 3-7) offers methodologically sound choices for
accomplishing a fish consumption survey. To help with decisions about investing adequate levels of
resources needed to obtain high-quality, defensible data, survey developers may use decision-making
tools such as Value of Information (Minelli & Baio, 2015) or influence diagrams (Carriger & Barron,
2011).
Resource-sensitive options are presented using the following levels.
Level 1 (least expensive): Low- to moderate-cost partnerships that leverage existing surveys
Level 2: Low- to moderate-cost options for developing a new survey
Level 3 (most expensive): Moderate- to high-cost options for developing a new survey
Level 1: Low- to Moderate-Cost Partnerships that Leverage Existing Surveys
If resources are limited, a good approach would be for survey developers to first determine whether
there may be a survey already in place within their geographic area of interest focusing on the
population of interest. Any existing survey should be carefully evaluated to determine the feasibility,
cost, and schedule for adding desired questions. This has the advantage of utilizing an existing
survey methodology that has been established, without having to invest in developing an entire
sample design and survey infrastructure. However, the frame and survey methodology of the
existing survey should be carefully examined to determine whether or not it will result in a sample
that is representative of the population of interest. Also this analysis should consider any potential
biases or inaccuracies specific to the population of interest. For example, is seasonality addressed?
Will the existing survey yield data that are consistent with survey goals in terms of any time
43
-------
constraints? Another consideration is whether or not the size of the existing effort is sufficient to
support derivation of the desired statistics.
While there may be costs associated with adding questions to an existing survey, these costs should
be less than what would be required to develop and implement a new survey. The process for
adding questions onto existing surveys and cost for doing so will vary, depending on which survey is
chosen as the best vehicle for obtaining the desired data. All of the level 1 survey options involve
building on existing data collections.
There are several different types of survey designs that could be leveraged. One is a random digit
dialing telephone survey (e.g., BRFSS). BRFSS is conducted by the Centers for Disease Control and
Prevention (www.cdc.gov/brfss) in all 50 states, the District of Columbia, and three U.S. territories.
It is the Nation's leading telephone survey, collecting state data about U.S. residents regarding their
health-related risk behaviors, chronic health conditions, and use of preventive services. BRFSS
completes more than 400,000 adult interviews each year, making it the largest continuously
conducted health survey system in the world. With technical and methodological assistance from
CDC, state health departments use in-house interviewers or contract with telephone call centers or
universities to administer the BRFSS surveys continuously throughout the year. States use a
standardized core questionnaire, optional modules, and state-added questions. The survey uses RDD
techniques for both landlines and cell phones.
While the content of the core BRFSS questionnaire is determined by state coordinators and the
CDC, state coordinators may choose to add new questions based on submitted proposals
(http://www.cdc.gov/brfss/about/brfss faq.htm). There is a review process in place for BRFSS to
decide whether additional questions can be accommodated and what fees will be charged. Before
beginning this review process, the first step is for survey developers to determine whether a
telephone survey is appropriate for their population of interest, and whether BRFSS would provide
adequate coverage within the population(s) most of interest for the proposed surveys. There may
also be limitations on the number of questions that can be added to the existing survey. Adding
questions to BRFSS would yield fish consumption rates for the general population of the state.
However, using BRFSS may not provide valid estimates for subpopulations of interest such as local
fishers, pregnant women, or subsistence fishers.
Other types of surveys that could be leveraged include in-person intercept creel surveys, or web or
mail surveys conducted within certain areas (i.e., surveys conducted by local agencies). For example,
fisheries managers in a number of states have current or planned creel surveys that are done on a
continuous basis, conducted by state, tribal, or local agency staff (examples include Florida FWC,
2016; Michigan DNR, 2014; Oregon DFW, 2014; Texas PWD, 2014; Washington DFW, 2014;
Minnesota DNR, 2011;). There are a variety of mail and online surveys that are also conducted by
state and local authorities. Examples include the Tennessee Wildlife Resources Agency's (TWRA)
BITE (Bass Information from Tournament Entries) program, which is a coordinated effort to
obtain tournament data to the TWRA Fisheries Management Division via an online reporting form
or a mail-in tournament report card. This information supplies information such as the number of
participants, total catch, size, and weight structure of the tournament catch (Tennessee Wildlife
Resources Agency, 2014). Also, the Florida Fish and Wildlife Conservation Commission created an
online web survey to indicate where they harvest scallops, how many they collect, and how long it
takes to harvest the shellfish (Florida Fish and Wildlife Conservation Commission, 2014). Also, New
Jersey has undertaken the Recreational Saltwater Angler Survey to collect information on
44
-------
recreationally important marine finfish species (New Jersey Department of Environmental
Protection, 2014). These are just a few examples of local surveys.
If it is decided that a survey's objectives can be fulfilled by leveraging an existing survey effort, the
researchers should conduct a thorough assessment of the current possibilities that exist within the
locales of interest, including local colleges or universities as they may have on-going or up-coming
research concerning fish consumption. Then, negotiations would need to be held with the agency
that conducts the survey to assess the cost and feasibility of adding the desired fish consumption
questions to an existing survey.
Table 3-5 shows how each of these modes and sample designs could address survey objectives, and
the pros/cons of each approach within the overall structure of a partnership with an existing survey.
Level 2: Low- to Moderate-Cost Options for Developing a New Survey
At this level, there are four modes of administration to choose from (mail only, mail-to-web hybrid,
web only, or an in-person survey), and three different sample designs (address-based sampling, list
sample, and an intercept design for the in-person mode). Under Level 2, these would be stand-alone
new survey efforts, and all of the designs rely on relatively low-cost methodologies that can provide
statistically sound results.
Table 3-6 shows how each of these modes and sample designs could address survey objectives, and
the pros/cons of each approach. The mail-to-web option mode involves sending a mail invitation to
potential study participants, inviting them to visit a web site to complete the survey. If necessary,
some screening can occur at the time of the mail contact to allow the survey to be targeted toward
respondents with characteristics of interest (e.g., fishers).
Level 3: Moderate- to High-Cost Options for Developing a New Survey
This highest level presents some other generally accepted survey approaches that require more
extensive resources to develop and implement. These are presented in this document to provide an
understanding of what these approaches can offer compared to the low and moderate cost options,
and when they may be most appropriate to consider. Three administration modes are within this
level: telephone surveys, in-person surveys, and mail-to-telephone hybrid. Within each mode of
administration, there are different sampling approaches that could be employed, depending on the
characteristics of the population under study and other survey objectives. Table 3-7 shows how each
of these modes and sample designs could address survey objectives, and the pros/cons of each
approach.
In sum, Tables 3-5 through 3-7 describe major factors affecting the decision about an appropriate
sample design and mode of data collection within these three budgetary frameworks. Each table lists
independent survey objectives that the designer may wish to address, along with potential
characteristics of the population to be surveyed. These tables are designed to aid the reader in
choosing an appropriate sample design and data collection mode based on multiple independent
factors, considering the known level of resources available for the planned survey. Listed on each
table are the key factors to consider in choosing an approach, as follows.
45
-------
Required response rates: Achieving the highest possible response rates is optimal. Higher
response rates generally increase the precision of the estimate and maximize the number of
observations available for analysis.
Need for use of visuals to aid respondent: Some modes of administration may not be suitable if
visual aids are needed to obtain accurate information from the respondent. For example, for fish
consumption surveys, visual aids showing portion sizes, species of fish, and/or maps of fishing
locations may be helpful in improving the accuracy of responses.
Need for data over multiple seasons or points in time: Fish consumption rates can be heavily
influenced by the time of year, and the design of consumption surveys should take into account
when the survey will be administered. To estimate UFCR, interviews should take place at multiple
points in time.
Size of geographic area to be covered: Some data collection modes and sampling frames may be
better suited to smaller or larger geographic areas. For example, an in-person interview that needs to
cover a large geographic area will be resource intensive due to the need for interviewers to move
throughout the area.
Expected prevalence of the characteristic and/or behavior among the target population:
What percentage of the population within the defined geographic area is likely to possess the
demographic characteristic (s) of interest or behavior (e.g., pregnant women, children, elderly
persons, high fish consumers, etc.)? If the characteristic or behavior is expected to be relatively rare
among the population, then a larger sample size will be required so that completed interviews are
obtained from a sufficient number of individuals with the characteristic and/or behavior of interest.
Literacy rates among population of interest: Certain modes of administration (e.g., any self-
administered survey) may not be feasible for populations that are known to have low literacy rates.
Fixed addresses within population of interest: Certain modes of administration (e.g., a mail
survey) are not feasible for highly mobile populations (e.g., fishers who may move with the seasons)
because they will not receive the survey in the mail in a timely manner.
Web penetration rates within population of interest: While proven to be an efficient mode of
data collection among certain groups, web surveys should not be planned for a population that is
known to have limited Internet accessibility.
Telephone coverage rates for population of interest: Telephone surveys are not generally used
for a population without a known telephone number (e.g., list survey) or availability of a telephone
(e.g., RDD survey). Ideally, both landline and cell phone numbers will be available.
Length of the interview: Some modes of administration are not generally used for longer survey
instruments. In general, the shorter the interview, the more options the researcher has for choosing
a mode.
46
-------
Table 3-5.
Level 1 framework for choosing a sample design and data collection mode based on independent factors
YES = APPROPRIATE
LEVEL 1: POTENTIAL LOW TO MODERATE COST PARTNERSHIPS TO
MAYBE = MAY BE APPROPRIATE
LEVERAGE EXISTING SURVEYS
NO = NOT APPROPRIATE
RDD Telephone Survey
In-Person Intercept Survey
Web or Mail Survey
SURVEY OBJECTIVES
What response rates are needed from the survey to meet precision requirements?
High (>80%)
MAYBE
Moderate (60-79%)
MAYBE
MAYBE
Low (<60%)
YES
YES
YES
Are visuals needed for validity (e.g., river maps, pictures offish species, portion
sizes, etc.)?
Yes
MAYBE
YES
YES
No
YES
YES
YES
Are data needed over multiple seasons, or are multiple observation points needed?
Yes
YES
MAYBE
YES
No
YES
YES
YES
What is the size of the geographic area to be covered?
Large
YES
YES
Moderate
YES
MAYBE
YES
Small
YES
YES
YES
POPULATION CHARACTERISTICS
Prevalence of characteristic or behavior of interest among target population (e.g.,
pregnant women and/or high fish consumers)?
High percent of population
YES
YES
YES
Moderate percent of population
MAYBE
MAYBE
YES
Low percent of population
MAYBE
YES
Unknown
MAYBE
MAYBE
YES
What are the literacy rates among the target population?
High
YES
YES
YES
Moderate
YES
YES
MAYBE
Low
YES
YES
Unknown
YES
YES
NO
How important is it to have high coverage of persons with no fixed address?
Very important
YES
Moderately important
MAYBE
YES
MAYBE
Not important
YES
YES
YES
47
-------
Table 3-5. Level 1 framework for choosing a sample design and data collection mode based on independent factors (continued)
RDD Telephone Survey
In-Person Intercept Survey
Web or Mail Survey
What are the web penetration rates for the target population? *
High
YES
YES
YES
Moderate
YES
YES
YES
Low
YES
YES
WEB MAYBE MAIL YES
Unknown
YES
YES
WEB IMO MAIL YES
What are the telephone coverage rates for the target population? *
High
YES
YES
YES
Moderate
MAYBE
YES
YES
Low
YES
YES
Unknown
YES
YES
PROS/CONS OF EACH APPROACH
Coverage
Without dual frame, would
Depends on coverage of
Depends on coverage of
exclude HHs w/o landline
existing survey
existing survey
telephone numbers. In mid-2014,
43% of HHs nationally were cell
only (Blumberg & Luke, 2014)
Response rate
Low, less likely to get "buy-in" for
May be high, although the
Moderate
survey over the phone
presence of interviewers in
certain locations may act as
deterrent. If interviewers are
perceived as law
enforcement, there may be
cooperation issues
Need for visuals
Can be provided to respondents
High degree of flexibility in
Can be provided with mail
only if mailing precedes RDD call
use of visuals
materials, or in advance of
or refer respondents to common
web survey invitation
household objects
Size of geographic area
No impact
Depends on range of existing
No impact
survey but large geographic
area may be difficult to cover
and/or prohibitively expensive
Screening costs
High, may need to screen HHs to
Low - screening may be
Moderate, would have to
identify fishers or people who
unnecessary, or minimal
screen to identify fishers or
consume fish
people who consume fish,
but less costly than phone
or in-person screening
48
-------
Table 3-5. Level 1 framework for choosing a sample design and data collection mode based on independent factors (continued)
RDD Telephone Survey
In-Person Intercept Survey
Web or Mail Survey
Literacy rates among the target population
No effect on ability to
participate in survey
No effect on ability to
participate in survey
Those with low literacy may
be unable to participate,
leading to potential biases
Web penetration for target population
No effect on ability to
participate in survey
No effect on ability to
participate in survey
No effect on mail survey,
major impact on web
survey
Telephone coverage for target population
Persons with no landline
numbers would be unable
to be sampled or contacted
No effect on ability to
participate in survey
No effect on ability to
participate in survey
Questionnaire length
(see Section 4.9 for details)
Screener up to 5 min., 20-
30 min. for main survey,
longer lengths can be used
but potential for breakoffs
will increase
10-20 minutes, but
situational factors need to
be considered
Web 15-20 mins., but
higher potential for
breakoffs with longer
length;
Mail 15-30 mins, although
up to 45 min. may be
feasible
Relative costs
Low - Moderate
Moderate - High
Low
* This refers to how many have an email address (proxy for web access) or have a telephone number available.
49
-------
Table 3-6. Level 2 framework for choosing a sample design and data collection mode based on independent factors
YES = APPROPRIATE
MAYBE = MAY BE APPROPRIATE
NO = NOT APPROPRIATE
LEVEL 2: LOW TO MODERATE COST OPTIONS FOR DEVELOPING A NEW SURVEY
MAIL ONLY
MAIL TO WEB
WEB ONLY
IN-PERSON
Address
Based- List
Sample Sample
Address-
Based List
sample Sample
Address-
Based List
sample Sample
Intercept
Survey
SURVEY OBJECTIVES
What response rates are needed from the survey to meet precision requirements?
High (>80%)
Moderate (60-79%)
Low (<60%)
NO MAYBE
MAYBE MAYBE
YES YES
NO MAYBE
MAYBE MAYBE
YES YES
NO MAYBE
MAYBE MAYBE
YES YES
MAYBE
MAYBE
YES
Are visuals needed for validity (e.g., river maps, pictures offish species, portion sizes,
etc.)?
Yes
No
MAYBE MAYBE
YES YES
MAYBE MAYBE
YES YES
YES YES
YES YES
YES
YES
Are data needed over multiple seasons or are multiple observation points needed?
Yes
No
YES YES
YES YES
YES YES
YES YES
YES YES
YES YES
MAYBE
YES
What is the size of the geographic area to be covered?
Large
Moderate
Small
YES YES
YES YES
YES YES
YES YES
YES YES
YES YES
YES YES
YES YES
YES YES
NO
MAYBE
YES
POPULATION CHARACTERISTICS
Prevalence of characteristic or behavior of interest among target population (e.g.,
pregnant women and/or high fish consumers)?
High percent of population
Moderate percent of population
Low percent of population
Unknown
YES YES
MAYBE MAYBE
NO MAYBE
MAYBE MAYBE
YES YES
MAYBE MAYBE
NO MAYBE
MAYBE MAYBE
YES YES
MAYBE MAYBE
NO MAYBE
MAYBE MAYBE
YES
MAYBE
NO
MAYBE
What are the literacy rates among the target population?
High
Moderate
Low
Unknown
YES YES
MAYBE MAYBE
NO NO
YES YES
MAYBE MAYBE
NO NO
YES YES
MAYBE MAYBE
NO NO
YES
YES
YES
YES
50
-------
Table 3-6.
Level 2 framework for choosing a sample design and data collection mode based on independent factors (continued)
MAIL ONLY
MAIL TO WEB HYBRID
WEB ONLY
IN-PERSON
Address-Based
Address-
Address-
Intercept
sample
List Sample
Based sample
List Sample
Based sample
List Sample
Survey
How important is it to have high coverage of
persons with no fixed address?
Very important
NO
YES
Moderately important
MAYBE
MAYBE
MAYBE
MAYBE
MAYBE
MAYBE
YES
Not important
YES
YES
YES
YES
YES
YES
YES
What are the web penetration rates for the
target population? *
High
YES
YES
YES
YES
YES
YES
YES
Moderate
YES
YES
MAYBE
MAYBE
MAYBE
MAYBE
YES
Low
YES
YES
YES
Unknown
YES
YES
NO
NO
NO
NO
YES
What are the telephone coverage rates for the
target population? *
High
YES
YES
YES
YES
YES
YES
YES
Moderate
YES
YES
YES
YES
YES
YES
YES
Low
YES
YES
YES
YES
YES
YES
YES
Unknown
YES
YES
YES
YES
YES
YES
YES
PROS/CONS OF EACH APPROACH
Coverage
Highest
Will not
Highest
Will not
Will not
Will not
Difficult and
coverage of
represent a
coverage of
represent a
represent a
represent a
costly to
population
general
population
general
general
general
design
(excluding area
population.
(excluding area
population.
population
population.
statistically
probability
The extent to
probability
The extent to
unless web
The extent to
representative
samples).
which a
samples).
which a
penetration is
which a
survey;
Potential for
targeted
Potential for
targeted
100% among
targeted
excludes
higher rates of
population
higher rates of
population
population.
population will
individuals
undercoverage
will be
undercoverage
will be
be covered
who fish from
in some rural
covered will
in some rural
covered will
will depend
private ponds.
areas.
depend upon
areas. Depends
depend upon
upon list
list quality.
on web
list quality.
quality. Also
penetration.
Depends on
depends on
web
web
penetration.
penetration.
51
-------
Table 3-6.
Level 2 framework for choosing a sample design and data collection mode based on independent factors (continued)
MAIL ONLY
MAIL TO WEB HYBRID
WEB ONLY
IN-PERSON
Address-
Address-
Address-Based
Intercept
Based sample List Sample
Based sample List Sample
sample List Sample
Survey
Response rate
Moderate
Moderate
Low
May be high,
although if
interviewers
are perceived
as law
enforcement,
there may be
problems with
obtaining
cooperation
Need for visuals
Can be provided with mail
Variable, can be included in mail
Can be shown on web survey
High degree of
materials
survey or shown on web survey
flexibility in
use of visuals
Size of geographic area
No impact
No impact
No impact
Large
geographic
areas may be
difficult to
cover and/or
prohibitively
expensive
Screening costs
Moderate, may Generally
Moderate, may Generally not
Moderate, may Generally not
Low -
have to screen not
have to screen necessary,
have to screen necessary,
screening may
to identify necessary,
to identify but will
to identify but will
be
fishers/people but will
fishers/ people depend on
fishers/ people depend on
unnecessary,
who consume depend on
who consume target
who consume target
or minimal
fish, but less target
fish, but less population in
fish, but less population in
costly than population
costly than the list
costly than the list
phone or in- in the list
phone or in- sample
phone or in- sample
person sample
person
person
screening
screening
screening
Literacy rates among the target population
Those with low literacy may be
Those with low literacy may be
Those with low literacy may be
Little or no
unable to participate, leading
unable to participate, leading to
unable to participate, leading to
impact
to potential biases
potential biases
potential biases
52
-------
Table 3-6.
Level 2 framework for choosing a sample design and data collection mode based on independent factors (continued)
MAIL ONLY
MAIL TO WEB HYBRID
WEB ONLY
IN-PERSON
Address-
Address-
Address-
Intercept
Based sample List Sample
Based sample List Sample
Based sample List Sample
Survey
Web penetration for target population
No effect on ability to
Sampled persons without
Sampled persons without
No effect on
participate in survey
Internet access will be unable to
Internet access will be unable to
ability to
fully participate, leading to
participate, leading to potential
participate in
potential biases
biases
survey
Telephone coverage for target population
No effect on ability to participate in survey
Questionnaire length
15-30 mins., but up to 45 min.
Mail invitation only, no data
15-20 mins., but higher
10-20 minutes,
(refer to Section 4.9 for details)
may be feasible
collected; Web-15-20 mins., but
potential for breakoffs with
but situational
higher potential for breakoffs
longer length
factors need
with longer length
to be
considered
Relative costs
Low
Low
Low
Moderate
(with relatively
small sample
size)
* This refers to how many have an email address (proxy for web access) or have a telephone number available.
53
-------
Table 3-7.
Level 3 framework for choosing a sample design and data collection mode based on independent factors
YES = APPROPRIATE
LEVEL 3: MODERATE TO HIGH COST OPTIONS FOR DEVELOPING A NEW SURVEY
TELEPHONE
IN-PERSON
MAIL TO PHONE HYBRID
MAYBE = MAY BE APPROPRIATE
Area
NO = NOT APPROPRIATE
Probability
Address-
Ad dress-Based
List
RDD
List Sample
Sample
Based Sample
List Sample
Sample
Sample
SURVEY OBJECTIVES
What response rates are needed from
the survey to meet precision
requirements?
High (>80%)
NO
MAYBE
MAYBE
MAYBE
MAYBE
MAYBE
Moderate (60-79%)
MAYBE
MAYBE
MAYBE
MAYBE
MAYBE
MAYBE
MAYBE
Low (<60%)
YES
YES
YES
YES
YES
YES
YES
Are visuals needed for validity (e.g.,
river maps, pictures offish species,
portion sizes, etc.)?
Yes
MAYBE
MAYBE
YES
YES
YES
MAYBE
MAYBE
No
YES
YES
YES
YES
YES
YES
YES
Are data needed over multiple seasons
or are multiple observation points
needed?
Yes
YES
YES
YES
YES
YES
YES
YES
No
YES
YES
YES
YES
YES
YES
YES
What is the size of the geographic area
to be covered?
Large
YES
YES
YES
YES
YES
YES
Moderate
YES
YES
YES
YES
MAYBE
YES
YES
Small
YES
YES
YES
YES
YES
YES
YES
POPULATION CHARACTERISTICS
Prevalence of characteristic or
behavior of interest among target
population (e.g., pregnant women
and/or high fish consumers)?
High percent of population
YES
YES
YES
YES
YES
YES
YES
Moderate percent of population
MAYBE
MAYBE
MAYBE
MAYBE
MAYBE
MAYBE
MAYBE
Low percent of population
NO
MAYBE
NO
MAYBE
Unknown
MAYBE
MAYBE
MAYBE
MAYBE
MAYBE
MAYBE
MAYBE
54
-------
Table 3-7.
Level 3 framework for choosing a sample design and data collection mode based on independent factors (continued)
TELEPHONE
IN-PERSON
MAIL TO PHONE HYBRID
Area
Probability
Address-
Ad dress-Based
List
RDD
List Sample
Sample
Based sample
List Sample
sample
Sample
What are the literacy rates among the
target population?
High
YES
YES
YES
YES
YES
YES
YES
Moderate
YES
YES
YES
YES
YES
MAYBE
MAYBE
Low
YES
YES
YES
YES
YES
Unknown
YES
YES
YES
YES
YES
NO
NO
How important is it to have high
coverage of persons with no fixed
address?
Very important
NO
MAYBE
YES
NO
Moderately important
MAYBE
MAYBE
MAYBE
MAYBE
YES
MAYBE
MAYBE
Not important
YES
YES
YES
YES
YES
YES
YES
What are the web penetration rates for
the target population? *
High
YES
YES
YES
YES
YES
YES
YES
Moderate
YES
YES
YES
YES
YES
YES
YES
Low
YES
YES
YES
YES
YES
YES
YES
Unknown
YES
YES
YES
YES
YES
YES
YES
What are the telephone coverage rates
for the target population? *
High
YES
YES
YES
YES
YES
YES
YES
Moderate
MAYBE
MAYBE
YES
YES
YES
MAYBE
MAYBE
Low
YES
YES
YES
Unknown
MAYBE
MAYBE
YES
YES
YES
MAYBE
MAYBE
PROS/CONS OF EACH APPROACH
Coverage
Excludes HHs w/o
Will not represent
Highest coverage
Highest coverage
Will not represent a
Highest coverage
Will not represent a
landline telephone
a general
of population.
of population.
general population.
of population.
general population.
numbers,
population. The
Potential for
The extent to which a
Potential for
The extent to which
although could
extent to which a
higher rates of
targeted population
higher rates of
a targeted
supplement with
targeted
undercoverage in
will be covered
undercoverage in
population will be
cell frame.
population will be
covered depends
upon list quality.
some rural areas.
depends on list quality.
some rural areas.
covered will depend
upon list quality.
55
-------
Table 3-7. Level 3 framework for choosing a sample design and data collection mode based on independent factors (continued)
TELEPHONE
IN-PERSON
MAIL TO PHONE HYBRID
RDD List Sample
Area
Probability Address-
Sample Based Sample List Sample
Ad dress-Based
Sample List Sample
Response rate
Low, less likely Variable,
to get "buy-in" depending
for survey over upon list quality
the phone. and data
Potential for collection
lower response mode,
for cell phone
respondents.
Relatively high
Relatively low; Variable,
Potential survey depending upon
"buy-in" list quality and
increases data collection
w/previous mail mode,
contact.
Need for visuals
Can only be provided through
advance mailing to respondents or
refer respondents to common
household objects.
High degree of flexibility in use of visuals.
Can be provided only through
advance mailing to respondents.
Size of geographic area
No effect on ability to participate
in survey.
Large geographic areas will Large geographic
require sample clustering. areas may be
difficult to cover
and prohibitively
expensive.
No effect on ability to participate
in survey.
Screening costs
High, may have Generally not
to screen to necessary, but
identify fishers will depend on
or fish target
consumers. population in
the list sample.
Variable, depends on if only Generally not
interested in fishers or fish necessary, but will
consumers. depend on target
population in the
list sample.
Moderate-High, Generally not
if screening is necessary, but
part of mail will depend on
protocol. If target
screening part population in
of phone the list sample,
protocol,
screening costs
will increase.
Literacy rates among the target
population
No effect on ability to participate
in survey.
No effect on ability to participate Little or no impact,
in survey.
Those with low literacy may be
unable to respond to phone
number request, leading to
potential biases.
56
-------
Table 3-7. Level 3 framework for choosing a sample design and data collection mode based on independent factors (continued)
TELEPHONE
IN-PERSON
MAIL TO PHONE HYBRID
RDD List Sample
Area
Probability Address-
Sample Based Sample List Sample
Ad dress-Based
Sample List Sample
Web penetration for target population
No effect on ability to participate in survey.
Telephone coverage for target
population
High if cell Those with no
phones telephone
included. number on the
list frame will
be unable to be
contacted.
No effect on ability to participate in survey.
Sampled Those with no
persons telephone
without number on the
telephones list frame will
would be be unable to be
unable to contacted,
provide a
phone number
for contact.
Questionnaire length
(see Section 4.9 for details)
20-30 min., although longer
lengths can be used but potential
for breakoffs will increase.
30-60 minutes
30-60 minutes
Mail invitation only, no data
collected; phone 20-30 mins. with
higher potential for breakoffs with
longer length.
Relative costs
Moderate
High
High
Moderate
* This refers to how many have an email address (proxy for web access) or have a telephone number available.
57
-------
4
QUESTIONNAIRES FOR COLLECTING FISH
CONSUMPTION DATA
4.1
Background
We have previously presented a variety of data collection options (based on budgetary level) in
Tables 3-5 through 3-7. These tables address various types of sampling frames and modes that can
be used, based on available resources and research goals. In addition to selecting a sampling frame
and mode to satisfy the goals of the planned research, a variety of factors such as characteristics of
the population (e.g., literacy rates) and desired outcomes of the survey (e.g., response rates) also
factor into the decision. These are also addressed in Tables 3-5 through 3-7 to provide guidance for
the researcher. In this chapter, we build upon this information by presenting questionnaire
approaches that can be used with any of the survey approaches previously discussed. This is the
fourth of the five basic steps in the design and development of a survey for estimating fish
consumption rates listed in Chapter 1— Development of an appropriate survey questionnaire. Note
that this chapter does not cover methodologies for determining heritage rates. Those methodologies
are covered in Chapter 6. As additional resources, Appendix A provides a discussion of
considerations for instrument development to help the researcher develop survey questions that are
methodologically sound. Appendix B provides examples of validated survey instruments, as well as
additional information on structuring questions aimed at collecting fish consumption data. See also
Willet, 2012, and Thompson and Subar, 2008, for general information and discussion on dietary
assessment.
The decision of which dietary assessment tool to use depends on many things. Two of the most
important considerations are the ultimate objective of the survey (e.g., estimate the distribution of
usual intake) and the planned analysis method. The NCI has published the Dietary Assessment
Primer, which provides background on dietary data collection methodologies, the benefits and
drawbacks of the various methods, and approaches to use given the study objective
(http://dietassessmentprimer.cancer.gov/). If the survey's goal is to estimate the usual intake
distribution, the NCI's Diet Assessment Primer endorses multiple administrations of a 24-hour
recall or a single 24-hour recall on the full sample and multiple 24-hour recalls on a sub-sample
(http://dietassessmentprimer.cancer.gov/approach/table.html#intake). NCI notes that more than
two administrations of the 24-hour recall are beneficial for foods that are consumed by less than 5
percent to 10 percent of the population of interest. However, further research is needed to guide
specific recommendations on the number of 24-hour recalls needed in relation to frequency of
consumption. They therefore suggest that the best approach is to collect two 24-hour recalls from
the entire sample (http://dietassessmentprimer.cancer.gov/approach/intake.html).
In Section 1.2.3, five analysis methods were introduced to estimate UFCR from dietary data
collected through a survey. These methods are the NCI Method (Tooze et al., 2010; Tooze et al.,
2006), MSM (Harttig et al., 2011), the Iowa State University Method (Nusser et al., 1996), SPADE
(Dekkers et al., 2014), and the use of a fish-specific FFQ. Table 4-1 provides the data needs and the
advantages and limitations of each of these analytical approaches. Other analysis methods requiring
custom programming might be used, including Bayesian methods. In some cases, where the data are
not adequate for the NCI method, additional data may be available that can be used to estimate
58
-------
UFCR, in particular see Carroll et al., 2012. In these cases, consultation with a statistician is
recommended.
The NCI Method for data analysis utilizes statistical modeling to estimate usual intake of nutrients
and foods, and it is currently the best method for estimating usual intake of episodically consumed
foods such as fish when a sufficient number of respondents and resources are available. The method
has a general requirement that dietary data are collected at two or more different time points, during
which a respondent provides a 24-hour recall. The method provides an estimate of UFCR
representing the long-term average grams of fish consumed per unit of time (e.g., day or week).
There is a requirement that at least some respondents have reports of consumption of the fish of
interest for more than one time point. It is generally accepted that a survey must yield at least 50
respondents who report fish consumption at least two different times (Kipnis, et al., 2009). In order
to achieve this requirement, a survey must interview a large number of people or make many
contacts with each respondent, especially if the prevalence of the behavior of interest (e.g.,
consumption of a certain type of fish) is low. While lengthening the recall period at each interview
time point per contact (e.g., 24 hours to 7 days) increases the number of fish consumption reports,
the accuracy of a participant's memory of what he or she ate diminishes as the recall period
lengthens (Gersovitz et al., 1978).
If there is a concern that the survey will not yield enough reports of fish consumption from
individuals on multiple days for certain subpopulations or specific fish of interest (e.g., freshwater
fish) to successfully implement the NCI Method, it would be beneficial to also collect long-term
frequency of consumption data from participants. It may be possible to approximate the distribution
of fish consumption in small subpopulations or for rarely consumed fish species based on an
analysis of the relationship among mean fish consumption rates from the FFQ and the distributions
of fish consumption in those populations where it can be estimated. As a simple example, if a survey
yields enough reports of multiple days of total fish consumption to utilize the NCI Method, but not
enough reports of multiple days of freshwater fish consumption, the UFCR distribution for
freshwater fish might be estimated by scaling the UFCR distribution for all fish based on the ratio of
the mean fish consumption rate from the FFQ data. More research is needed to assess the validity of
this approach.
59
-------
Table 4-1.
Data needs, advantages, and limitations of various analytical methodologies for estimating UFCR
NCI Method
Multiple Source Method
(MSM)
Iowa State University
Method
Statistical Program for
Age-Adjusted Dietary
Assessment (SPADE)
FFQ Alone
Data needs
24-hour fish
24-hour fish
24-hour fish
24-hour fish
Frequency of
consumption recall
consumption recall
consumption recall
consumption recall
consumption over a long
period of time; some
measure of portion size.
Minimum
2
2
2
2
1
number of
contacts
with
respondents
Advantages
Current best method for
Can utilize covariates to
Can include survey
Implemented in the
Least expensive due to
estimating usual intake
improve estimates (such
weights.
statistical package R and
dietary data collection
of episodically consumed
as frequency of
freely available.
requirements and
foods like fish.
consumption).
Can include age as a
easiest to implement.
Can utilize covariates to
Web-based interface is
covariate.
Can include survey
improve estimates (such
easy to use.
Can include survey
weights.
as frequency of
weights.
Useful for smaller
consumption).
populations and rarely
Can include survey
consumed fish species.
weights.
Limitations
Requires SAS software
Does not provide
Does not account for
Does not account for
Estimates of UFCR may
(expensive).
standard errors.
correlation between
correlation between
be biased. The amount
Difficult to implement if
Cannot include survey
consumption amounts
consumption amounts
of bias is hard to
the target population is
weights.
and consumption
and consumption
quantify and will be
very small or the survey
Difficult to implement if
frequency.
frequency.
different for estimates of
is interested in specific
the target population is
Difficult to implement if
Difficult to implement if
upper and lower
species offish or
very small or the survey
the target population is
the target population is
percentiles.
narrowly defined groups
is interested in specific
very small or the survey
very small or the survey
of fish.
species offish or
narrowly defined groups
of fish.
is interested in specific
species of fish or
narrowly defined groups
of fish.
is interested in specific
species offish or
narrowly defined groups
of fish.
60
-------
4.2
Questionnaires for Fish Consumption Surveys
Defensible and quantifiable fish consumption rates are needed for development of water quality
standards for tribes and states, as well as assessment of seafood contaminant risks to the population.
There are various options that can be considered when choosing a questionnaire for a fish
consumption survey. In the overview provided in Appendix B, the methods of a 24-hour recall, fish
focused recall, food record/diary, food frequency questionnaire, and fish diet screener are described.
Depending on the needs of the survey, any of these may be appropriate for use. Appendix B
enumerates the advantages and limitations of each approach.
For purposes of this discussion, we focus on a previously used instrument that utilizes the 24-hour
recall approach and also collects fish consumption frequency data. In 2003, EPA developed a
Microsoft Access-based software package, henceforth called the "Fish Consumption Survey Tool"
(or FCST), which could be used to conduct well-designed fish consumption surveys (Kissinger et al.,
2010). While the software, derived from a 2000 Suquamish Tribe seafood consumption survey and
other seafood consumption surveys used in the Pacific Northwest, was originally developed for
tribal use, it is also usable by other entities and other populations. If consumption of foods in
addition to fish (e.g., marine mammals, wetland plants) is of interest to the study, additional
questions can be added to the FCST, modeled after the fish-specific questions. If there are many
different foods of interest to the study, a total diet 24-hour recall would be the preferred data
collection methodology (over a fish-focused recall) in order to capture all foods of interest.
The FCST is sophisticated and well developed and provides reasonably good estimates of fish
consumption. Multiple administrations allow for the assessment of usual intake and population
distributions (based on the NCI analysis method described in Chapter 7). It also requires the
respondent to estimate the portion of fish consumed, as would be the case in all retrospective
surveys.
The FCST is a configurable, computerized interview that collects a 24-hour fish consumption recall
and a FFQ, annual and seasonal consumption (including source and parts consumed), purchased
versus caught fish, consumption by children, and consumption at gatherings and special events. Its
greatest strength is the capture of quantitative fish consumption data. It allows researchers to
populate the software with fish species, parts, portion sizes (including dishes) of interest and to
characterize seasonal variability in fish consumption. Fish consumption may be characterized on the
basis of groups of fish having similar feeding behavior/trophic level associated with contaminant
uptake. The fish database includes 77 species of fish. The FCST automates logic flow in question
branching, simplifying survey administration. Range checking of answers enhances accuracy and
reliability. The need for data entry of hard copy surveys is eliminated. Individual interviewers can use
the FCST on multiple devices and can download their data to a master survey database. Query and
data reporting capabilities allow calculation of basic seafood consumption statistics on either a
complete survey data set or data subsets. The data could be used with the NCI Method for
estimating UFCR or if survey circumstances make the NCI Method impractical, the FFQ data
collected through the FCST could be used to estimate UFCR. The survey includes a booklet with
images of a variety of species, portion amounts and fish preparation methods. The questions that
collect more qualitative information have been well vetted, although users may wish to modify these
questions. A drawback of the FCST is that modification of these questions may require the services
of a Microsoft Access programmer and associated consultants.
61
-------
As it stands, the automated version of the FCST can be conducted in-person or via telephone using
a desktop computer, laptop, or other mobile device. If an automated survey is not possible, then the
interview can also be implemented as a hard-copy questionnaire. This would allow for the mail data
collection mode to be used. Additionally, using the hard-copy questionnaire as a guide, the
instrument could be converted into a web survey. Links to both the automated version and a hard-
copy version are available on EPA's web site (https://www.epa.gov/fish-tech/epa-guidance-
developing-fish-advisories). along with a User's Guide, Supervisor's Guide, and other accompanying
documentation (copies of the guides, documentation, and hard-copy questionnaire are provided in
Appendix D). These resources are being provided to researchers free of charge by EPA. Other
survey instrument options are presented in Appendix B.
4.3 Use of Visual Aids
When consumption of fish from specific geographic areas is to be assessed, maps of the area of
interest can be provided to ensure that respondents understand the questions. Use of a map helps to
standardize the frame of reference when reporting local consumption. Geographic information
systems resources are readily available at the federal, tribal, state, and county levels.
As local names for fish may vary, it may also be helpful to provide photographs of the fish species
of interest, keeping in mind that some consumers may not be familiar with the appearance of the
intact fish. Inquiries should be performed with local experts prior to finalizing the wording of the
survey to make sure that common local names are used when possible.
Visual aids are commonly used to help respondents estimate portion sizes when collecting dietary
intake. These aids may include life size photographs of relevant portions of various fish
preparations, physical portion size models, or common objects. There are three types of portion size
models: organism (e.g., clams, mussels or shrimp consumed), preparation (e.g., fish fillets, fish jerky),
and volumetric (e.g., cups). For each model used, there should be an associated raw and cooked
weight. Volumetric models might be used to assess consumption of fish/shellfish tissue (e.g., crab
meat) or composite dishes (e.g., fish stews). Model depiction is particularly important in cases where
there is a great deal of difference in tissue mass as a result of preparation techniques (e.g., dried fish).
Lately, there has been interest by dietary researchers in using common objects as portion aids, such
as a checkbook or deck of cards, rather than the traditional measuring cups and spoons. Appendix
D contains information on preparing photographs on portion size.
It should also be noted that misidentification of types of fish can present a major obstacle for
consumption surveys that are interested in a limited set of species. To reduce misidentification,
visual aids are often used to assist the respondent with correct identification of fish consumed. This
step is a challenge for telephone surveys and other interviews conducted without the presence of an
interviewer. If the interview is not conducted in-person, procedures should be implemented for
sending, or otherwise making available, visual aids such as maps and fish species photographs for
use during the interview.
62
-------
4.4
Respondent Body Weight
The body weight of survey respondents is an important variable in analyzing fish consumption data.
Tribal, state, and local governments often have an interest in reporting fish consumption on a
per-body weight unit basis, as lighter individuals may be able to consume less fish before they reach
safety thresholds for contaminants in fish. In order to understand variability in fish consumption
within a population, it can be useful to collect and then control for the weight of the survey
respondents. If study respondents are, on average, heavier or lighter than the overall population,
then the consumption rate may differ (Trondsen et al., 2004). Such differences can only be
accounted for if respondent weight is collected. If respondents are expected to report on their entire
household, then the weights of all household members in question should also be captured. Ideally
the bodyweight data would be collected by weighing survey respondents, although in many instances
this may not be practical.
Although it is known that self-reported weight can be inaccurate (Connor et al., 2007), obtaining
self-reported weight is still a generally accepted survey practice. Using data from the NHANES, it
has been found that men tend to overreport and women tend to underreport their weights.
Variations can also be seen depending on the age of the respondent and whether or not he/she is at
a normal weight, underweight, or obese (Merrill & Richardson, 2009).
CDC's BRFSS survey includes a question on self-reported weight: "About how much do you weigh
without shoes?" Average weights for populations within states are available from the BRFSS annual
data bases (http://www.cdc.gov/brfss/annual data/2Q13/pdf/CODEBOOK13 LLCP.pdfi. These
weights could be used if the population of interest is the state-wide general population.
The amount of fish consumed is the main variable of interest when estimating UFCR. It is
important to understand that the terms "portion" and "meal" are not necessarily equivalent. For
example, a meal may consist of shrimp consumed as an appetizer and baked bass as a main course.
When collecting the portion size data from the respondent, the survey team needs to differentiate
between the fish species in this example and collect a portion size for the shrimp and a separate
portion size for the bass. Additionally, if the same species is consumed more than one way (i.e., fried
shrimp and boiled shrimp) then the respondent needs to provide portion size data for both
preparations.
Generally, respondents estimate the size or amount of the fish that they consumed (for example, the
size of a deck of cards or a checkbook or a half a cup). For mixed fish dishes, the weight of fish per
unit volume may be ascertained from recipes and combined with estimates of consumption volume
to yield estimates of fish intake. These amounts are converted to gram or ounce equivalents for
In order for respondents to more accurately estimate their portions, visual cues are used. Pictures of
fish portions of varying sizes (e.g., 4 oz., 6 oz.) on standard size plates or portion models help
respondents to estimate the amount of fish that they consumed. As discussed in Section 4.3, if the
4.5
Portion Sizes
analysis.
63
-------
interview is not conducted in-person, these pictures of fish meal sizes may be sent (or otherwise
made available) to respondents for their use during the interview.
While it is ideal for respondents to provide the amount they consumed in raw weight, they may not
know the weight of the fish unless they were the one who bought or prepared it. Many surveys
collect the as-consumed weight or amount and convert it to raw weight during analysis (see Section
7.3.1). It is important for the survey to consistently collect either raw or as-consumed amount or to
ask and record which the respondent is providing.
If respondents are providing portion sizes for as-consumed fish, it is necessary to ask how the fish
was prepared (e.g., breaded, battered) and how it was cooked (e.g., poached, fried, baked). This will
allow for conversion from as-consumed amounts to raw weight if that is the estimate of interest. See
Section 2.3.4 for a list of preparation attributes that may be of interest.
If the survey is using the FCST, and a respondent does not know the weight of the fish he/she
consumed, it is designed so that a respondent views a photograph of a fish portion (e.g., a baked
salmon filet) that has a known raw weight (e.g., 4 oz.) and then estimates the weight of the fish they
consumed. This requires the survey team to weigh and cook fish meals of species of interest to
photograph for use in the survey. See Appendix C, Developing the Booklet of Photographs, for
more information.
Asking respondents to estimate frequency of consumption by species can lead to overestimation of
fish consumption. For example, individuals may overestimate their salmon consumption if they are
asked about consumption of many individual salmon species. This overestimation is likely lower in
cultures where fish consumption is important and members of the population are very familiar with
consumption of specific salmon species. An alternative approach could be to ask about frequency of
consumption of a group, e.g., salmon, and then ask the respondents to provide a percentage
breakdown of the individual species they consume (Chinook, Coho, pink, sockeye, etc.).
Another consideration regarding collecting data by species is that it may be difficult for respondents
to accurately answer questions concerning factors that vary by species. For example, a survey
question may ask what fraction of the fish that is consumed is harvested. This is quite difficult if
some species of fish are almost always harvested and others are almost always purchased. The
respondent has to engage in a difficult cognitive exercise to come up with assignment of fish source
percentages. Thus these types of questions are better posed on a species-specific basis, for example,
"what fraction of salmon are harvested or caught by you or someone you know?"
The variety of fishing practices, whether it is recreational catch and release, recreational catch and
consumption, or subsistence-based, will affect a respondent's report of consumption. It will be
important to collect fishing behavior preferences from those respondents who report fishing
4.6
Collecting Data by Fish Species
4.7
Fishing Practices and Locations
64
-------
behavior. This can include whether they consume the fish they catch, preferences for type(s) of fish
they may fish for and subsequently consume or give away, and locations respondents may prefer.
For on-the-bank or intercept surveys, it will be important to evaluate locations frequented by fishers.
Some locations may be avoided by fishers due to an advisory or information from other fishers that
some locations may be undesirable. If fishing locations are not well identified and missed, this may
result in some groups or minority populations not being covered by the survey, especially if they
tend to avoid well-populated areas. For location-based surveys, good sources for identifying
locations are park officials or personnel from state departments of fish and game. Also,
reconnaissance of the waterbody from boats to identify where individuals spend time fishing can
assist with determining the best locations. It is also important to note that if the objective of the
study is to estimate unsuppressed FCR, then conducting surveys at locations with fish advisories
would not be preferable. Another option is to conduct the survey of other waterbodies with similar
attributes without an advisory.
Advisories, creel limits, and fisher observation of caught fish will influence fishing practices and
subsequent consumption rates (for example, see Beehler et al., 2001). Collecting information on
these behaviors can help to identify possible reasons for consumption suppression and may allow
for quantification of suppression for specific populations of fishers.
4.8 Data Capture and Management Methods
A number of interrelated factors can help to determine the optimal data capture method for a
consumption survey, of which the most important may be data collection mode, questionnaire
complexity, sample size, and available resources to process data.
For self-administered mail questionnaires, a paper-and-pencil instrument can be used to collect the
data from respondents; once the questionnaire is returned, the answers need to be captured in a way
that facilitates analysis. If the number of questionnaires is small, then it may suffice to have a staffer
extract the questionnaire data. However, as the number of questionnaires grows and/or the
complexity of the questionnaire increases, so, too, does the effort necessary to capture the data and,
in turn, the likelihood of coders introducing error into the data capture process. A computerized
data-capture system can help to reduce the time necessary to code a large number of questionnaires
while reducing the likelihood of introducing error into the process through coding and key entry
errors. A computerized data-capture system requires sophisticated software that can accurately read
respondents' answers on each page. The FCST offers a computerized format.
For self-administered web-based questionnaires, data capture is automatic as the respondent inputs
responses into the computer directly. Up-front time and costs are required to upload and/or
program the questionnaire; however, once this is complete, the survey can be replicated across
respondents and survey administrations (e.g., seasons) and results can be extracted at any point. If a
self-administered web-based survey is planned, web accessibility by the target population should be
taken into account.
Questionnaire complexity should be a consideration when using self-administered (i.e., paper and
pencil) instruments as there is no interviewer to assist respondents when they're confused or if they
65
-------
make a mistake. Particularly with self-administered mail questionnaires, respondents must be able to
understand what the questions are asking of them, which questions apply to them, and how to
appropriately answer and navigate the instrument. If this is not clear to the respondent, then the
questionnaire may yield inaccurate and/or incomplete data, or in some cases result in item or
respondent nonresponse. Having a toll-free number that respondents can call with any questions
may help. An advantage of a web-based or computerized instrument is that the questions
respondents see or hear can be tailored based on their responses to previous items. For example,
respondents who indicate they have not consumed fish can be directly moved past likely follow-up
questions on fish consumption. Such tailoring is not possible in a paper-and-pencil questionnaire but
can be approximated by including clear instructions to respondents about which questions they
should answer based on their previous answers. However, unlike a web or computerized survey,
respondents cannot be forced to adhere to such instructions and data cleaning will be necessary,
along with procedures to address errors of commission or omission. When transcribing data from
paper forms into electronic data storage, it is important to implement checks on transcription
quality.
For surveys conducted over the phone or in-person, it is the interviewer who must capture the data
provided by respondents. Similar guidelines as described in the previous paragraph for self-
administered questionnaires apply to interviewers capturing data; it may suffice to have the
interviewer fill out a paper-and-pencil questionnaire if the questionnaire is short and simple and the
resources are available to extract the data and enter it into an analyzable format. Similar to a web-
based questionnaire, a Computer-Assisted Telephone Interviewing (CATI) program can be used to
facilitate more complex questionnaire administration and simple and error-free data processing.
As mentioned, the FCST (discussed in Section 4.2) can be used as a computerized instrument, although a
hard-copy version is also available. Based on the availability of computers and programming resources
that may be available for a particular study, a decision will need to be made as to whether automation is
feasible. Computerized data collection (vs. capturing data on hard copy) is usually desirable and results in
higher quality data, due to the ability to automatically conduct edit checks during the data collection
process and directly input data into a file.
4.9 Interview/Questionnaire Length
It is widely accepted that interview/questionnaire length is correlated with a respondent's likelihood to
participate in the survey. The longer the survey, the greater the burden and the less likely respondents will
be willing to participate. Groves et al., 1992, cites interview length as a deciding factor in the decision to
participate in face-to-face surveys. However, Bogen, 1996, has found the influence of interview length on
participation behavior to be "weak and inconsistent." Instead, the effect of interview length on survey
participation may actually be related to timing of the survey request or the perception of length. In a
Danish study, Hansen, 2007, observed increases in participation when the expected length told to the
respondent was shortened from 20 minutes to 15 minutes (which reflected an actual decrease in survey
length) for a telephone interview. The question for interview length then becomes, how long is too long?
Other factors should be considered, including the relevance, or saliency, of the survey topic. Tolerance
levels of participants for lengthier interviews may be increased if the subject matter is particularly salient
or important to them.
66
-------
Another concern with interview length is the effect that a long or burdensome interview can have
on data quality. Holbrook et al., 2003, found that as respondents got to the end of a long telephone
interview, they observed more satisficing behaviors. Respondent satisficing can take many forms,
but generally involves respondents taking cognitive shortcuts to select the first acceptable (or easiest)
response. Satisficing behaviors can lead to substantial measurement biases (for more on satisficing,
see Krosnick, 1991).
While it can be tempting to researchers to develop an exhaustive questionnaire covering a wide
range of topics of interest, it is critical that the questions included in the survey instrument map to
research goals or objectives. Researchers need to consider the response burden placed upon the
participant to ensure that quality data are collected. It is also important to consider the population
that will be interviewed when thinking about interview length. For example, creel or on-the-bank
intercept surveys may need to be shorter as fishers may be unwilling to provide time for a lengthy
interview after a day of fishing (Washington State Department of Ecology, 2012). Conversely, the
narrative structure of sharing or conveying information in many Native American cultures may lead
to much longer interviews and for these situations guided or structured interviews may be more
appropriate as opposed to strict adherence to written questions.
4.10 Qualitative Methods to Support Survey Instrument
Development
Qualitative research is beneficial during the questionnaire development process or development of
other survey components (e.g., location maps, identification of fishing locations, fish pictures). This
will help ensure that survey questions appropriately map to the research objectives. Testing will also
help identify difficult questions and concepts or ambiguous terms that are not familiar to the
respondent. Key informants that are familiar with local or customary fishing practices, culture, and
fish preparations can be identified and assist at this stage. Terminology can be clarified through this
qualitative research, to ensure that data being collected are those which were intended. Names for
certain species of fish, or how to identify fish that are bought versus those that are caught, are just a
few of the types of clarifications that can be ascertained. There are commercial software packages
available that are useful for analyzing qualitative data.
In using qualitative research methods, we must carefully consider what is necessary to reach the
research goal. These methods can, for instance, be used to identify access points and locations (for a
specific body of water) that may be frequently used. Fishers may use specific points for early fishing
hours but may migrate to a different location as water temperature changes. To understand this type
of pattern and ultimately build a frame for a creel or on-the-bank survey, the researcher could
perform cognitive interviews on a broad spectrum of fishers. Frequent fishers' behavior may differ
from that of recreational fishers. The researcher should also provide materials so the participants can
identify locations (on a map, for example).
When the objective is to define how to measure consumption, qualitative research methods can be
useful to help understand how people identify various types of fish or how they think of
consumption. Frequent fishers may identify fish more easily by sight or be familiar with the types of
fish they catch. Recreational fishers may better recognize specific names for fish, or local names,
rather than other names/labels that are used.
67
-------
A difficult question to answer is how many respondents are needed for qualitative research
approaches. This type of research can be done in groups or on an individual basis. Identifying
fishing locations or how people identify fish is an activity that can be done in a group or individual
setting. Testing questionnaires is something that should be done in a one-on-one setting. The
number of participants can vary but should not be so many that it overwhelms the research team.
The number also should be sufficient enough to identify common themes or problems that can be
distinguished from idiosyncrasies that may not be an indicator of a real problem. For more guidance
on qualitative testing methods, see Willis, 2005.
4.11 Pilot Testing
After the survey instrument and accompanying procedures have been developed, pilot tests are
conducted to identify possible shortcomings within a survey instrument or procedures that generally
cannot be identified during qualitative testing or expert review. A pilot test is a small-scale test of
survey instruments and procedures conducted under actual survey conditions (e.g., within a realistic
environment using respondents similar to the survey's targeted respondents). It may also serve to
identify unique interviewing issues specific to the populations of interest. For example, a common
interviewing technique when a respondent does not understand the question is to repeat the
question. This may present a cultural issue in some populations, for example when a younger tribal
member is interviewing a tribal elder, as the younger tribal member may feel that they are insulting
the intelligence of an elder. Ideally, a pilot test can be done early enough in the survey period to
allow for small changes, as needed, to the questionnaire and/or study procedures prior to the start
of actual data collection. Data resulting from pilot testing are typically not included as part of the
analytic data set, and the individual respondents interviewed during the pilot test are typically not
interviewed during the actual study.
Pilot tests can vary in size and scale, depending on resources and time available, but they should
always have set goals or objectives before they are conducted. For example, a pilot test can be
conducted to evaluate new procedures, or for data collection efforts involving multiple processes
(e.g., multiple contacts, diaries, etc.), to ensure that all appropriate procedures are in place and
operating as planned. A pilot test can also be used to help estimate sample sizes (by obtaining
information about response) or prevalence of a behavior in the population. Finally, pilot testing can
be useful to help identify to what degree respondents are able to provide information desired by the
researcher. For example, if a researcher is interested in knowing the consumption of a specific type
of fish, pilot testing can help identify whether respondents are aware of the fish species they
consume.
Ideally, the pilot test will try out not only the questionnaire but also give a thorough workout to
production or field procedures, including methods of finding participants and use of data capture
instruments (paper and pencil, computer-assisted telephone interview, or computer-assisted personal
interview), as well as coding and key entry procedures. Attention should be paid to the secure
storage of collected survey data to protect personal identifying information (PII). The pilot test of
the questionnaire should include pilot respondents who will give all aspects of the questionnaire a
thorough workout. The following list of characteristics is an example of participants who might be
recruited for a pilot test of a questionnaire in an ethnic minority group.
68
-------
¦ Age: elders and younger participants
¦ Gender: male and female
¦ Fishing: fishers and non-fishers
¦ Source of fish: primarily eat at home vs. eat out frequently
¦ Income: low-income and medium- or high-income;
¦ Food preparation: persons who do and who do not usually prepare food for themselves
or the household
¦ Recent immigrants vs. members of the ethnic group who have lived in the U.S. for
longer periods of time
¦ Members of all of the various ethnic groups that might be encountered in the survey
It is also important to include individuals across the spectrum of lifestyle from traditional to modern,
and if surveys are planned in other languages, these other versions should also be tested during the
pilot.
A given pilot participant may satisfy more than one of the criteria. For example, someone may be an
elder who follows a traditional lifestyle and fishes. Generally, a minimum of 10 persons should be
recruited as pilot participants, although more may be needed. Note that for federally funded
research, approval by the Office of Management and Budget (OMB) may be required and if so, pilot
testing is strongly encouraged, but may be limited to 9 or fewer respondents thus not requiring the
additional time needed for formal approval of an information collection request (ICR). See OMB
Standards and Guidelines for Statistical Surveys at
https://www.whitehouse.gov/sites/default/files/omb/inforeg/statpolicy/standards stat survevs.p
df. More information can be found at https://www.whitehouse.gov/omb/inforeg infocoll.
Sequential improvement of the questionnaire after every participant or after several participants
during the pilot is a process that ends when no more changes to the questionnaire or procedures are
needed. It is likely that some minor changes to the questionnaire or procedures will be needed
during the main survey, but if the pilot is well done, these changes should be minimal.
69
-------
5
SURVEY IMPLEMENTATION AND OPERATIONAL
CONSIDERATIONS
5.1 Introduction
Once the sampling frame, sample size, mode of administration, and type of survey instrument have
been established, there are important considerations for fielding a study. The quality of a survey is
dependent on the quality of the data collected. Data collection is usually the most labor intensive
aspect of a survey and also the most expensive. Therefore, it is important to consider
implementation and operational issues during the development process. This is the fifth and final of
the basic steps in the design and development of a survey for estimating fish consumption rates
listed in Chapter 1 — Consideration of implementation and operational issues. This chapter describes
some of the key issues. There are many other factors to consider in operationalizing a survey,
including hiring and training of interviewers, interviewer supervision, interviewer assignments, etc.
These topics are not within the scope of this document; however, it is important for the researcher
to consult with a survey specialist to address these issues.
5.2 Defining Eligible Respondents
For operational purposes, it is necessary to determine who is eligible to respond to the survey. Once
the target population has been determined, consideration should be given to situations in which
sampled persons might not respond for themselves. Several common situations are discussed in this
section.
5.2.1 Household Reporters
If the survey will be collecting information about all eligible persons in the household (this includes
both screening surveys and consumption surveys that ask about the behaviors of each eligible
household member), it is important for study procedures to specify who is eligible to serve as a
respondent for the household. (See Section 3.3 for a discussion of within-household sampling, in
which a sampled person is selected from among all eligible persons in the household.) Generally,
surveys require that household respondents be adult members of the household. More restrictive
requirements might necessitate it be the primary respondent for the household (e.g., the person
most knowledgeable about the household members' consumption).
5.2.2 Children
If the target population includes children, it will be necessary to determine who is eligible to respond
for the child. Young children themselves may not be able to serve as respondents due to both data
quality concerns and the need to obtain parental consent. Researchers should also be advised that
there are special regulatory requirements generally accepted across Federal, state, and local agencies
70
-------
that provide additional protection for children who would be involved in research. Additional
information can be found at http: / /www.hhs.gov/ohrp/policy/populations /children.html and
https:/ /www.epa.g-ov/osa/products-and-publications-related-human-subjects-research.
However, with carefully considered procedures, food consumption surveys do sometimes obtain
information from children directly, depending on their age. For example, NHANES collects food
intake data for children under the age of 6 from a proxy respondent; proxy-assisted interviews are
conducted with children 6-11 years of age; and participants 12 years and older complete the dietary
interview on their own.
In many cases, the respondent for a child is specified as the parent or guardian of the child. For fish
consumption surveys, it may be preferable to require that the respondent be a parent or guardian of
the child who is knowledgeable of the child's food intake. In shared custody arrangements or other
situations in which the child resides in more than one location, other special considerations may
come into play when collecting consumption information.
There may be other situations in which proxy respondents are necessary. For example, if a person is
disabled or elderly and cannot respond for themselves, a person who is knowledgeable about their
intake may be interviewed. While proxy respondents can be used to collect information about food
consumption, the concern is with data quality. However, in some situations, the only alternative to
proxy reporting might be nonresponse. When weighing whether to permit proxy respondents, the
researcher should consider what particular situations might give rise to a need for proxy respondents
(i.e., a preference for proxy respondents over complete nonresponse). Consideration should also be
given to what stipulations might be placed on the potential proxy respondent (e.g., a requirement
that he/she is knowledgeable of the sampled person's food intake).
Careful pre-contact procedures, along with a persuasive introduction to the survey, can serve to
increase study participation. The use of pre-notification letters is a useful method for introducing the
study and alerting potential respondents of future contact by the survey organization. Research has
shown that pre-notification letters can lead to increased response in mail surveys (Dillman et al.,
2009), RDD telephone surveys (de Leeuw et al., 2007; Link & Mokdad, 2005), and telephone
surveys utilizing lists (Goldstein & Jennings, 2002). For in-person surveys, these letters may be
delivered by the interviewer to provide the respondent additional information or amplify study
legitimacy.
The content of the letter should include an appeal for participation, highlight the study sponsor,
convey the importance of the study, and provide information for what is expected of the
respondent. Letters that are lengthy (generally more than one page) or provide an overabundance of
5.2.3
Proxy Respondents
5.3
Pre-contact Procedures and Introduction of the
Study
71
-------
information should be avoided, as respondents are unlikely to read letters that appear burdensome.
For more information on letter content and design, see Dillman et al., 2009.
The timing of the delivery of the letter is important so that it directly precedes the substantive
encounter (e.g., a mailed paper survey or interviewer contact). If the letter is mailed too far in
advance, the connection between the pre-notification letter and the survey contact may be forgotten.
Pre-notification letters should be sent a few days to a week in advance of the survey or initial
interviewer contact.
Other pre-survey activities can include community outreach using radio and/or TV announcements
or social media messages targeted to the study population. Identification of the primary media
utilized by the target population and preparation of communications designed for those media to
publicize the survey may also be helpful. See, for example, a fish consumption survey for Asians and
Pacific Islanders in King County (U.S. EPA, 1999). Such outreach efforts must be carefully planned
so they will not bias or otherwise affect the outcome of the survey or influence the types of
individuals who may or may not volunteer to take part. Although these types of efforts can be
helpful in recruiting individuals with rare characteristics (e.g., those who consume certain types of
fish), sometimes the breadth of such media coverage results in more volunteers than the research
can sustain (particularly in light of resource constraints).
Additionally, eliciting support from community leaders for surveys of targeted groups can also be
important for the success of a survey. Including endorsements of the survey by community leaders
in communications may enhance survey response and credibility.
5.4 Informed Consent for Interview
Informed consent is the process by which the prospective research participant is provided with
sufficient opportunity and information about the study to decide whether or not to participate.
Ethical standards specify that researchers may not conduct research on human subjects unless they
have obtained consent from the respondent or the respondent's proxy. Depending on the type of
data collection, oral or implied consent may be appropriate. For in-person interviews, written
consent is needed.
For Federally-funded studies, as well as many state and local surveys, planned consent procedures
must be reviewed and approved by an Institutional Review Board (IRB). IRB review is also
important and sometimes required for non-Federally funded studies. The IRB will ensure the
protection of human subjects based on regulations from the Office for Human Research Protections
(OHRP) within the U.S. Department of Health & Human Services. If the project will include
protected populations such as pregnant women or children, additional review may be required, and
review boards will require assurance that the study procedures include protection of such groups.
Additional information about regulatory requirements that govern research on human subjects may
be found at http://www.hhs.gov/ohrp/index.html. If appropriate for a specific research endeavor,
some consideration should be given to using IRBs that have experience with the unique cultural,
social, and psychological impacts on Native Americans.
72
-------
The IRB of record for the study will require information on the informed consent procedures to be
followed. For many study designs such as mail or telephone data collection, some informed consent
procedures will be impractical. The IRB can waive requirements for written informed consent
procedures, but generally only under the following conditions:
¦ Research poses minimal risk to study participants;
¦ Subjects' rights or welfare will not be affected by participation;
¦ It would be otherwise impractical to conduct the research; and
¦ Full disclosure of the risk and benefits of participation will be provided to participants.
Although written informed consent may be waived in certain instances, respondents must always be
informed of their rights as research participants. If approved by the IRB, verbal consent may be
used to inform them of these rights. Informed consent requires that the consent to participate in
research is informed, understood, and voluntary. Informed consent must be conducted in a manner
that minimizes coercion or undue influence. These ethical principles are in the Belmont Report and
have been codified in the Federal regulations to protect human research subjects, found at 45 C.F.R.
§ 46.116. Researchers must ensure that informed consent documents meet the regulatory
requirements, including state and local laws where the study takes place. All materials submitted
need to be in final form. If changes are made after submittal, the new materials needed to be re-
submitted. This could increase costs. The following nine required elements of informed consent
must be communicated to a potential research subject.
1. Identify the project as research, explain its purpose, and state the expected duration of
the subject's participation
2. Describe the procedures to be followed
3. Describe any foreseeable risks (i.e., probability of physical, psychological, social, or
economic harm or injury occurring as a result of participating in the study)
4. Describe any potential benefits to either the individual or society that can be expected
from participating in the research
5. Disclose alternative medical treatments, if any
6. Explain how the confidentiality of any data that identifies the subject will be maintained
7. Provide the names and contact information for two people—one to contact to ask
questions about the research and another to contact about the subject's rights as a
research participant
8. State that participation in the research is voluntary
9. Explain whether compensation or medical treatment will be available if any injury
occurs (required for more than minimal risk studies)
IRBs may vary in having additional requirements, and there may be additional criteria for protected
populations. Researchers are encouraged to meet with their IRB early in the project design to
establish necessary requirements for approval. Fish consumption surveys may often be timed to
coincide with specific seasons; therefore, it is important that adequate time is allocated for
appropriate review of the project before any data collection activities are scheduled to begin.
Tribes are sovereign nations, and researchers are required to obtain specific permission from the (or
each) Tribe to conduct research. In addition to the study IRB, the study should be prepared to apply
for a tribal research permit, which may require seeking approval from a separate tribal IRB or the
tribal government. The Indian Health Service (IHS) has regional IRBs, such as the Portland Area
73
-------
Indian Health Board. Some tribes use other IRBs such as the Northwest Indian College's IRB. This
will vary from tribe to tribe. In many cases, studies will need to seek an agreement to the use of the
intellectual property generated by the study from the tribe and be prepared to share all information
collected with the tribal authority. In addition to IRB requirements, tribes may also have provisions
that require the research activity does not damage their cultural heritage or identity.
EPA also has information on human subjects review that can be accessed at
http://www.epa.gov/osainter/phre/hsr.htm and https://www.epa.gov/osa/products-and-
publications-related-human-subjects-research. For additional information, the reader is encouraged
to review a thorough discussion on this topic by Harding et al., 2012.
As mentioned previously, the timing of the data collection will affect consumption rate estimates.
Ideally, a survey would capture seasonal variation in fish consumption by planning on a full year of
data collection. However, this may not be feasible. Questions determining how soon information
from the study will be needed, or over what seasons the data are needed, should be considered to
help determine when the study should be conducted. Fishing activity might be undertaken by the
majority of fishers only during the summer; however, ice fishing is popular in some areas of the
northern United States. In addition, fish caught in one season might be preserved (e.g., smoked or
frozen) and consumed later, indicating that exposure to tissue contaminants might be equally
important year-round. If the data collection period misses the peak seasonal periods, an
underestimate of consumption would result. Conversely, measuring consumption only during
seasonal peaks may result in an overestimate of consumption. One way to address seasonality is to
divide the sample into equal proportions and interview each over fixed time periods until the entire
sample has been interviewed. Unequal numbers of interviews per time period can be accommodated
by weighting approaches. Another approach might be to conduct interviews during two periods, one
of low consumption and one of high consumption. The research goals should dictate when and,
possibly, how often consumption data will be needed.
Low literacy rates can be problematic if prevalent within the target population or an area that has
been sampled. The effectiveness of some study designs can be impacted by low literacy rates. For
example, self-administered survey designs (mail, web) will likely see a negative impact on survey
response. This can result in biased estimates if fish consumption behavior is correlated with literacy.
The National Center for Education Statistics (NCES) provides state and county estimates of low
literacy, defined as the percent lacking Basic Prose skills in the National Assessment of Adult
Literacy (NAAL), which includes those who score Below Basic in prose and those who could not be
tested due to language barriers (NCES, 2003). NCES has an online tool that provides state and
county estimates of the percentage of the population lacking basic prose literacy skills
(http://nces.ed.gov/naal/estimates/stateestimates.aspx). In some counties in the United States, the
illiteracy rate is over 30 percent (NCES, 2003).
5.5
Impact of Seasonality on Data Collection Schedule
5.6
Language and Literacy Issues
74
-------
Literacy of the population should be considered both when developing survey questions and in
choosing a mode of administration. If a self-administered survey mode is under consideration,
estimated literacy rates should be reviewed for the target population and taken into consideration at
the study design phase. For study designs where interviewers administer the survey, there may be
self-administered components (e.g., food log or diary) that could be affected by literacy rates.
Interviewers should be trained to identify respondents who may have difficulty with these
components. In many cases, respondents may be reluctant to reveal literacy issues to the interviewer.
In the design of survey questions (as well as consent forms, letters, etc.), the designer should strive
to keep language to less than a 12th grade level. See Willis, 2005, and Fowler, 1995, and for further
guidelines on question design.
If the target population or the sampled area has a high percentage of non-English speakers, it is
important to initially review the prevalence of other languages spoken by members of the target
population. The language differences may be addressed through the use of questionnaires translated
into the predominant non-English language(s) spoken for self-administered surveys. For
interviewer-mediated surveys, it will be important to have trained data collectors who speak the
predominant languages within the sample. Best practices in survey translation should be followed in
order to tailor questions to the needs of a given audience while retaining the measurement properties
of the source. Planning for translation should be part of the study design (Harkness et al., 2010).
5.7 Survey Nonresponse
Survey response is a critical component of any survey and will have potentially strong influence on
measurement bias, sample size, and costs in terms of data collection effort. Trends for response
rates have shown an overall decrease in response over time for household surveys (Brick &
Williams, 2013). This same effect has been observed for natural resource based surveys (Connelly et
al., 2003). It is highly recommended that an experienced survey statistician be involved in planning
for nonresponse analysis, as it is not possible to proscribe a "one-size-fits-all" approach within this
guidance document. The application of methods for analyzing and ameliorating the effects of
nonresponse are extensive and require a solid background in survey methods as well as training in
nonresponse analysis.
Nonresponse is unavoidable and should be anticipated. Plans for dealing with nonresponse should
be a part of the study design and analysis methods. To minimize nonresponse, possible incentives
should be considered and approaches for encouraging cooperation should be built into the survey
design. These may include multiple contact attempts with selected respondents, as well as follow-up
reminders for nonrespondents such as postcards, letters, and telephone calls, if feasible. For self-
administered surveys, generally mail surveys, replacement surveys should be sent to nonrespondents
(Dillman et al., 2009). Following protocols outlined by Dillman and his colleagues, replacement
surveys are mailed to sample units that have not responded to an initial survey. For interviewer-
mediated surveys (e.g., in-person or telephone surveys), this will require procedures for re-contacting
the household and specialized training for interviewers attempting to convert nonrespondents.
The survey mode will have an effect on the response rate that can be anticipated. In-person surveys
generally have the highest response but also the highest per interview costs. Mail surveys often
75
-------
outperform or do as well as telephone surveys in terms of response rates. However, a telephone
survey may be necessary for complicated interviews. Web surveys generally have the lowest overall
response and often lack an available sample frame, but web surveys mixed with mail invitations to
complete the survey have shown promise (Millar & Dillman, 2011).
The degree to which nonresponse has a biasing effect will depend upon the relationship between the
propensity to respond to the survey and the characteristic of interest from the survey (Groves,
2006). If fish consumers have a higher propensity to respond than do non-consumers, this will result
in higher per capita consumption estimates. (Per capita consumption estimates include both
consumers and non-consumers as part of the population.) This will not be an issue if consumers
only are the population of interest (i.e., if non-consumers are not part of the target population for
the survey). However, if nonrespondents have a higher consumption rate than respondents, this will
result in an underestimation of fish consumption in both per capita and consumer-only
consumption rates.
Also, it should be noted that avidity bias can be a problem for fish consumption surveys. This can
arise when sample members disproportionately participate due to interest, perceived relevance, or
activity in the survey topic. For example, recreational fishers may be more likely to participate, but
family members or others who also consume the sport-caught fish may feel the survey is of less
interest or relevance.
Determining the degree to which nonresponse bias may be present requires planning for a
nonresponse bias study. While nonresponse analyses may not always be feasible for smaller or lower
budget surveys, it may be of interest to note that for studies requiring approval by OMB, standards
dictate that a nonresponse bias analysis be done when unit or item nonresponse suggest the
presence of bias, e.g., if response rates fall below 80 percent (OMB, 2006). Any nonresponse analysis
should be done in consultation with a survey expert.
The purpose of the nonresponse study is to detect whether there is a difference in the behavior of
interest between those who responded to the survey and those who did not. The number of cases
necessary will be driven by the estimate of the behavior for respondents. If the behavior of interest
is of relatively low frequency, it may be necessary to sample and interview a larger pool of
nonrespondents in order to have enough power to detect a difference between the two groups. For
studies with small sample sizes, a nonresponse bias study may not be feasible.
Generally, nonresponse bias studies include a subset of the survey items that are related to the
measure of interest or that will generate estimates. In large studies where the pool of
nonrespondents is large, cases from the pool of nonrespondents are subsampled, since these cases
may have previously received multiple contact attempts and will require much additional effort and
costs to contact all nonrespondents. Where possible, it is best to attempt a different mode of
contact. For example, if the survey was conducted by telephone, an in-person contact with
nonrespondents may be more successful.
76
-------
5.8
Methods for Respondent Retention
For some study designs it may be necessary to re-contact survey participants to complete data
collection. In the context of collecting data on fish consumption, an example of a design where re-
contact will be necessary is when multiple dietary recalls are completed (e.g., separate 24-hour or
multiple-day recalls). It will be necessary to have procedures in place for re-contacting respondents
to collect the second recall.
In order to mitigate respondent attrition for study designs where re-contact will occur over a short
period of time, researchers should inform the respondent of plans for future contact. In the case of
reluctant respondents, it can be helpful to inform them of the purpose of the later contact.
Additionally, it is beneficial for retention purposes to set a time or an appointment for the next
contact during the first data collection, although this could affect respondents' consumption
behavior for the day they know they will be reporting about. An appointed day/time for the
interview can help commit the respondent to the task and identify optimal times for re-contact.
Finally, if possible, it may be helpful to have the same interviewer conduct the first and subsequent
data collections, as research has shown this can help with respondent retention (Hill & Willis, 2001).
For study designs that will follow respondents over a longer period of time (e.g., multiple seasons or
years) to measure individual-level change in consumption over seasons, for example, one or more
intermediate contacts with the respondent may be necessary. Intermediate contact (i.e., contact
between substantive data collections) will help to keep the respondent involved in the study, be
reminded of past cooperation, and help maintain up-to-date contact information. At baseline, it is
important to obtain alternate contact information for each respondent, including alternate telephone
numbers, e-mail addresses, and/or contact information for close relatives or friends. This
information is useful in the event that original contact information becomes outdated during the
data collection period. A protocol for interim contacts between data collection events should be
developed based on the study schedule and budget, and can include one or more of the following
approaches: periodic telephone or e-mail reminders to the respondent, reminder postcards regarding
upcoming activities, and/or letters that could include information about the study progress, as
appropriate. To ensure adequate sample sizes for analysis, it is very important to retain respondents
in the study who have previously participated, and careful attention to retention efforts should take
place as part of overall study planning
Maintaining the confidentiality of information or responses provided by respondents is important to
the integrity of all surveys. Respondents will be asked to voluntarily provide data that they may
generally be unwilling to share with others. While fish consumption can appear to be an innocuous
behavior or topic, under some circumstances it can be quite sensitive. For example, pregnant women
may be advised to avoid certain species of fish or shellfish. Respondents may not want others to be
aware of how much (or how little) fish they consume. Assuring respondents that their information
and privacy is of utmost importance to the researcher can help foster cooperation with the survey
request.
5.9
Confidentiality
77
-------
Many state agencies offer legal protections for the confidentiality of survey data and also may have
specific requirements for researchers to follow. Researchers should verify they are in compliance
with any state, local, or tribal regulations for maintaining respondent confidentiality. Additional
protections can be granted for surveys in the form of certificates of confidentiality that are available
through several government agencies (for example, see
http: / /grants.nih.gov/grants /policy/coc/index.htm). For EPA studies see
https:/ /nepis.epa.gov/EPA/html/DLwait.htm?url=/Exe/ZyPDF.cgi /PI 0012LY.PDF?Dockey=P
10012LY.PDF
Assurances of confidentiality provided to respondents are an important part of gaining cooperation
and trust during the interview process. All survey information must be kept confidential, and any
plans to share data must be explained to participants during the informed consent process prior to
data collection. Steps must be taken to separate survey data from personal identifying information
(PII), and procedures for doing this must be described in the IRB submission(s). PII should be
removed from the data files. At a minimum, this involves replacing identifiers, such as names, with
anonymous ID numbers. For public use data files, more intensive procedures involve data disclosure
analysis to verify respondents cannot be identified due to a combination of sampling (e.g.,
geographic area) and their unique survey data. Procedures for anonymizing data are available that
preserve the estimates generated by the survey data (see
http://fcsm.sites.usa.gov/committees/cdac/ for more information).
Each person working on the study must be continuously aware of his or her responsibility to
safeguard survey data and identifying information. Names or other information about study
participants should never be divulged to anyone except the research team. Data collectors should be
instructed to never interview someone that they know personally. This may lead to response bias
and a possible breach of confidentiality. All members of the research team are under the same legal,
moral, and ethical obligations to protect the privacy of those participating in the research. All
members of the research team, including data collectors, should sign a Pledge of Confidentiality that
is prepared by the survey team. Additionally, specific training on confidentiality and privacy
procedures may be required for all staff working on a research project. For more information related
to confidentiality and tribal communities, see Harding et al., 2012.
5.10 Data Availability
There is an expectation that data used for regulatory purposes, e.g., for developing water quality
standards, would be made publically available. It is important for the data to be made available for
review in order for all stakeholders to be able to understand, evaluate, and comment on the basis for
the regulatory requirements that are developed. The survey team can ensure confidentiality and
address privacy concerns by anonymizing the data. Ensuring the availability of the data preserves
transparency of regulatory decisions. As described in Section 5.9, procedures for anonymizing data
are available that preserve the estimates generated by the survey data (see
http://fcsm.sites.usa.gov/committees/cdac/ for more information).
78
-------
6
SUPPRESSION
As described in Section 2.3.5, a suppression effect occurs when a FCR for a given population,
group, or tribe reflects a current level of consumption that is diminished from an appropriate
baseline level of consumption for that tribe, population, or group. For most survey goals and
objectives, it is important to assess suppression, in particular if the resulting FCR will be used in
regulatory action. Additionally, if one objective of the survey is to investigate consumption trends
over time, and there are no survey data available from comparable populations, the techniques
presented in this chapter can be helpful in estimating past consumption.
There are a variety of methods and approaches that can be used to estimate an unsuppressed FCR.
One approach is to recreate the historic, ecological conditions through the use of historical records
and documents and apply knowledge of nutrition, natural resource use, and activity levels to
estimate the historic FCR or heritage rate. In many cases, heritage rates may be the only practical
way to estimate unsuppressed rates — that is, free from the biasing influence of suppression effects.
While this approach could be particularly useful for tribal fishing populations, it may not be feasible
for other populations or situations where there is a lack of necessary historical information or other
data. Alternative methodologies for establishing unsuppressed rates are presented in Section 6.2.
Heritage rates of fish consumption can be validly estimated by detailing the ecological conditions
under which the population lived and how the population utilized the available natural resources.
The subsistent (or traditional, historic) diet can then be recreated and the fraction of that diet which
was fish can be determined. (Harper & Walker, 2015a, 2015b; Harper et al., 2012; Harper et al.,
The data and information can come from a variety of sources. The use of historic, ecological data
can provide the types and amounts of natural resources that were available to the population during
the time period of interest and how they varied throughout the year. This information includes
ecoregion maps, climatological data, watershed data, and species composition data. Ecoregion maps
and detailed descriptions of the regions can be found at
http: / /www.epa.gov/wed/pages /ecoregions.htm. The National Oceanic and Atmospheric
Administration National Climate Data Center provides access to historical climate data (available at:
https:/ /www.ncdc.noaa.gov/cdo-web /) and paleoclimatology data (available at:
http: / /www.ncdc.noaa.gov/data-access /paleoclimatologv-data/datasets). The United States
Geological Survey has watershed maps that are available at
http://water.usgs.gov/wsc/map index.html. Species information can be found in state Natural
Heritage Programs (for example see http://www.azgfd.gOv/w c/edits/species concern.shtml) and
archeological records can be used to determine historical populations.
Anthropological and ethnographic documents can provide the type and amount of activities
undertaken by the population, how the natural resources were used, and the cultural significance of
these natural resources. Some populations may have historical written accounts of their fishing
activities, including quantities harvested and consumed, in which case this may be enough
6.1
Estimating Heritage FCR
2007).
79
-------
information to estimate the heritage FCR. The population's traditional environmental knowledge
(TEK) identified through cultural interviews, ethno-historical and ethno-botanical literature, oral
history, and historical records, such as those from early observers, can also provide the basis for the
determination of historical or traditional resource use patterns. A great deal of valuable information
can be gained through discussing experiences and oral histories with tribal members, especially tribal
elders.
From the ecological data and the anthropological and ethnographic information, a food pyramid or
food wheel can be constructed that represents the subsistent diet and shows the relative importance
of the identified dietary staples, such as fish, game, roots, honey, etc. An example is shown in Figure
6-1 (modified from Harper et al., 2007).
Figure 6-1. Food wheel showing the relative proportion of dietary staples in an example
subsistent diet
i Anadromous and marine fish
i Game, large and small
i Resident fish
i Roots and tubers
i Fruits and berries
Other vegetables
Fowl
Seeds, nuts, and grain
Honey
Note that one population may have more than one subsistent diet depending on the ecological
findings. For example, three subsistent diets have been identified for the Wabanaki — inland
freshwater, inland anadromous, and coastal based on three major ecological settings in the state
(Harper & Ranco, 2009; Harper et al., 2007). Other examples are available at
http:/ / superfund.oregonstate.edu/conducting-research-tribal-communities.
A nutritionally complete diet can be constructed that shows the grams per day of each food
consumed by determining the representative foods in each staple category (e.g., for anadromous and
marine fish it may be salmon, oysters, and herring and for game it may be deer, rabbit, and beaver)
and assigning caloric counts to each (Kcal/g) from the USDA National Nutrient Database for
Standard Reference (USDA, 2010). For example, if it was determined that anadromous and marine
fish comprised 30 percent of the total caloric intake, and the average Kcal/g given the species mix
was 1.55 Kcal/g, we can estimate that to reach a daily calorie goal of 2,000 Kcal, a typical individual
would have consumed 600 Kcal (30% of 2,000 Kcal) of anadromous and marine fish each day,
which is 387 g/day (600 Kcal divided by 1.55 Kcal per g). It is important for researchers to clearly
80
-------
document each calculation and estimate at each step, providing the assumptions and documenting
decisions made.
To obtain information on the uncertainty of an estimated heritage rate, the researcher can vary the
assumptions and develop rates under each set of assumptions. While this range is not as formal as a
confidence interval, it will give the user some feeling for the uncertainty in the rates.
Unlike the methodology for a contemporary survey of current consumption, the heritage rate
estimates may depend on professional judgment. That judgment may be informed not only by the
literature, but also by site visits to the population and use of historic ecological data. Heritage rates
should not be discarded simply because professional judgment or a reasonable assumption has been
used in the calculations. The user of the rates takes into consideration the underlying basis of the
heritage rate.
The methods described in this section provide a peer-reviewed approach (Harper et al., 2012) for
estimating an unsuppressed FCR. Additionally, examples of studies include the following (posted at
http:/ /superfund.oregonstate.edu/conducting-research-tribal-communities and
https:/ /www.deq.idaho.gov/media/60177351 /58-0102-1201-fish-consumption-rates-kootenai-
tribe.pdf):
¦ Confederated Tribes of the Umatilla Indian Reservation (CTUIR) (Harris & Harper,
2004)
¦ The Spokane Tribe's Multipathway Subsistence Exposure Scenario and Screening Level
RME (Harper et al., 2002)
¦ Washoe Tribe Human Health Risk Assessment Exposure Scenario for the Leviathan
Mine Superfund Site (Harper, 2005)
¦ Quapaw Traditional Lifeways Scenario (Harper, 2008)
¦ Wabanaki Traditional Cultural Lifeways Exposure Scenario (Harper & Ranco, 2009)
¦ Heritage Fish Consumption Rates of the Kootenai Tribe of Idaho (Kootenai Tribe of
Idaho, 2014)
Although this methodology provides a reasonable estimate of an unsuppressed rate, it cannot be
assumed that this approach could be reproduced for all populations. This depends on the types of
historical records and/or data that are available for the population.
6.2 Alternative Methodologies to Measure Suppression
In some cases, a similar population may be divided among two areas, one contaminated and one not.
To the extent that one area can be a control site for another area, those two populations might be
compared, assuming that the populations are truly similar. In other cases, a single population might
have had data collected before a significant contamination event, in which case the population serves
as its own before- and after- control. These types of studies are typical epidemiological study
designs, and are described in environmental epidemiology texts along with ways to control for
confounding factors.
81
-------
If they are available, fish consumption rate data from areas with similar fish and human populations
to the area of interest that are less impacted by fish advisories or fears of chemical contamination
can be used. This method measures suppression by comparing one contemporary site to another
control site. This is not the same as suppression in fish consumption from a traditional or heritage
rate. The control site is a baseline with respect to the general population, but is not a baseline for a
tribal heritage rate. A key factor in this decision is the habitat quality associated with areas from
which the population harvests fish. This approach has been employed in examining chemical
contaminant risks for American Indians/Alaska Natives consuming seafood in the vicinity of Puget
Sound Superfund sites (U.S. EPA, 2007). The suppression effect due to the fish advisories or the
fears of chemical contamination could then be estimated by taking the difference between the two
rates. Note that the resulting estimate of the suppression effect would not account for any
suppression in the less impacted area (unrelated to fish advisories or chemical contamination), such
as suppression from a heritage rate.
A similar methodology to assess suppression due to changing fish availability could be investigated
by looking at the relationship between fish populations and fish consumption rates. For example, in
areas with plentiful fish, what are the FCRs for the population and how do those FCRs compare to
FCRs in areas with fewer fish? Investigating these differences could provide some insight into the
magnitude of suppression due to changing fish availability, but will not measure any suppression
from the heritage rate.
Discrete choice modeling, using individual decision rules to formulate econometric models of
population choice (Greene, 2009; McFadden, 1974), could be applied to the problem of measuring
suppression. It has been used to measure the economic effects of fish consumption advisories
(McNair & Desvousges, 2007; Jakus et al., 1998; Jakus et al., 1997).
Another appropriate alternative is to include suppression questions in a fish consumption survey.
For example, a person might want to eat more fish, but they are not readily available and/or
expensive or a family might want to eat more fish if they had time to fish. These questions can be
designed to determine whether fish consumption may be lower than a population would normally
consume or wish to consume. Suppression questions can take several forms, but should be
developed in such a way so that the causes and reasons behind suppression can be identified. Since
the measure of suppression relies upon recall of past (in terms of years or decades), or even
historical, consumption patterns, studies often incorporate more qualitative or ethnographic
approaches.
Quantification of suppression can be very difficult using traditional survey methods, but could be
addressed at a qualitative level through use of well-structured questions. One principle in the design
of survey questions is to beware of hypothetical questions as people generally are unprepared to
predict future behavior or quantify past behavior, especially if it has been a generation or more since
suppression began. Estimating future behavior is difficult as there are often a number of influences
or situational factors that a hypothetical question is unable to account for. Another reason that
responses to hypothetical questions can be troublesome is that the question or questions are often
based on hypothetical changes in a state or behavior that the respondent must create or imagine.
These imagined states may be difficult for a question to fully describe (Fowler, 1995).
From a methodological perspective, it is important that good standard survey practices be employed
if these topics are important for a proposed survey. In particular, respondents may be asked, "If the
82
-------
fish were not contaminated, how much would you eat?" or "How much fish would you like to eat if
you could do so safely?" These types of questions are speculative and based on hypothetical
situations only. Estimates derived from such speculative questions may yield data that say little or
nothing about actual future behavior or about heritage cultural practices.
However, if a survey requires that questions about future behaviors be asked, it is generally
appropriate that the questions be based as much as possible on past experience or direct knowledge
by the respondent. This helps to provide some frame of reference for the respondent. For example,
if respondents are asked if they plan to continue current fish consumption levels in the future,
current fish consumption habits and experiences should be included in the questioning. This is in
contrast to simply asking respondents if their consumption habits would change based on
hypothetical changes in other factors (e.g., availability or safety). The respondents may have no past
experience on which to base their response. When this occurs, respondents are likely to make
cognitive shortcuts in the response process, rely on heuristics, or provide socially desirable answers.
Some approaches to asking about current behaviors/experiences as they may relate to future
activities include the following:
¦ Determine if the respondents have tried to eat fish but the fish they wanted was
unavailable
¦ Ask about the awareness and influence of fish consumption advisories on fish
consumption
¦ Ask about economic influences that may prevent them from eating fish (e.g., cost of
purchased fish relative to income)
¦ Ascertain the importance of fish as a dietary staple or the cultural significance of fish
consumption
¦ Ask about the current availability of the fish within the respondents' local area
¦ Ask whether a family or respondent's access to fishing locations has changed
¦ Ask whether laws or fishing regulations have altered the ability of the respondent to
obtain fish
¦ Ask whether changes in family duties related to working in the contemporary job
market have reduced the time family members have to engage in fishing activities
While responses to these questions will not be able to quantify a hypothetical change in a behavior,
they can indicate whether a change in any of these states or situations may be a factor in influencing
future behavior.
Another approach to capture suppressed or time-altered consumption is to ask respondents whether
their fish consumption or fishing activities have increased, decreased, or stayed the same in
comparison to a period X years ago (e.g., 10 years). Note that long-term suppression or
contamination means that the heritage, unsuppressed, or baseline rate cannot be ascertained by
asking these types of questions because they may not remember or have not studied the issue of
population-specific traditional rates. Such a question—with a fixed prior time point—cannot be
asked of the younger respondents. Only respondents who were teenagers or older at the prior time
point can give meaningful answers. For respondents who indicate a change in consumption, the
reason for the change can be captured. The FCST (the EPA developed Microsoft Access-based
software package) includes the question, "Sometimes for various reasons, people's consumption offish and
shellfish changes. Has the amount offish and/ or shellfish consumed by you or yourfamily changed over the last twenty
83
-------
years?" and the follow-up questions, "How has the amount offish and/ or shellfish you have eaten over the past
20years changed? (eat more now, eat less now, eat different types now) and Please tell me what you think has caused
the change in the amount or type offish and shellfish you now eat" These questions can be used to determine
if there is a suppression effect, gauge the possible magnitude of it, and find ways to adjust the FCR
that results from the survey for the suppression effect. For example, if respondents report that they
are consuming less fish now and the reasons provided are related to environmental contamination,
comparing the resulting FCR from the survey to a FCR from an uncontaminated area could provide
an adjustment factor. Another example would be if respondents reported that they eat less fish now
due to the availability of fish in lakes and streams for harvest, researching changes in fish
populations could provide an adjustment factor.
It is recognized that, in some cases, it may be necessary or possibly unavoidable to ask hypothetical
questions. When these situations arise, careful development of such questions is encouraged. The
area of survey methodology called "contingent behavior" is based on the concept that people are
capable of accurately predicting their future actions, given well-worded questions about specific
scenarios (Parsons et al., 2006).
Although quantitative data about future consumption often cannot be derived from the
questionnaires described in this guidance document, by asking people to predict their level of future
use under the change of a single condition (e.g., alleviation of their concerns about contamination), a
survey can provide useful information on the qualitative scale of change that usage rates are likely to
undergo as remediation and/or risk communication progresses.
84
-------
7
ANALYTIC APPROACHES
7.1 Introduction to Analysis
This section provides a general discussion on statistical analysis, and Section 7.2 focuses on the types
of data analysis available for estimating UFCR using the FCST described in Chapter 4, or other
similar dietary data collection tools (e.g., providing both 24-hour fish consumption recall data and
frequency of fish consumption data).
7.1.1 Weighting
Care must be taken when analyzing the data because the statistical methods appropriate for
calculating unbiased estimates of the population parameters will depend on the sampling method
(e.g., simple random sampling, stratified sampling, proportional stratified sampling). There are
important statistical issues to consider when making adjustments for the various types of sampling
inaccuracies. Weights might need to be applied during the estimation of population parameters
whereby the weights account for different sizes of subpopulations, for differential nonresponse
rates, or for disproportionate sample selection probabilities. For example, there might be cases
where probabilities of respondent selection become disproportionate in field implementation such
that the sample population disproportionately represents different demographic groups. In those
cases where probabilities change between the design and implementation stages, post-stratified
weights are used to estimate population parameters that are derived from a sample distribution that
does not correspond to the known population distribution. An experienced survey statistician
should be consulted to facilitate the appropriate weighted analysis and presentation of survey results.
7.1.2 Estimating Usual Fish Consumption Rates
Estimating the UFCR requires particular dietary data collection and analysis methodologies.
As of the writing of this document, the NCI Method is the most generally accepted method for
estimating usual intake. Implementing the NCI Method requires statistical knowledge and the use of
SAS software. Individuals experienced with using the NCI Method can be found in academia and
statistical consulting and survey research firms. Of the four methods mentioned in Section 1.2.3 for
estimating usual intakes using 24-hour recall data (the NCI Method, MSM, the Iowa State University
Method, and SPADE), the NCI Method is generally accepted as the method that produces the least
biased estimates. Souverein et al., 2011, provide a comparison of the methods. Also see Table 4-1.
The NCI Method (Tooze et al., 2010; Tooze et al., 2006) utilizes statistical modeling to estimate
usual intake of nutrients and foods, and it is especially appropriate for episodically consumed foods
like fish. It can be used to estimate the distribution of usual intake for a population or
subpopulation. The premise of the NCI method is that usual intake is equal to the probability of
consumption on a given day times the average amount consumed on a "consumption day." For
episodically consumed foods, such as fish, a two-part model is used to estimate usual intake. The
85
-------
first part uses logistic regression to estimate the probability of consumption with a person-specific
random effect. The second part estimates the consumption day amount using linear regression, also
with a person-specific effect. The reported consumption amounts are transformed to be roughly
normally distributed for linear regression. The regression predictions are then back-transformed to
estimate UFCR in the original units (see Tooze et al., 2006 for more details). The two parts are
linked by allowing the person-specific effects to be correlated and by including common predictors
in both parts of the model. The method assumes all subjects included in the analysis are consumers,
even if some are assigned (through modeling) a very small probability of consumption. Thus it
produces per capita estimates. Since the method provides long-term averages of consumption, even
subjects who report consuming fish once in a year should be included in the analysis. However, if a
survey is interested in consumer-only estimates, and wants to exclude those who never consume fish
(e.g., strict vegetarians and vegans), the survey should include an appropriate question to be able to
exclude those from the analysis.
Data from one or more non-consecutive recalls provide the values for the dependent variable. At
least a subset of the population, generally believed to be 50 respondents, needs to have consumption
data from two or more recalls. Predictors related to either the probability of consumption or
consumption amount, such as gender, age, race, and income, can be included in the modeling. If the
survey is interested in obtaining UFCRs for population subgroups, such as various age groups or by
gender or race, then, unless the criteria above applies to each subgroup, the subgroups of interest
need to be defined by covariates when implementing the NCI method. In most cases, the most
important predictor is a measure of frequency of consumption of the food of interest obtained from
a food frequency questionnaire. In addition, the model can incorporate the following within-person
predictors: 1) differences between weekends (Friday to Sunday) and weekdays (Monday to
Thursday), and 2) consistent differences between multiple 24-hour recalls (sequence effect) as
captured by an appropriate covariate. The resulting model parameters are then used to estimate
population and subpopulation distributions. Note that research has shown differences in weekend
consumption compared to weekday consumption and that Friday is more comparable to Saturday
and Sunday than to the rest of the week (Haines et al., 2003). Evidence for the validity of the
method has been published in a series of papers in the Journal of the American Dietetic Association and
The Journal of Nutrition. (Freedman et al., 2010; Kipnis et al., 2009; Tooze et al., 2006). The NCI
method can be implemented using two SAS programs (the MIXTRAN and DISTRIB macros)
available from the NCI web site
(http: / /appliedresearch.cancer.gov/diet/usualintakes /macros single.html).
While the MSM approximates the NCI Method and thus may not provide as accurate estimates of
usual fish consumption and does not provide standard errors of estimates, it is available interactively
through this website: https:/ /nugo.dife.de/msm. where users can interactively import data sets,
define analysis models, and review and export results and graphs. The use of the program is
supported by online help and a user guide, and communication between users and the program web
site is encrypted, securing transmitted data against unauthorized use (Harttig et al., 2011). Using this
method may reduce analysis costs and thus may be a good option for some users.
Utilizing a FFQ to estimate UFCR involves asking respondents how often they consume fish in a
given time period (i.e., past year, past 6 months) and asking how much they usually consume when
they consume it. To estimate the usual intake for each respondent, the amount reported (converted
to a substitute value if collected as a range) is multiplied by the frequency reported and then
converted to a per day amount. The mean and percentiles of usual intake can then be estimated
86
-------
from these individual rates. Note that this method results in more biased estimates compared to the
NCI Method as FFQs result in more measurement error and can be more prone to systematic error
than 24-hour recalls, which are used for the NCI Method (Kipnis et al., 2003). Note that research
concerning the bias in FFQs compared to that in the NCI Method have been conducted on total
diet FFQs and not fish-specific FFQs. If circumstances are such that a survey must rely on FFQ
data for estimating fish consumption rates, it is necessary to be aware of and consider adjusting for
the known bias. Further research on adjusting for bias is necessary.
In Chapter 4 we discussed dietary assessment methods to collect data for the purpose of estimating
UFCR. Section 7.3 describes how 24-hour recall data collected by the FCST or another instrument
should be organized, along with the ancillary data, in order to use the NCI Method for analysis. It
also walks through the application of both the MIXTRAN and DISTRIB macros. Section 7.4
describes how to analyze data collected through a FFQ, by the FCST, or by another instrument to
obtain estimates of UFCR.
7.1.3 Reporting of Usual Fish Consumption Rates
FCR can be reported as "as-consumed" weight or "raw" weight of the fish. There needs to be
consistency in how the data are collected in terms of cooked weight (or as-consumed) or raw weight.
Depending on the goals of the study, if cooked weights are reported and collected they may need to
be converted to raw weights or vice versa prior to analysis. If the as-consumed weight is collected
and the data of interest are actually raw weight, it is necessary to collect data on how the fish was
prepared and cooked and if the weight is for the whole fish or the edible portion only. Different
preparation and cooking methods reduce the moisture in the fish by different amounts. Generally,
dry heat (e.g., baking) reduces the weight of the fish by 25 percent, moist heat (e.g., steaming)
reduces the weight by 21 percent, and frying reduces the weight by 12 percent (U.S. EPA, 2014; U.S.
EPA, 2002). The USDA Food and Nutrient Database for Dietary Studies (FNDDS) (USDA, 2014)
can be used to determine average recipes for various foods (e.g., breaded and baked trout) to then
estimate the weight of the cooked fish in the reported as-consumed amount. If raw weight is the
measure of interest, this estimate would then need to be converted to raw weight using a moisture
loss conversion factor.
If the survey objectives included estimating UFCR for various sub-populations, then more than one
rate will need to be estimated, for example age-specific rates, rates by race/ethnicity, rates by report
of sport fishing, etc. If a survey covered a population with a wide-range of different lifestyles, it may
be important to estimate and present data by these subpopulations to ensure that subpopulation
rates are not masked.
Consumption data can be presented in several different ways. Consumption estimates can be given
as point estimates or as distributions illustrating the variability in the population. A point estimate is
a single value such as 50 g/day, whereas a distribution can be summarized by a measure of central
tendency (e.g., mean, median), a standard deviation, and a shape of the distribution curve (e.g.,
lognormal). For many risk assessments, risk estimates for individuals at both the central tendency
and high-end portions of the exposure distribution are made. The Ambient Water Quality Criteria
(AWQC) Human Health Methodology (U.S. EPA, 2000d) suggests that the arithmetic mean or 90th
and 95th percentiles of fish consumption distributions be used for developing AWQC. In
87
-------
characterizing reasonable maximum exposure, such as for Comprehensive Environmental Response,
Compensation, and Liability Act (CERCLA) sites, commonly known as Superfund sites, EPA
guidance for exposure assessment specifies that media intake (e.g., fish consumption) rates be set at
a high-end exposure level (around the 90th percentile of the media intake distribution) to ensure
adequate protection of potentially exposed individuals, lifestages, or populations (U.S. EPA, 2016;
U.S. EPA, 1992). Estimates should be presented with a measure of precision (e.g., a standard error
or 95 percent confidence interval). To preserve the maximum amount of flexibility for future uses of
the data, researchers may present consumption data as a distribution.
7.2 Uncertainty
The estimated FCR resulting from the survey and analysis may be uncertain due to either bias
(systematic error) or random variation. Bias results in a consistently high or consistently low fish
consumption rate relative to the true or desired value. Variation results in an uncertain fish
consumption rate that might be either higher or lower than the true value.
The primary sources of random variation are the following:
¦ Sampling error associated with the random selection of respondents. For example, if
different counties and individuals had been selected for the data collection, the data and
FCRs would be different.
¦ Random differences due to the simulation of usual fish consumption for each
respondent. This source of variation can be reduced by increasing the number of
simulations.
The confidence intervals for the fish consumption rates account for both of these sources of
variation. If there are fewer respondents with reported fish consumption in two (or more) 24-hour
recalls, there is less data to estimate the parameters and particularly the variance components,
resulting in more uncertainty in the fish consumption estimates and wider confidence intervals.
There are multiple sources of bias that can affect the fish consumption rates, including:
¦ Seasonality
¦ Respondent bias
¦ Use of standard recipes to calculate fish consumption amounts from the 24-hour recalls
¦ Bias associated with the estimation method (e.g., NCI Method, MSM) and its
assumptions
Each of these sources of bias is discussed in more detail in the following subsections.
88
-------
7.2.1 Seasonality
Fish consumption, especially of recreationally or sport-caught fish, is likely to vary by season. If the
data collection plan does not take into consideration seasonality, the estimates may overestimate or
underestimate usual intake. Refer to Section 5.5 for more information.
7.2.2 Reported Fish Consumption
The reported fish consumption may be biased if respondents tend to report consistently more or
less fish consumption in the 24-hour recall or FFQ than actually occurred. Assessing if the reported
values are biased requires comparing reported values to estimates obtained using other data
collection approaches, such as analysis of duplicate meals. Over the years, much research has gone
into assessing dietary intake. The OPEN study found that both 24-hour recalls and total diet FFQs
yield biased estimates; however, total diet FFQs are more prone to systematic error compared to 24-
hour recalls, and both are prone to random error. Further research is needed to analyze bias in 24-
hour recall vs. FFQ approaches for fish consumption surveys. Random error can be reduced by
greater sample sizes, while systematic error cannot be. For more information on measurement error,
see NCI's measurement error webinar series at http: / /epi.grants.cancer.gov/events /measurement-
error/.
7.2.3 Use of Standard Recipes
The FNDDS can be used to obtain the amount of fish in a recipe (see Section 7.1.3). It utilizes
standard recipes for foods such as breaded and fried trout. If the survey team decides to use this
database as opposed to having respondents supply specific recipes of the foods they consumed, the
resulting amounts of fish consumed will be biased. For example, the standard recipe for the food
"scallops and noodles with cheese sauce" is approximately 35 percent fish. However, the true recipe
for the food consumed by a respondent may have less fish or more fish than the standard recipe.
Additionally, there is uncertainty associated with the moisture loss values for processing and cooking
methods that will be applied to the amounts to determine raw weight. They are generally average
values of moisture loss given the various processing and cooking methods. If respondents cooked
their fish a bit longer, then the moisture loss would be a bit greater than average, and if they cooked
it a bit less, the moisture loss would be a bit less than average. Populations with unique food
preparation practices may wish to document the mass of fish in specific preparations and to
document the moisture loss associated with unique preparation practices.
7.2.4 Estimation of Usual Fish Consumption
Measurements of usual fish consumption are very difficult to obtain. Since usual fish consumption is
a long-term average, many 24-hour recalls over a long time are needed to approximate what "usual
intake" is trying to assess; therefore, survey teams should rely on a statistical model and associated
89
-------
assumptions to estimate usual intake. As a result, the estimates of usual fish consumption depend in
part on the statistical assumptions.
The statistical model makes certain assumptions, such as 24-hour recalls provide unbiased estimates
of fish consumption, all respondents are fish consumers (at least occasionally, unless they are
omitted from the analysis), and the distribution of fish consumption among those reporting
consumption in a 24-hour recall is normally distributed for some power transformation. The validity
of these assumptions can be discussed and to some extent evaluated using data.
The estimates of the frequency of fish consumption depend in part on how non-consumers (those
who never eat fish or don't eat fish for a long time) are treated. From two 24-hour recalls it is not
possible to separate true non-consumers from those who did not happen to report fish consumption
in either recall. A similar problem relates to consumption of small amounts of fish. Should a person
who never eats an identifiable piece of fish but uses a salad dressing with a small amount of fish in it
be considered a regular consumer of a very small amount or a non-consumer of fish? Whether a
meal is classified as having fish may depend on the procedures used to ask the questions and the
recipes used to estimate fish consumption. Having non-consumers in the data will lower the overall
probability of fish consumption (P) but increase the variance of the probability of fish consumption
among individuals. The resulting effect on the upper percentiles of the distribution is not clear. The
survey team needs to determine if and how they will identify non-consumers to be removed from
analysis if desired (e.g., an additional survey question).
The reported amount of fish consumption will vary from one 24-hour recall to another, in part
because the respondents may be poor at estimating the amount consumed and in part because the
consumption amounts are reported in rounded units, such as a cup or a pint, but not 1.267 cups.
The rounding adds some uncertainty to the estimates. The within-person variance component of the
statistical models accounts for uncertainty due to poor estimation by the respondent and rounding
that is part of the process. Because the definition of usual fish consumption does not include the
within-person variation, this source of error should contribute minimal bias to the estimates of usual
fish consumption.
The statistical models make some assumptions to simplify the computations, such as an assumption
that variance components are normally distributed, additive in the transformed scale, and linearly
correlated. The assumption that the person-specific random effect in the probability model is
normally distributed is difficult to test without many more 24-hour recalls for each person. The
assumption that the two variance components in the amount model are normally distributed is
generally consistent with the observation that the Box-Cox transformed consumption amounts are
roughly normally distributed. Nevertheless, other assumptions may imply a similar distribution for
the reported amounts while using a somewhat different assumption for the person-specific variance
component and thus somewhat different estimates of fish consumption. Because the estimated
parameters must be consistent with the reported data, the general center and spread of the predicted
distribution will be similar regardless of the distributional assumptions. Specific percentiles may be
either higher or lower using different assumptions or may be relatively insensitive to the
distributional assumptions. Although these assumptions are common in other statistical applications,
it is difficult to assess how the estimates might change using other assumptions.
If the model assumptions are accepted as reasonable, then the question is whether the estimates
from the model are biased. If the estimates are based on maximum likelihood (as in the NCI
90
-------
Method), this can produce biased estimates, particularly variance estimates, with small sample sizes.
Convergence theory says maximum likelihood is best with large sample sizes. Thus, if the sample
size is large enough, we expect the estimates to have relatively little bias compared to the size of the
confidence intervals.
The fish consumption estimates depend in part on the independent predictors used in the model.
When different predictors are used, the estimates change. It is impossible to know what the best set
of predictors is. A systematic approach can be used to selecting the independent predictors from the
available predictors in an effort to minimize any bias. The estimates have unknown bias due to the
decisions that are made.
7.3 Analyzing Recall Data
7.3.1 Preparing the Data
For purposes of this discussion, we will assume that a survey collected two 24-hour fish recalls at
least 3 weeks apart and that approximately 25 percent of the participants were contacted during the
spring, 25 percent during the summer, 25 percent during the fall, and 25 percent during the winter.
Additionally, we will assume that the survey is collecting data to estimate UFCR of fish from rivers
and lakes within a specific state, although there may be sub-objectives requiring data to be collected
on more than one species.
The fish consumption data collected through the use of the 24-hour fish recall need to be processed
and coded such that there are analysis variables consisting of the amount of fish of interest (in our
example, fish from rivers and lakes within the state) consumed for each day. Table 7-1 presents
example data for five participants. (Note that this is not an example of a data collection form; it is
only being used to display data abstracted from a 24-hour fish recall.)
Table 7-1. Example of 24-hour fish recall data, ounces
Participant
First 24-hour recall
Second 24-hour recall
Not local
y caught
Locally caught
Not locally caught
Locally caught
Tuna
Crab
Bass
Walleye
Tuna
Crab
Bass
Walleye
1
6.5
0
0
0
0
0
4.0
0
2
0
3.0
4.5
0
6.0
0
0
8.5
3
0
8.0
0
3.0
0
0
0
5.5
4
0
0
0
0
0
0
0
0
5
0
0
6.0
0
—
—
—
—
For the desired analysis objective (UFCR of locally caught fish), these data would then be processed
to sum up the ounces of reported locally caught fish for each participant-recall day, as shown in
Table 7-2.
91
-------
Table 7-2. Example of summary of 24-hour fish recall data to derive analysis variables,
ounces
Participant
First 24-hour recall
Second 24-hour recall
Locally caught
Locally caught
1
0
4.0
2
4.5
8.5
3
3.0
5.5
4
0
0
5
6.0
—
The general format of the analytic data set is one record per person-recall. Thus, for our example,
each participant will have two records in the data set, as shown in Table 7-3. Note that if some
participants did not provide a second recall (lost to follow-up, refused, etc.), as is the case with
participant 5 in our example, the data from their first contact can still be used in the analysis.
Ancillary data will be appended to the data set for each record.
Table 7-3. Example format of analytic data set
Participant
Recall
Day
Grams
Fish
Weekend
Season
Gender
Age
Body
weight
(lbs.)
Fish
Freq
Seq
1
1
0
0
Summer
M
35
197
12
0
1
2
113.4
1
Summer
M
35
197
12
1
2
1
127.6
0
Fall
F
22
150
6
0
2
2
241.0
1
Fall
F
22
150
6
1
3
1
85.0
0
Spring
F
51
168
9
0
3
2
155.9
1
Spring
F
51
168
9
1
4
1
0
1
Winter
M
72
164
3
0
4
2
0
0
Winter
M
72
164
3
1
5
1
170.1
0
Summer
F
46
133
8
0
The column headers represent the variables to be used in the analysis.
¦ "Recall Day" is the first or second 24-hour fish recall
¦ "Grams fish" is the number of ounces of fish reported consumed on each recall day
converted to grams (28.35 g = 1 ounce, 227 g = 8 ounces or V2 a pound)
¦ "Weekend" is an indicator variable that denotes whether the recall was for a weekend
day (Friday, Saturday, or Sunday) or a weekday (Monday, Tuesday, Wednesday, or
Thursday), with a "1" indicating weekend and a "0" indicating weekday
¦ "Season" is the season during which the recalls occurred
¦ "Gender" is the gender of the participant
¦ "Age" is the age of the participant
¦ "Body weight" is the reported weight of the participant (in this example it is reported in
pounds, but the pounds could later be converted to kilograms if necessary.)
¦ "Fish Freq" is the frequency that fish is consumed, as reported by the participant (This
could be the reported frequency for the season during which a participant is
92
-------
interviewed, or it could be the reported frequency for 30 days. It obviously depends on
how the questions are asked. Ideally, they will ask about the season during which the
interview is being conducted.)
¦ "Seq" is a variable used in the MIXTRAN macro that accounts for effects due to the
sequence number of a subject's records (Since our example data has two records per
person, we will need just one seq indicator variable, set to 0 for the person's first record
and set to 1 for the second record. If we collected three fish recalls per person, then two
indicator variables would be used to identify a subject's second and third records,
respectively.)
Note that other ancillary data could also be used (e.g., race and ethnicity).
7.3.2 Application of the NCI Method
The following is a brief discussion of the MIXTRAN and DISTRIB macros. Note that the use of
these macros requires knowledge of SAS software and a familiarity with nonlinear mixed modeling
for implementation. The macros and more detailed information, a user's guide, and examples can be
found on NCI's web site at: http: / /epi.grants.cancer.gov/diet/usualintakes /macros.html.
"The MIXTRAN macro is used for the analysis of episodically consumed
foods, foods consumed every day, and nutrients, and output from the
MIXTRAN macro is used by the DISTRIB macro for estimation of the
distribution of usual intake. For episodically consumed foods, the
MIXTRAN macro fits a two-part nonlinear mixed model where the first
part considers the probability of consumption and the second part
considers the consumption-day amount. The model allows for covariates
and includes a random effect in both parts and allows for correlation
between the random effects (Tooze et al., 2006). To fit this nonlinear mixed
model with correlated random effects (i.e., the correlated model), starting
values for the two parts of the model are obtained by first using the
GENMOD procedure to fit a probability model and an amount model.
Then a nonlinear mixed model with uncorrelated random effects (i.e., the
uncorrelated model) is fit using two calls to the NLMIXED procedure, and
the parameter estimates from this model are used as starting values for the
correlated model." (NCI Method, MIXTRAN macro, version 2.1)
"The DISTRIB macro uses results from the MIXTRAN macro and
estimates the distribution of usual intake for episodically consumed foods,
foods consumed every day, and nutrients (Tooze et al., 2006). The data can
then be used to calculate percentiles, and optionally, the percent meeting or
failing to meet the recommended daily intake for a population. The
DISTRIB macro contains two main functions. First, the DISTRIB macro
reads data sets of parameter estimates and predicted values output by the
MIXTRAN macro. Monte Carlo simulation of the random effect(s) is used
to estimate the distribution of usual intake. This data set can be saved.
Second, once the data containing the estimated usual intake are available,
93
-------
percentiles and outpoints can be calculated. The addition of a subgroup
variable is accommodated, so that statistics can be calculated by subgroup
and for the overall data set. Optionally the percent who meet
recommended daily intake values can be calculated. To accomplish this and
allow flexibility, the DISTRIB macro contains two sub-macros and some
general code to set up and call the macros as requested. The macro MC
uses Monte Carlo simulation of the random effect(s) to estimate the
distribution of usual intake. The output data set can be saved for future use.
The macro PC reads in the usual intake values calculated in the macro MC,
normalizes the weights, calculates the percentiles of usual intake, cutpoints
if requested, and optionally the percent meeting the recommended intake.
A single subgroup variable can be accommodated by the macro PC. The
resulting data set can be saved for future use." (NCI Method, DISTRIB
macro, version 2.1)
In some rare situations in which a study with weighted data and a large number of primary sampling
units is interested in estimating FCR for a number of different fish types (e.g., freshwater fish,
estuarine fish, marine, fish for specific trophic levels, and any combination of these) and a number
of different populations (e.g., women of childbearing age, children of various age groups, different
racial and ethnic groups, varying income levels, etc.), the NCI Method may be too time-consuming
(in terms of computer time) to be of practical use. Additionally, the NCI Method has a maximum
number of predictors that can be included in the model. Thus if a study finds that there are many
statistically significant interactions of important predictors, this may also preclude the use of the
NCI Method. Please see U.S. EPA, 2014, for further information and an alternative approach which
approximates the NCI Method. Additionally, the MSM, the Iowa State University Method, and
SPADE may also be considered.
7.4 Analyzing Frequency Data
For purposes of this discussion, assume that the survey used the FCST or a similar set of questions
as the data collection instrument to collect the frequency data. Appendix D provides an example
FCST hard-copy questionnaire. Parts two and three of the questionnaire represent the FFQ portions
of the instrument. Before analysis can begin, there are two conversions that need to be made.
First, the frequency questions to which participants provide the number of times they consume fish
(or a species of interest) over a time period need to be transformed to represent a consistent time
period. Converting all responses to days is preferable as it is likely that the survey's goal is to have
the final UFCR estimates in g/day. It is important to maintain full precision (no rounding) for
intermediate results. However, in the following tables, results have been rounded for presentation
purposes. The final UFCR estimates reported by the survey should be rounded. Table 7-4 provides
an example of converting reported frequency data to a consistent unit of time.
94
-------
Table 7-4. Example of conversions to consistent time periods
Reported Data
Conversion
Converted Data
Participant
# times
Unit of time
Factor1
# times
Unit of time
1
1
week
1/7
0.1429
day
2
1
day
1
1.0000
day
3
8
month
1/30.44
0.2628
day
4
6
year
1/365.24
0.0164
day
5
3
week
1/7
0.4286
day
6
3
season of year
1/91.32
0.0329
day
7
10
fish season2
1/63 (9 weeks)
0.1587
day
*An average year has 365.2425 days and an average month has 30.44 days (365.2425/12)
2The fish season is defined by the survey team. The species of the fish and the geographic
location are determining factors. For this example, we are assuming the species' season is 9
weeks.
Second, the usual portion size, or reported amount consumed per meal, needs to be converted to a
consistent unit of weight. Participants may report their portion size in ounces, pounds, or grams.
The desired unit should be determined based on the survey's goals. For example, if a survey wanted
to estimate UFCR in g/day, then each participant's usual portion needs to be converted to grams.
The FCST asks participants to report the usual portion size and provides pictures or portion size
models to assist participants in determining the portion size. Some instruments ask participants to
select a range that represents the usual portion size. In these situations, you will need to convert the
range to a point value. Generally, the mid-point of the range is used. Using narrower ranges on the
data collection instrument or using methods to elicit a more precise estimate of amount (such as in
the FCST) are preferred. Table 7-5 provides an example of converting reported portion size to a
consistent unit of weight.
Table 7-5. Example of conversions to consistent unit of weight
Reported Data
Converted Data
Conversion
Unit of
Participant
Size
Unit of weight
Factor
Size
weight
1
4
ounces
28.3495 g/oz
113.40
g
2
6.5
ounces
28.3495 g/oz
184.27
g
3
3
ounces
28.3495 g/oz
85.05
g
4
0.5
pounds
453.592 g/oz
226.80
g
5
2
ounces
28.3495 g/oz
56.70
g
6
1
ounces
28.3495 g/oz
28.35
g
7
0.75
pounds
453.592 g/oz
340.19
g
Once the frequency of consumption data and the usual portion size have been converted to
consistent units, they can be combined to get a UFCR for each individual. The frequency in days is
multiplied by the amount in g to get g/day. Table 7-6 presents an example.
95
-------
Table 7-6. Example of calculating UFCR for each individual
Frequency
Usual
Portion
Size
Participant
# times
Unit of time
g
UFCR, g/day
1
0.1429
day
113.40
16.20
2
1.0000
day
184.27
184.27
3
0.2628
day
85.05
22.68
4
0.0164
day
226.80
3.72
5
0.4286
day
56.70
24.30
6
0.0329
day
28.35
0.94
7
0.1587
day
340.19
53.99
The resulting individual estimates of UFCR can then be summarized to estimate the mean UFCR
and its standard error, and percentiles of UFCR and their standard errors for the survey population,
using the survey weights as discussed in Section 7.1.1. The data can also be summarized by
population characteristics (e.g., gender, age) to estimate UFCR for sub-populations of interest.
The FCST asks participants to provide frequency of consumption and usual portion size for two
time periods, when the fish is in season and when it is not in season. This allows for estimates of
UFCR to be made for these two time periods independently or to combine them to get an annual
UFCR. If the survey's goal is to estimate an annual UFCR, then the in-season UFCR would be pro-
rated for only the number of days in the season (e.g., 63 days if the season is 9 weeks long) and the
out-of-season UFCR would be prorated for the remaining days in the year (302 days). Table 7-7
provides an example.
Table 7-7. Example of estimating annual UFCR with seasonal data
In
Out-of-
UFCR, g/day
Season1
Season
Column A
Column B
(Column C +
UFCR
UFCR
multiplied by
multiplied by
Column D,
Participant
(g/day)
(g/day)
631days
3021days
divided by 365
A
B
C
D
days/year)
1
16.20
3.78
1020.60
1141.56
5.92
2
184.27
24.31
11609.01
7341.62
51.92
3
22.68
2.83
1428.84
854.66
6.26
42
3.72
3.72
234.36
1123.44
3.72
5
24.30
8.10
1530.90
2446.20
10.90
62
0.94
0.94
59.22
283.88
0.94
7
53.99
22.67
3401.37
6846.34
28.08
^he fish season is defined by the survey team. The species of the fish and the geographic location
are determining factors. For this example, we are assuming the species' season is 9 weeks or 63
days.
2Assuming participants 4 and 6 reported no difference in frequency or amount consumed by
season.
96
-------
8
SUMMARY AND APPENDIX OVERVIEW
The main body of this document presents a variety of survey approaches and other considerations
for conducting a fish consumption survey. Tables 3-5 through 3-7 present a graphical overview of
the various types of sampling frames and modes that can be used, including the strengths and
weaknesses of each approach. Choosing an appropriate approach must take into consideration (1)
the objectives of the survey, (2) the population being surveyed, and (3) the resources available for
the survey. In addition to selecting a sampling frame and mode to satisfy the goals of the planned
research, a variety of factors such as characteristics of the population (e.g., literacy rates) and desired
outcomes of the survey (e.g., response rates) also factor into the decision. During the planning of
the survey, the trade-offs between data desires, data needs, data quality, survey length,
representativeness, survey cost, and other factors must be considered. Usually, one or more of these
factors will limit and/or more clearly define what the survey can expect to accomplish.
Understanding these factors and their impact on the feasibility of an approach can assist in the
selection of the best survey approach to meet research needs.
This document provides guidance on dietary collection methodologies using the approaches
summarized in Tables 3-5 through 3-7. Specifically, the reader is provided with a description of an
existing instrument, EPA's FCST that can be used to collect data on fish consumption. Various
factors that are relevant for instrument development, such as respondent recall periods, use of visual
aids, portion sizes, etc., are also discussed. The use of qualitative methods and other pilot testing that
can be used to finalize a survey instrument are described.
For some populations or data uses, a typical survey approach may not be appropriate due to
suppression of fish consumption. For this reason, this document provides guidance on estimating
the historic or heritage FCR by recreating the historic, ecological conditions through the use of
historical records and documents and applying knowledge of nutrition, natural resource use, and
activity levels. In many cases, heritage rates may be the only practical way to estimate unsuppressed
rates — that is, free from the biasing influence of suppression effects. While this approach could be
particularly useful for tribal fishing populations, it may not be feasible for other populations or
situations where there is a lack of necessary historical information or other data. Therefore,
alternative methodologies for establishing unsuppressed rates are presented.
To support the analysis of collected survey data, Table 4-1 presents the data needs, advantages, and
limitations of various analytical methodologies for estimating a usual fish consumption rate (UFCR).
In the following appendices (A-D), there are useful tools, materials, and information to assist the
reader with the application of guidance presented in the main body of this document. Appendix A
includes a discussion of additional considerations for instrument development to help the researcher
develop questions that are methodologically sound, and provides example questions for use in
existing surveys. Appendix B presents background on a variety of generally accepted dietary
assessment methods. Appendix C provides details on the application of the NCI Method, and
Appendix D provides documentation for the publicly available, automated FCST. This includes a
hard-copy version of the instrument, guidance for developing a booklet of photographs to assist
respondents in accurately identifying fish species and portion sizes, a supervisor's guide to setting up
and configuring the automated tool, and an interviewer's guide.
97
-------
The appendices, in combination with the main body of this guidance document, provide the reader
with an overall strong foundation from which to begin the process of designing, executing, and
analyzing a fish consumption survey tailored to meet specific research needs.
98
-------
9
REFERENCES
Ahmed, F.E., Hattis, D., Wolke, R.E., and Steinman, D. (1993). Human health risks due to
consumption of chemically contaminated fishery products. 'EnvironmentalHealth Perspectives,
101(Suppl. 3): 297-302.
Andrews, W.R., Brick, J.M., Mathiowetz, N., and Stokes, L. (2010). Pilot test of a dualframe two-phase
mail survey of anglers in North Carolina. Final report. Prepared for the National Marine Fisheries
Service, National Oceanic and Atmospheric Administration, November 2010.
Ashford, J.R., Jones, C.M., and Fegley, L. (2009). A license registry improves sampling efficiency for
a marine recreational survey. Transactions of the American Fisheries Society, 138(5), 984-989.
Beehler, G.P., McGuinness, B.M., and Vena, J.E. (2001). Polluted fish, sources of knowledge, and
the perception of risk: Contextualizing African American anglers' sport fishing practices.
Human Organisation, 60(3), 288-297.
Birch, R.J., Bigler, J., Rogers, J., Zhuang, Y., and Clickner, R.B. (2014). Trends in blood mercury
concentrations and fish consumption among U.S. women of reproductive age, NHANES,
1999-2010. Environmental Research. 131:431-8.
Blumberg, S.J. and Luke J.V. (2014). Wireless substitution: Early release of estimatesfrom the National Health
Interview Survey, January-June 2014. National Center for Health Statistics. Available at:
http://www.cdc.gov/nchs/nhis.htm [Accessed on December 14, 2014].
Bogen, K. (1996). The effect of questionnaire length on response rates: A review of the literature. In
Proceedings of the Section on Survey Research Methods (pp. 1020-1025).
Brick, J. M., Brick, P. D., Dipko, S., Presser, S., Tucker, C., & Yuan, Y. (2007). Cell phone survey
feasibility in the US: Sampling and calling cell numbers versus landline numbers. Public Opinion
Quarterly, 71(1), 23-39.
Brick, J.M., Andrews, W.R., Brick, P.D., King, H., Mathiowetz, N.A., and Stokes, L. (2012).
Methods for improving response rates in two-phase mail surveys. Survey Practice, 5(3).
Brick, J.M., and Williams, D. (2013). Explaining rising nonresponse rates in cross-sectional surveys.
The ANNALS of the American Academy of Political and Social Science, 645(1), 36-59.
Burger, J., Stephens, W.L., Boring, S., Kuklinski, M., Gibbons, J.W., Gochfeld, M. (1999). Factors in
exposure assessment: ethnic and socioeconomic difference in fishing and consumption of fish
caught along the Savannah River. Risk Analysis', 19:427—38.
Carriger, J.F. and Barron, M.G. (2011). Minimizing risks from spilled oil to ecosystem services using
influence diagrams: The Deepwater Horizon spill response. EnvironmentalScience <& Technology.
45(18):7631-9.
99
-------
Carroll, R.J., Midthune, D., Subar, A.F., Shumakovich, M., Freedman, L.S., Thompson, F.E., Kipnis,
V. (2012). Taking advantage of the strengths of 2 different dietary assessment instruments to
improve intake estimates for nutritional epidemiology. American Journal of epidemiology. 2012
Feb 15;175(4):340-7.
Connelly, N.A., Brown, T.L., and Decker, D.J. (2003). Factors affecting response rates to natural
resource-focused mail surveys: empirical evidence of declining rates over time. Society and
Natural Resources, 16(6), 541-549.
Connelly, N.A., Lauber, T.B., Niederdeppe, J., and Knuth, B.A. (2012). Factors affectingfish consumption
among licensed anglers living in the Great Lakes Region. HDRU Publ. No. 12-3. Dept. of Nat.
Resour., N.Y.S. Coll. Agric. and Life Sci., Cornell Univ., Ithaca, N.Y. 78 pp.
Connelly, N.A., Smith, K.K., Lauber, T.B., Niederdeppe, J., and Knuth, B.A. (2013). Factors affecting
fish consumption among new mothers living in Minnesota, Pennsylvania, and Wisconsin. Human
Dimensions Research Unit Series Publication No. 13-1. Dept. of Natural Resources, Cornell
University, Ithaca, NY.
Connor, G.S., Tremblay, M., Moher, D., and Gorber, B. (2007). A comparison of direct vs. self-
report measures for assessing height, weight, and body mass index: A systematic review.
Obesity Reviews. 8(4): 307-36.
CRITFC (Columbia River Inter-Tribal Fish Commission) (1994). A fish consumption survey of the
Umatilla, Nez Perce, Yakama, and Warm Springs Tribes of the Columbia River Basin.
CRITFC Technical Rport 94-3. Portland, Oregon.
Dekkers, A.L., Verkaik-Kloosterman, J., van Rossum, C.T., and Ocke, M.C. (2014). SPADE, a new
statistical program to estimate habitual dietary intake from multiple food sources and dietary
supplements. The Journal of Nutrition, Dec;144(12):2083-91.
de Leeuw, E.D. (2005). To mix or not to mix data collection modes in surveys. Journal of Official
Statistics, 21(2), 233.
de Leeuw, E.D., Callegaro, M., Hox, J., Korendijk, E., and Lensvelt-Mulders, G. (2007). The
Influence of Advance Letters on Response in Telephone Surveys: A Meta-Analysis. Public
Opinion Quarterly, 71(3), 413-443.
Dietay Assessment Primer. National Institutes of Health, National Cancer Institute (2014). Available at:
(http://dietassessmentprimer.cancer.gov/) [Accessed on November 12, 2014].
Dillman, D.A., Smyth, J.D., and Christian, L.M. (2009). Internet, mail, and mixed-mode surveys: The
tailored design method. Hoboken, NJ: Wiley.
Donatuto, J., and Harper, B.L. (2008). Issues in evaluating fish consumption rates for Native
American Tribes. Risk Analysis, 28(6), 1497-1506.
Ellis, J.H., Friedman, D.B., Puett, R., Scott, G.I., and Porter, D.E. (2014). A qualitative exploration
of fishing and fish consumption in the Gullah/Geechee culture. J Community Health. Dec;
39 (6): 1161 -70.
100
-------
Florida Fish and Wildlife Conservation Commission. (2014). Available at:
http: / / myfwc.com/research/ freshwater/ sport-fishes / management-resources / freshwater-
angler-survevs/ [Accessed on December 5, 2016].
Forsberg, N.D., Stone, D., Harding, A.K., Harper, B., Harris, S., Matzke, M.M., Cardenas, A.,
Waters, K.M., Anderson, K.A. (2012). Effect of Native American fish smoking methods on
dietary exposure to polycyclic aromatic hydrocarbons and possible risks to human health.
Journal of Agricultural and Food Chemistry, 60(27), 6899-6906.
Fowler, F.J. (1995). Improving survey questions: Design and evaluation (Applied Social Research Methods Series,
Volume 38). Thousand Oaks, CA: Sage Publications.
Freedman, L.S., Midthune, D., Carroll, R.J., Krebs-Smith, S., Subar, A.F., Troiano, R.P., Dodd, K.,
Schatzkin, A., Ferrari, P., and Kipnis, Y. (2004). Adjustments to improve the estimation of
usual dietary intake distributions in the population. Journal of Nutrition.l34:1836-1843.
Freedman, L.S., Guenther, P.M., Dodd, K.W., Krebs-Smith, S.M., Midthune, D. (2010). The
population distribution of ratios of usual intakes of dietary components that are consumed
every day can be estimated from repeated 24-hour recalls. The Journal of Nutrition, Jan;
140(1): 111 -6.
Gaziano, C. (2005). Comparative analysis of within-household respondent selection techniques.
Public Opinion Quarterly, 69(1), 124-157.
Gersovitz, M., Madden, J.P., Smiciklas-Wright, H. (1978). Validity of the 24-hour dietary recall and
seven-day record for group comparisons. Journal of the American Dietetic Association, 73, 48-55.
Goldstein, K.M., and Jennings, M.K. (2002). The effect of advance letters on cooperation in a list
sample telephone survey. Public Opinion Quarterly, 66(4), 608-617.
Greene, W. (2009). Discrete Choice Modeling. In Palgrave Handbook of Econometrics: Vol 2, Applied
Econometrics. T. Mills and K. Patterson (Eds.), London, UK: Palgrave Macmillan.
Groves, R.M. (2006). Nonresponse rates and nonresponse bias in household surveys. Public Opinion
Quarterly, 70(5), 646-675.
Groves, R.M., Cialdini, R.B., and Couper, M.P. (1992). Understanding the decision to participate in a
survey. Public Opinion Quarterly, 56(4), 475-495.
Haines, P.S., Hama, M.Y., Guilkey, D.K., and Popkin, B.M. (2003). Weekend eating in the United
States is linked with greater energy, fat, and alcohol intake. Obesity Research, 11: 945—949.
Hansen, K.M. (2007). The effects of incentives, interview length, and interviewer characteristics on
response rates in a CATI-study. International Journal of Public Opinion Research, 19(1), 112-121.
Harding, A., Harper, B., Stone, D., O'Neill, C., Berger, P., Harris, S., and Donatuto, J. (2012).
Conducting research with tribal communities: sovereignty, ethics, and data-sharing issues.
Environmental Health Perpectives, 120(1), 6.
101
-------
Harkness, J.A., Villar, A., and Edwards, B. (2010). Translation, Adaptation, and Design. In Survey
methods in multinational, multiregional, and multicultural contexts. Harkness, J.A., Braun, M.,
Edwards, B., Johnson, T.P., Lyber, L., Mohler, P., Pennell, B. E., and Smith, T.W. (Eds.),
Hoboken, NJ: John Wiley and Sons.
Harper, B., Flett, B., Harris, S., Abeyta, C., Kirschner, F. (2002). The Spokane Tribe's multipathway
subsistence exposure scenario and screening level RME. Risk Analysis, 22(3):513-26.
Harper, B. (2005). Washoe Tribe Human Health Risk Assessment Exposure Scenario for the Leviathan Mine
Supefund Site. Available at:
http: / /health.oregonstate.edu/ sites / default/files / research/pdf/tribal-grant/Washoe-HHRA-
Exposure-Scenario-FINAL-3-17-05.pdf [Accessed on 16 April 2015].
Harper, B.L., Harding, A.K., Waterhous, T., Harris, S.G. (2007). Traditional Tribal Subsistence
Exposure Scenario and Risk Assessment Guidance Manual. Corvallis: Oregon State
University. Available at: http://superfund.oregonstate.edu/conducting-research-tribal-
communities.
Harper, B. (2008). Quapaw TraditionalUfeways Exposure. AESE, Inc.; Richland, WA, USA: 2008.
Available at:
http://superfund.oregonstate.edu/sites/default/files/harper 2008 quapaw scenario final.pd
f [Accessed on 16 April 2015].
Harper, B., and Ranco, D. (2009). Wabanaki traditional cultural Ufeways exposure scenario. Available at:
http://superfund.oregonstate.edu/sites/superfund.oregonstate.edu/files/harper ranco 2009
wabanaki scenario final.pdf [Accessed on 3 December 2014].
Harper, B., Harding, A.K., Harris, S., Berger, P. (2012). Subsistence exposure scenarios for tribal
applications. Human and Ecological Risk Assessment: An InternationalJournal, 18(4):810-831.
Harper, B., Walker, D.E. (2015a). Comparison of Contemporary and Heritage Fish Consumption
Rates in the Columbia River Basin. Human Ecology: Volume 43, Issue 2. Page 225-236. DOI:
10.1007/sl0745-015-9734-4
Harper, B. and Walker, D.E. (2015b). Columbia Basin Heritage Fish Consumption Rates. Human
Ecology: Volume 43, Issue 2. Page 237-245. DOI: 10.1007/sl0745-015-9735-3
Harris, S.G. and Harper, B.L. (2004) Exposure Scenario for CTUIR Traditional Subsistence Ufeways.
Department of Science & Engineering, Confederated Tribes of the Umatilla Indian
Reservation, P.O. Box 638, Pendleton, OR 97801. Available at:
http: / /health.oregonstate.edu/ sites / default/files / research/pdf/tribal-grant/CTUIR-
SCENARIO.pdf [Accessed on 16 April 2015].
Harttig, U., Haubrock, J., Kniippel, S., Boeing, H.; EFCOVAL Consortium (2011). The MSM
program: Web-based statistics package for estimating usual dietary intake using the Multiple
Source Method. Eurcpean Journal of ClinicalNutrition, 2011 Jul; 65 Suppl 1:S87-91.
Hill, D.H., and Willis, R.J. (2001). Reducing panel attrition: A search for effective policy instruments.
Journal of Human Resources, 36(3), 416-438.
102
-------
Holbrook, A.L., Green, M.C., and Krosnick, J.A. (2003). Telephone versus face-to-face interviewing
of national probability samples with long questionnaires: Comparisons of respondent
satisficing and social desirability response bias. Public Opinion Quarterly, 67(1), 79-125.
Holt, D. and Smith, T.M.F. (1979). Post stratification. Journal of the Royal Statistical Society, Series A,
142, Part 1, 33-46.
Idler, D.R., and W.A. Clemens. (1959). The Energy Expenditures ofFraser River Sockeye Salmon During the
Spawning Migration to Chilko and Stuart Lakes. Intern. Pacific Salmon Fish. Comm., Prog. Rept.
No. 6, New Westminster, B.C.).
Imm P., Knobeloch, L., Anderson, H.A., and the Great Lakes Sport Fish Consortium. (2005). Fish
consumption and advisory awareness in the Great Lakes Basin. Environmental Health Perpectives,
113(10): 1325-1329.
Israel, B.A., Schulz, A.J., Parker, E.A., Becker, A.B. (1998). Review of community-based research:
assessing partnership approaches to improve public health. Annual Review of Public Health.
19:173-202.
Israel, B.A., Eng, E, Schulz, A.J., Parker, E.A., editors. (2012). Methods in Community-BasedParticipatoy
Research for Health, (2nd Edition). San Francisco, CA: Jossey-Bass.
Jakus, P. M., Downing, M., Bevelhimer, M.S., and Fly, J.M. (1997). Do Sportfish Consumption
Advisories Affect Reservoir Anglers' Site Choice? Agricultural and Resource Economics Review, 26
(2): 198-204.
Jakus, P. M., Dadakas, D., and Fly, J.M.. (1998). Fish Consumption Advisories: Incorporating
Angler-Specific Knowledge, Habits, and Catch Rates in a Site Choice Model. American Journal
of Agricultural Economics, 80 (5): 1019—24.
Judd, N.L., Griffith, W.C., Faustman, E.M. (2004). Consideration of cultural and lifestyle factors in
defining susceptible populations for environmental disease. Toxicology, 198(1-3): 121 -33.
Katner, A., Ogunyinka, E., Sun, M.H., Soileau, S., Lavergne, D., Dugas, D., and Suffet, M. (2011).
Fishing, fish consumption and advisory awareness among Louisiana's recreational fishers.
Environmental Research, 111(8), 1037-1045.
Kinnell, J.C., Bingham, M.F., Hastings, E.A., Ray, R., Craven, V., and Freeman, M. (2007). A survey
methodology for collecting fish consumption data in urban and industrial water bodies (part
1). Journal of Toxicology and Environmental Health, Part A, 70(6), 477-495.
Kipnis, V., Subar, A.F., Midthune, D., Freedman, L.S., Ballard-Barbash, R., Troiano, R.P., Bingham,
S., Schoeller, D.A., Schatzkin, A., and Carroll, R.J. (2003). Structure of dietary measurement
error: Results of the OPEN biomarker study. American Journal of Epidemiology. 158:14-21.
Kipnis, V., Midthune, D., Buckman, D.W., Dodd, K.W., Guenther, P.M., Krebs-Smith, S.M., Subar,
A.F., Tooze, J.A., Carroll, R.J., Freedman, L.S. (2009). Modeling data with excess zeros and
measurement error: application to evaluating relationships between episodically consumed
foods and health outcomes. Biometrics, Dec; 65(4):1003-10.
103
-------
Kissinger, L., Lorenzana, R., Mittl, B., Lasrado, M., Iwenofu, S., Olivo, V., Helba, C., Capoeman, P.,
Williams, A.H. (2010). Development of a computer-assisted personal interview software
system for collection of tribal fish consumption data. Risk Analysis. Dec; 30(12):1833-41. doi:
10.1 lll/j.1539-6924.2010.01461.x.
Kootenai Tribe of Idaho. (2014). Heritage Fish Consumption Rates of the Kootenai Tribe of Idaho.
Available at: https://www.deq.idaho.gov/media/60177351 /58-0102-1201-fish-consumption-
rates-kootenai-tribe.pdf.
Kroke, A., Klipstein-Grobusch, K., Voss, S., Moseneder, J., Thielecke, F., Noack, R., and Boeing, H.
(1999). Validation of a self-administered food-frequency questionnaire administered in the
European Prospective Investigation into Cancer and Nutrition (EPIC) Study: comparison of
energy, protein, and macronutrient intakes estimated with the doubly labeled water, urinary
nitrogen, and repeated 24-h dietary recall methods. American Journal of Clinical Nutrition,
Oct;70(4):439-47.
Krosnick, J.A. (1991). Response strategies for coping with the cognitive demands of attitude
measures in surveys. Applied Cognitive Psychology, 5, 213-236.
Link, M.W., and Mokdad, A. (2005). Advance letters as a means of improving respondent
cooperation in random digit dialing studies: A multistate experiment. Public Opinion Quarterly,
69(4), 572-587.
Link, M. W., Battaglia, M. P., Frankel, M. R., Osborn, L., & Mokdad, A. H. (2007). Reaching the US
cell phone generation comparison of cell phone survey results with an ongoing landline
telephone survey. Public Opinion Quarterly, 71(5), 814-839.
Lohr, S. (2010). Sampling. Design and analysis, (2nd edition). Boston, MA: Brooks/Cole, Cengage
Learning.
MacNair, D.J. and Desvousges, W.H. (2007). The Economics of Fish Consumption Advisories:
Insights from Revealed and Stated Preference Data. Land Economics, 83 (4): 600—616.
Mahaffey, K.R., Clickner, R.P., Jeffries, R.A. (2009). Adult women's blood mercury concentrations
vary regionally in the United States: Association with patterns of fish consumption (NHANES
1999-2004). Environmental Health Perpectives, 117(1), 47-53.
McFadden, D. (1974). Conditional logit analysis of qualitative choice behavior. In Frontiers in
Econometrics. P. Zarembka (ed.), New York, NY: Academic Press.
Mercer, A., Caporaso, A., Cantor, D., Townsend, R. (2015). How Much Gets You How Much?
Monetary Incentives and Response Rates in Household Surveys. Public Opinion Quarterly, 79(1),
Spring 2015, 105-129.
Merrill, R.M. and Richardson, J.S. (2009). Validity of self-reported height, weight, and body mass
index: Findings from the National Health and Nutrition Examination Survey, 2001-2006.
Preventing Chronic Disease, 6(4). Available at:
http://www.cdc.gov/pcd/issues/2009/oct/08 0229.htm [Accessed on December 18, 2014].
104
-------
Michigan Department of Community Health. (2007). California Recreational Fisheries Survey.
Available at:
http://www.michigan.gov/documents/mdch/FCS Final rpt 061407 199288 7.pdf
[Accessed on January 15, 2015].
Michigan Department of Natural Resources. (2014). Available at:
http://www.michigan.gOv/dnr/0.4570.7-153-10364 52259 10951 11301-97829--.00.html.
Millar, M.M., and Dillman, D.A. (2011). Improving response to web and mixed-mode surveys. Public
Opinion Quarterly, 75(2), 249-269.
Minelli, C. and Baio, G. (2015). Value of Information: A Tool to Improve Research Prioritization
and Reduce Waste. PLoS Medicine, 12(9).
Minkler. M, and Wallerstein, N., editors. (2003). Community-Based Participatoy Research for Health. San
Francisco: Jossey-Bass Publishers.
Minnesota Department of Health. (2012). Fish consumption andfish advisory awareness amongMinnesota
women who recently gave birth. Minnesota Department of Health Environmental Division,
Environmental Surveillance and Assessment Section, Site Assessment and Consultation Unit.
Available at: http: / /www.health.state.mn.us /divs /eh/fish/techinfo /momsurvevl .pdf.
Minnesota Department of Natural Resources. (2011). Available at:
http: / /www.dnr.state.mn.us / areas / fisheries / montrose / creelsurvev.html.
National Cancer Institute, Register of Validated Short Dietary Assessment Instruments. (2013).
Applied Research Program, National Cancer Institute. Available at:
http:/ /appliedresearch.cancer.gov/diet/shortreg/.
National Center for Education Statistics. (2003). U.S. Department of Education. National assessment of
adultliteray. Available at: http://nces.ed.gov/naal/index.asp. [Accessed on December 3,
2014].
National Environmental Justice Advisory Council (NEJAC). Fish Consumption and Environmental
Justice; 2002. Available at:
https:/ /www3.epa.gov/environmentaljustice/resources /publications /nejac/fish-consump-
report 1102.pdf.
National Research Council. (2000). Committee on the toxicological effects of methylmercuy. Washington, DC:
National Academies Press.
New Jersey Department of Environmental Protection. (2014). Division of Fish and Wildlife.
Recreational Saltwater Angler Survey. Available at:
http://www.njfishandwildlife.com/marinesurvey.htm. [Accessed on December 18, 2014].
Nusser, S.M., Carriquiry, A.L., Dodd, K.W., and Fuller, W.A. (1996). A semi-parametric
transformation approach to estimating usual daily intake distributions. Journal of the American
Statistical Association, 91, 1440-1449.
105
-------
Office of Management and Budget 2006. "Standards and Guidelines for Statistical Surveys."
Available at:
http://www.whitehouse.gov/sites/default/files/omb/inforeg/statpolicy/standards stat surv
evs.pdf.
Oh, H. and Scheuren F. (1983). Weighting adjustments for unit nonresponse. In Incomplete Data in
Sample Surveys, Volume 2: Theory and Bibliographies. W. Madow, I. Olkin and D. Rubin. (Eds.),
143-184. New York: Academic Press.
O'Neill, C. Fishable waters, /. American Indian Taw Journal, 181 (2013). Available at:
http://papers.ssrn.com/sol3/papers.cfm?abstract id=2263490.
Oregon Department of Fish and Wildlife. (2014). Biometrics & Angler Creel Surveys. Available at:
https://nrimp.dfw.state.or.us/crl/default.aspx?pn=BACS. [Accessed on December 18, 2014].
Parsons, G.R., Morgan, A., Whitehead, J.C., and Haab, T.C. (2006). The welfare effects of pfiesteria-
related fish kills: A contingent behavior analysis of seafood consumers. Agricultural and Resource
Economics Review, 35.2: 349-356.
Pew Internet and American Life Project. (2013).
Ray, R., Craven, V., Kinnell, J., Bingham, M., Freeman, M., and Finley, B. (2007a). A statistical
method for analyzing data collected by a creel/angler survey (part 2). Journal of Toxicology and
Environmental Health. Part A. Mar 15; 70(6):496-511.
Ray, R., Craven, V., Bingham, M., Kinnell, J., Hastings, E., and Finley, B. (2007b). Human health
exposure factor estimates based upon a creel/angler survey of the lower Passaic River (part 3).
Journal of Toxicology and Environmental Health. Part A. Mar 15; 70(6):512-28.
Razzaghi, H., and Tinker, S.C. (2014). Seafood consumption among pregnant and non-pregnant
women of childbearing age in the United States, NHANES 1999-2006. Food 'Nutrition
Research. Jun 11;58.
Salama, A.A., Mohamed, M.A.M., Duval, B., Potter, T.L., and Levin, R.E. (1998). Polychlorinated
biphenyl concentration in raw and cooked north Atlantic bluefish fillets. Journal of Agricultural
andFood Chemistry, 46:1359—1362.
Skea, J.C., Simonin, H.A., Harris, E.J., Jackling, S., Spagnoli, J.J., Symula, J., etal. (1979). Reducing
levels of mirex, Aroclor 1254, and DDE by trimming and cooking Lake Ontario brown trout
and smallmouth bass. Journal of Great Lakes Research, 5(2):153—159.
Soon, R., Dye, T.D., Ralston, N.V., Berry, M.J, and Sauvage, L.M. (2014). Seafood consumption and
umbilical cord blood mercury concentrations in a multiethnic maternal and child health
cohort. BMC Pregnany Childbirth. Jun 18;14:209.
Souverein, O.W., Dekkers, A.L., Geelen, A., Haubrock, J., de Vries, J.H., Ocke, M.C., Harttig, U.,
Boeing, H., van't Veer, P. (2011). Comparing four methods to estimate usual intake
distributions. European Journal of Clinical Nutrition, Jul;65 Suppl 1:S92-101.
106
-------
Stahl, L.L., Snyder, B.D., Olsen, A.R., and Walters, L.S. (2013). A national probabilistic study of
polybrominated diphenyl ethers in fish from US lakes and reservoirs. Environmental Monitoring
and Assessment, 185:10351-10364.
Stahl, L.L., Snyder, B.D., Olsen, A.R., Kincaid, T.M., Wathen, J.B., and McCarty, H.B. (2014).
Perfluorinated compounds in fish from U.S. urban rivers and the Great Lakes. Science of the
Total Environment, 499:185-195.
Subar, A.F., Thompson, F.E., Kipnis, V., Midthune, D., Hurwitz, P., McNutt, S., Mcintosh, A., and
Rosenfeld, S. (2001). Comparative validation of the Block, Willett, and National Cancer
Institute food frequency questionnaires: the Eating at America's Table Study. American Journal
of Epidemiology, 154:1089-99.
Subar, A.F., Kipnis, V., Troiano, R.P., Midthune, D., Schoeller, D.A., Bingham, S., Sharbaugh, C.O.,
Trabulsi, J., Runswick, S., Ballard-Barbash, R., Sunshine, J., and Schatzkin, A. (2003). Using
intake biomarkers to evaluate the extent of dietary misreporting in a large sample of adults:
The OPEN study. American Journal of Epidemiology.
The Suquamish Tribe. (2000). Fish Consumption Survey of the Suquamish Indian Tribe of the Port
Madison Indian Reservation. Puget Sound Region. August 2000. Report writing group: M.
Duncan, N.L. Polissar, S. Liao, D. LaFlamme.
Tarrant, M.A., and Manfredo, M.J. (1993). Digit preference, recall bias, and nonresponse bias in self
reports of angling participation. Leisure Sciences, 15, 231—238.
Tennessee Wildlife Resources Agency. (2014). Available at:
http:/ /www.tnfish.org/TournamentFishingTennessee TWRA/TWRA BiteProgram.htm.
Accessed on December 18, 2014.
Texas Parks and Wildlife Department. (2014). Available at:
https://www.tpwd.state.tx.us/fishboat/fish/didvouknow/creel.phtml. Accessed on
December 18, 2014.
Thompson, F.E., Subar, A.F. (2008). Dietary assessment methodology. In Nutrition in the prevention
and treatment of disease. Coulston, A.M., Rock, C.L., Monsen, E.R. (Eds.), San Diego (CA):
Academic Press; 2008.
Thompson, M.R., and Boekelheide, K. (2013). Multiple environmental chemical exposures to lead,
mercury and polychlorinated biphenyls among childbearing-aged women (NHANES 1999-
2004): Body burden and risk factors. Environmental Research, 2013 Feb;121:23-30. doi:
10.1016/j.envres.2012.10.005. Epub 2012 Nov 16.
Tooze, J.A., Kipnis, V., Buckman, D.W., Carroll, R.J., Freedman, L.S., Guenther, P.M., Krebs-
Smith, S.M., Subar, A.F., Dodd, K.W. (2010). A mixed-effects model approach for estimating
the distribution of usual intake of nutrients: The NCI method. Statistics in Medicine, 2010 Nov
30;29(27):2857-68.
Tooze, J.A., Midthune, D., Dodd, K.W., Freedman, L.S., Krebs-Smith, S.M., Subar, A.F., Guenther,
P.M., Carroll, R.J., Kipnis, V. (2006). A new statistical method for estimating the usual intake
107
-------
of episodically consumed foods with application to their distribution. Journal of the American
Dietetic Association, Oct;106(10):l 575-87.
Tourangeau, R., Edwards, B., Johnson, T.P., Wolter, K.M., and Bates, N. (Eds.). (2014). Hard-to-
survey populations. Cambridge University Press, United Kingdom.
Trondsen, T., Braaten, T., Lund, E., and Eggen, A. E. (2004). Consumption of seafood—the
influence of overweight and health beliefs. Food Quality and Preference, 15(4), 361-374.
USDA. (2010). Nutrient Data Laboratory Home Page. Agricultural Research Service, U.S.
Department of Agriculture. Available at:
http://www.ars.usda.gov/main/site main.htm?modecode=12-35-45-00.
USDA. (2014). Food and Nutrient Database for Dietary Studies. Agricultural Research Service, U.S.
Department of Agriculture. Available at:
http://www.ars.usda.gov/services/docs.htm?docID=12089.
U.S. EPA. (1992). Guidelines for Exposure Assessment. EPA-600-Z-92-001. Washington, DC: U.S.
Environmental Protection Agency, Risk Assessment Forum. Available at:
https:/ / cfpub.epa.gov/ncea/risk/recordisplay.cfm?deid= 15263
U.S. EPA. (1994). Guidance for the Data Quality Objectives Process. EPA-600-R-96-055. Washington, DC:
U.S. Environmental Protection Agency, Office of Research and Development. Available at:
https: / /www3.epa.gov/epawaste/hazard/correctiveaction/resources / guidance/qa/epaqag4.p
df.
U.S. EPA. (1995). Guidance for Assessing Chemical Contaminant Data for Use in Fish Advisories. Volume IV.
Risk Communication. 3rd edition. EPA-823-R-95-001. Washington, DC: U.S. Environmental
Protection Agency, Office of Water.
U.S. EPA. (1998). Guidance for Conducting Fish and Wildlife Consumption Surveys. EPA-823-B-98-007.
Washington, DC: U.S. Environmental Protection Agency, Office of Water.
U.S. EPA. (1999). Asian & Pacific Islander Seafood Consumption Study in Yang County, WA.
EPA-910/R99-003. Washington, DC: U.S. Environmental Protection Agency, Office of
Water.
U.S. EPA. (2000a). Guidance for Assessing Chemical Contaminant Data for Use in Fish Advisories. Volume I.
Fish Sampling and Analysis. 3rd edition. EPA-823-B-00-007. Washington, DC: U.S.
Environmental Protection Agency, Office of Water.
U.S. EPA. (2000b). Guidance for Assessing Chemical Contaminant Data for Use in Fish Advisories. Volume II.
Risk Assessment and Fish Consumption Umits. 3rd edition. EPA-823-B-00-008. Washington, DC:
U.S. Environmental Protection Agency, Office of Water.
U.S. EPA. (2000c). Guidance for Assessing Chemical Contaminant Data for Use in Fish Advisories. Volume
III. Overview of Risk Management. 2nd edition. EPA-823-B-00-009. Washington, DC: U.S.
Environmental Protection Agency, Office of Water.
108
-------
U.S. EPA. (2000d). Methodology for Deriving Ambient Water Quality Criteria for the Protection of Human
Health (2000). EPA-822-B-00-004. Washington, DC: U.S. Environmental Protection Agency,
Office of Science and Technology, Office of Water.
U.S. EPA. (2002). Estimated per capita fish consumption in the United States. EPA-821- C- 02-003.
Washington, DC: U.S. Environmental Protection Agency, Office of Science and Technology,
Engineering and Analysis Division, Office of Water.
U.S. EPA. (2006). Guidance on Systematic Planning Using the Data Quality Objectives Process. EPA-240-B-
06-001. Washington, DC: U.S. Environmental Protection Agency, Office of Environmental
Information. Available at: https://www.epa.gov/quality/guidance-svstematic-planning-using-
data-quality-objectives-process-epa-qag-4.
U.S. EPA. (2007). Framework for S electing and Using Tribal Fish and Shellfish Consumption Rates for Risk-
Based Decision Making at CERCTA and RCRA Cleanup Sites in Puget Sound and the Strait of Georgia.
U.S. Environmental Protection Agency, Region 10, Office of Environmental Cleanup, Office
of Air, Waste and Toxics, Office of Environmental Assessment. Available at:
http://vosemite.epa.gov/rl0/CLEANUP.NSF/7780249be8f2SlS3882S6S0f0Q70bd8b/el291
8970debc8e488256da6005c428e/$FILE/Tribal%20Shellfish%20Framework.pdf.
U.S. EPA. (2009). The National Study of Chemical Residues in Take Fish Tissue. EPA-823-R-09-006.
Washington, DC: U.S. Environmental Protection Agency, Office of Water.
U.S. EPA. (2011). Exposure Factors Handbook: 2011 Edition. National Center for Environmental Assessment,
Washington, DC: U.S. Environmental Protection Agency, Office of Research and
Development, EPA/600/R-09/052F. Available from the National Technical Information
Service, Springfield, VA, and online at
https: / / cfpub.epa.gov/ncea/risk/recordisplay.cfm?deid=236252.
U.S. EPA. (2013). Human Health Ambient Water Quality Criteria and Fish Consumption Rates: Frequently
Asked Questions. Washington, DC: U.S. Environmental Protection Agency, Office of Science
and Technology, Office of Water. Available at: https: / /www.epa.gov/wqc /human-health-
water-quality-criteria.
U.S. EPA. (2014). EstimatedFish Consumption Rates for the U.S. Population and Selected Subpopulations
(NHANES 2003-2010). EPA-820-R-14-002. Washington, DC: U.S. Environmental Protection
Agency, Office of Science and Technology, Office of Water.
U.S. EPA. (2016). Guidelines for Human Exposure Assessment. Peer Review Draft. Washington, DC: U.S.
Environmental Protection Agency, Risk Assessment Forum.
Vaske, J-J-, Beaman, J., Manfredo, M.J., Covey, D., and Knox, R. (1996). Response strategy, recall
frame and digit preference in self-reports of angling participation. Human Dimensions of Wildlife,
14, 54-68.
Voiland, M.P., Gall, K.L., Lisk, D.J., MacNeill, D.B. (1991). Effectiveness of recommended fat-
trimming procedures on the reduction of PCB and mirex levels in brown trout from Lake
Ontario. Journal of Great Takes Research, 17(4):454^160.
109
-------
Washington Department of Fish and Wildlife. (2014). Available at:
http:/ /wdfw.wa.gov/fishing/reports plants.html.
Washington State Department of Ecology. (2012). Supplemental Information to Support the Fish
Consumption Kates Technical Support Document. Olympia, WA: Author. Available at:
http: / /www.ecv .wa.gov/programs / tcp / regs / fish/ 2012/FCR-doc.html.
Wathen, J.B., Lazorchak, J.M. , Olsen, A.R., and Batt, A. (2015). A national statistical survey
assessment of mercury concentrations in fillets of fish collected in the U.S. EPA national
rivers and streams assessment of the continental USA. Chemophere, 122:52-61.
Wentland, E. J., and Smith, K.W. (1993). Survey responses: An evaluation of their validity. San Diego, CA:
Academic Press.
Willett, W. (2012). Nutritional Epidemiology. Oxford University Press.
Willis, G.B. (2005). Cognitive interviewing. A tool for improving questionnaire design. Thousand Oaks, CA:
Sage.
Zabik, M.E., Booren, A., Zabik, M.J., Welch, R., and Humphrey, H. (1996). Pesticide residues, PCBs
and PAHs in baked, charbroiled, salt boiled and smoked Great Lakes lake trout. Food
Chemistry, 55(3):231—239.
Zabik, M.E., Zabik, M.J. (1999). Polychlorinated biphenyls, polybrominated biphenyls, and dioxin
reduction during processing/cooking food. Advances in FxperimentalMedicine and Biology,
459:213-231.
Zabik, M.E., Zabik, M.J., Booren, A.M., Nettles, M., Song, J.H., Welch, R., et al. (1995). Pesticides
and total polychlorinated biphenyls in Chinook salmon and carp harvested from the Great
Lakes: effects of skin-on and skin-off processing and selected cooking methods .Journal of
Agricultural and Food Chemistry, 43:993—1001.
Zale, A.V., Parrish, D.L., and Sutton, T. M. (2013). Fisheries Techniques, Third Edition. Bethesda
(Maryland): American Fisheries Society.
Zhang, L.C. (2000). Post-stratification and calibration. American Statistician, 54, 178-184.
110
-------
Appendices
-------
APPENDIX A. ADDITIONAL INSTRUMENT DEVELOPMENT
CONSIDERATIONS
For a survey to be readily understood by respondents, the questions must be simple and
straightforward while accomplishing study objectives. Additionally, questions should consider
cultural aspects, such as possible unique common names or preparation techniques for fish species
that are specific to the population of interest. The design of each specific question must consider
both sentence structure, wording, mode of administration, and position of questions within the
survey (see, for example, Dillman & Christian, 2005). These topics are discussed in this appendix.
The reader is referred to, Groves et al., 2009; Saris and Gallhofer, 2007; Bradburn et al., 2004;
Tourangeau et al., 2000; and the references cited therein for more information. Pollock et al., 1994,
is a good resource for integrating the varied survey disciplines as they apply to fisher surveys. More
recent books on fishery survey methods have not focused as well on the issues of question design as
it applies to fisher surveys.
If possible the development of an instrument (or questionnaire) should rely heavily on the use of
previously validated questions. If local data are to be compared against other data sets (e.g., state or
national data), the questions should be asked in the same way so as to ensure consistency of the data
collected. Appendix C provides some examples of questions that can be adapted to obtain data on
fish consumption and portion sizes.
Question structure: Four general types of question structure are available (Pollack et al., 1994):
(1) open-end questions, (2) closed-end questions with ordered response choices; (3) closed-end
questions with unordered response choices, and (4) partially closed-end questions. Open-end
questions have no categories from which the respondent can choose; however, interpretation of all
but the simplest open-end questions can be quite difficult. It should also be noted that standardized
coding of open-end questions for analysis can be quite costly and time consuming. Closed-end
questions provide several answer categories, which can be ordered sequentially (e.g., numerically) or
unordered. The answers to closed-end questions are easy to summarize quantitatively. Response
options must be selected carefully so that the choices are mutually exclusive, inclusive of all
reasonable choices, and easy to understand. Categories also may provide cues to aid respondents'
recall (Bradburn et al., 2004). Partially closed-end questions often allow an open-ended option such
as "other." This option represents a good compromise between open-ended and closed-end
structures (Pollack et al., 1994). It also should be noted that standardized coding of the "other"
category responses may prove difficult and will require careful review to see if any responses should
be "up-coded"—that is, recoded into the closed-end categories provided.
For closed-end questions, the specific ranges for each response alternative can affect the way in
which the question is answered, resulting in impression-based responses. Values in the middle range
of the scale selected are often assumed by respondents to reflect the "average" or "typical" behavior,
whereas the extremes of the scale are assumed to represent the extremes of the distribution (Groves
et al., 2009; Schwarz & Hippler, 2004). Thus, respondents will assume that the response scale
provided indicates knowledge of the behavior and form an impression based on the scale provided.
Respondents will then adjust their response based on their impression of how they compare to the
scale provided.
A-l
-------
Question wording: The specific wording of questionnaires on fish consumption must be developed
very carefully to elicit nonbiased responses. Here are some guidelines for question wording (Pollack
et al., 1994):
¦ All alternatives of a multiple-choice question should be given
¦ As few words as necessary should be used
¦ The units that apply to each response should be given
¦ The time frame covered by the survey should be clear
¦ Only one concept or issue should be addressed by each question
Draft questions should be reviewed carefully for any ambiguities and tested through the use of
qualitative testing procedures (Presser et al., 2004). For a review of question wording and design
guidelines, the reader is directed to Bradburn et al., 2004.
Question order: Topic sections should be arranged for the convenience of the respondent, not that
of the researcher. There is likely a logical order to grouping questions that will aid in respondent
recall. The questions should build on each other. For example, rather than asking, "Did you wear
your seatbelt the last time you rode in a car?," the following series of questions may be more
effective: "When was the last time you rode in a car? How long was the trip? Did you wear your
seatbelt?" This type of cognitive design can be very effective in minimizing respondent error and
should be used for important questions (Groves et al., 2009).
The first few questions of the survey might be the most critical, particularly for self-administered
surveys (e.g., mail or web), since these might determine whether the respondent chooses to
complete the questionnaire. Sensitive questions or questions that are difficult to answer should be
asked near the end of the interview so as not to threaten the respondent and possibly compromise
the rapport between interviewer and respondent. Sensitive questions include demographic questions
such as age, income, and education (Bradburn et al., 2004), and questions about whether the fisher
has an applicable fishing license or is familiar with a particular advisory or regulation (Pollock et al.,
1994).
Question design and presentation: Just as question features, such as, context, response scale, and
number of response options can influence the answers provided by respondents, the presentation
and design of survey questions can also affect how respondents answer. For self-administered paper
surveys or mail surveys, the placement and highlighting of skip instructions can affect errors of
omission or commission (Dillman et al., 2009). An error of omission occurs when a respondent
skips a question that should have been answered, whereas an error of commission occurs when a
respondent answers a question that should have been skipped. Another example of design
influences happens when, for long response lists, response options are presented in two columns
rather than a single column. The use of two columns will result in more selections from the first
column, resulting in bias. Errors associated with question branching or skipping can be greatly
reduced by use of web or computer assisted interview software approaches that automate skip
patterns.
For web surveys, features such as layout, spacing, and proximity can influence how respondents
answer questions. Tourangeau et al., 2013, describe five heuristics, or rules, that respondents employ
to interpret response scales, as follows.
A-2
-------
¦ Middle means typical — respondents will use the visual midpoint as the mean or typical
value
¦ Left and top mean first — generally that response options will follow a logical order
¦ Near means related — relative proximity to other options may imply a relationship
between those options
¦ Similarity in appearance means close in meaning — the use of highlighting, or giving
similar appearance to another highlighted option, may make the options seem related
¦ Up means good — vertical position on the screen may imply value about the item
For response fields requiring specific formats, such as dates, the layout of response fields and labels
can influence how likely respondents are to follow directions and provide answers appropriately. For
example, in a review of a number of studies looking at the proportion of respondents correctly
entering a date, the highest proportion of correctly formatted entries occurred when labels
mimicking the expected formatting were used in conjunction with close proximity to the desired
entry box (Tourangeau et al., 2013). In summary, the survey designer should be aware of the
influences that survey design features can have on respondents' answers.
A-3
-------
REFERENCES
Bradburn, N.M., Sudman, S., and Wansink, B. (2004). Asking questions: The definitive guide to
questionnaire design—-for market research, politicalpolls, and social and health questionnaires. San
Francisco, CA: Jossey-Bass.
Dillman, D.A. and Christian, L.M. (2005). Survey mode as a source of instability in responses across
surveys. Field Methods, 17.1: 30-52.
Dillman, D.A., Smyth, J.D., & Christian, L.M. (2009). Internet, mail, and mixed-mode surveys: The tailored
design method. Hoboken, NJ: Wiley.
Groves, R.M., Fowler, F.J., Couper, M.P., Lepkowski, J.M., Singer, E., and Tourangeau, R. (2009).
Survey methodology. Hoboken, NJ: John Wiley & Sons.
Pollock, K.H., Jones, C.M., and Brown, T.L. (1994). Angler survey methods and their applications in
fisheries management. American Fisheries Society SpecialVublication 25, American Fisheries
Society, Bethesda, Maryland.
Presser, S., Rothgeb, J.M., Couper, M.P., Lessler, J.T., Martin, E., Martin, J., and Singer, E. (2004).
Methods for testing and evaluating survey questionnaires. Vol. 546. Hoboken, NJ: John Wiley & Sons.
Saris, W.E. and Gallhofer, I.N. (2007). Design, evaluation, and analysis of questionnaires for survey research.
Hoboken, NJ: John Wiley & Sons.
Schwarz, N. and Hippler, H.-J. (2004). Response alternatives: The impact of their choice and
presentation order. In Measurement errors in surveys. P. P. Biemer, R. M. Groves, L. E. Lyberg, N.
A. Mathiowetz, and S. Sudman. (Eds.), Hoboken, NJ: John Wiley & Sons,
doi: 10.1002/9781118150382.ch3.
Tourangeau, R., Rips, L.J., and Rasinski, K. (2000). The psychology of survey reponse. Cambridge, U.K.:
Cambridge University Press.
Tourangeau, R., Conrad, F., and Couper, M.. (2013). The science of web surveys. New York, NY: Oxford
University Press.
A-4
-------
APPENDIX B. OVERVIEW OF DIETARY ASSESSMENT METHODS
Method
Description
24-hour recall
(24HR)
A very structured method in which the respondent is asked, using standard probes,
to recall and describe all food and beverages consumed, or only a specific food like
fish, in the preceding 24 hours. Can be interviewer- or self-administered either in-
person, by telephone, or web. Typically, data are captured using a standard software
program. Portion size estimating aids assist the respondent to recall amounts
collected.
Advantages: Provides the highest quality and least biased food intake data for a
single day. Can be used to collect total diet or just fish consumption. Analyzes
individual data for grams consumed per food and per day. If interviewer-
administered, low respondent burden and few literacy issues. Single day used to
estimate average dietary intake of a group; multiple days (or when paired with FFQ
data) used to estimate population distributions (NCI method described in Chapter 7).
Limitations: Multiple days needed or combined with FFQ to model estimates of usual
intake (Carroll et al., 2012; Tooze et al., 2006). A measure of frequency of
consumption is needed to improve the statistical modeling. Moderate to high cost
depending on mode of administration. Some measurement error due to
underreporting; may require trained staff (interviewers and dietary coders)
depending on mode of administration. Software programs do not include specific
species of fish. Requires respondent to estimate portion of fish consumed.
Instruments:
• U.S. Department of Agriculture (USDA) Automated Multiple Pass Method
(AMPM). Interviewer-administered software used in the National Health and
Nutrition Examination Survey (NHANES) to collect surveillance data. Over 2,500
food probes (80 probes related to fish) used to collect very detailed descriptive,
source, and portion data. Uses the USDA Food and Nutrient Database for Dietary
Studies (FNDDS) to apply food code, portion sizes, and nutrients (includes over
230 primary fish codes and 120 fish dishes). Requires post-interview processing
system (PIPS) and Survey Net coding system to generate a data set with total fish
consumption by person. These programs can be requested from USDA A
Agricultural Research Service and may require setting up a cooperative
agreement.
• National Cancer Institute (NCI) Automated Self-Administered 24-hour Recall
(ASA24). English and Spanish versions of web-based software that can be self- or
interviewer-administered. Uses same food database and detailed probes to
collect descriptive, source, and portion data. Validated against the AMPM
(Kirkpatrick et al., 2014). Includes 95 fish varieties. All reported foods are auto-
coded, so no dietary coding is required. This software is freely available at
http://appliedresearch.cancer.gov/diet/usualintakes/macros.html.
• University of Minnesota Nutrition Coordinating Center (NCC) Nutrition Data
System for Research (NDS-R). Windows-based, interviewer-administered,
multiple-pass dietary analysis program designed to collect and analyze total diet
using the NDS database of over 18,000 foods including 8,000 brand names (over
B-l
-------
Method
Description
480 primary fish codes and 240 fish dishes). This software is available at
http://www.ncc.umn.edu/products/ndsr.html and requires a licensing fee.
Fish
Focused Recall
A method that combines the short-term recall used in 24HR with either a list of
specific groups of fish to probe respondent to report the fish and amounts consumed
the previous day. Mode may be in-person, telephone, or web. Automated
instrument exists for collecting 24HR/food list combination.
Advantages: Provides reasonably good estimates of consumption of fish for previous
day. Single day used to describe average dietary intake of a group; multiple days (or
when paired with FFQ data) used to estimate population distributions (NCI method
described in Chapter 7). Relatively low respondent burden.
Limitations: If using the 24-hour/food list combination, it mav be difficult to include
universe of local and non-local fish on food list. Increased measurement error due to
underreporting of foods consumed as part of mixed dishes. Requires respondent to
estimate portion of fish consumed.
Instrument:
• The FCST (discussed in Chapter 4). A configurable computerized interview
originally designed to record fish consumption information for American
Indians/Alaska Natives that collects 24-hour recall, and fish frequency estimates
of annual consumption (including source and parts consumed), consumption by
children, and consumption at gatherings and special events (Kissinger et al.,
2010). The automated survey tool allows researchers to populate the software
with fish species of interest and define the seasonality and forms of interest. The
fish database includes 77 species of fish. Survey includes a booklet with images
of a variety of species, portion amounts, and fish preparation methods. Available
from EPA free of charge at https://www.epa.gov/fish-tech/epa-guidance-
developing-fish-advisories.
Food
Record/Diary
(FR)
A method in which the respondent records all the foods and amounts consumed as
they are eaten, over 1-4 consecutive days. Although more than 4 days results in
decreased reporting (Gersovitz et al., 1978), 7-day records are sometimes used. The
interview modes may be mail or web. Portion size aids assist the respondent to
report amounts collected or portions may be weighed by the respondent (weighed
food record).
Advantages: Food description and amounts are recorded in real time and food
portions can be directly measured. Can be used to collect total diet or only fish
consumption. Single day used to describe average dietary intake of a group; multiple
days needed or combined with FFQ to estimate population distributions.
Limitations: Respondent training required; subject to bias because respondent mav
change diet or reporting over time; requires respondent motivation and literacy; high
respondent burden; increased measurement error with more days of reporting;
requires dietary coding and processing to generate amount consumed; weighed food
records require equipment and additional training. Instrument instructions will need
to emphasize fish consumption.
Instrument:
B-2
-------
Method
Description
• Fred Hutchinson Cancer Research Center (FHCRC). Multiple Day Food Record.
Use of template requires a fee. Contact FHCRC Nutrition Assessment at (800)
460-7270 or (206) 667-4161 or at
http://sharedresources.fhcrc.org/sites/default/files/MultipledavFoodRecord.pdf
Food Frequency
Questionnaire
(FFQ)
A method to assess usual frequency of consumption of each listed food over a
reference period (e.g., months or a year). FFQs may be based on an extensive list of
food items (total diet) or a short list of specific foods. Typically self-administered
using paper- or web-based formats, but can be interviewer administered, in-person,
or by telephone. May ask discrete portion sizes or assign default portions.
Advantages: Estimate usual intake over long time period up to 1 vear. Best suited to
rank individuals according to usual consumption to assess association between
dietary intake and disease risk. May be combined with 24-hour or FR to describe
usual intake. Forms can be optically scanned.
Limitations: Manv details of dietary intake not captured, portion sizes not accurate
due to categorical nature. Significant measurement error found with total diet FFQ.
Food list must reflect universe of foods of interest (including mixed dishes) in order
to capture usual intake. Self-administered FFQ may require a high level of literacy; if
however, the FFQ is administered by a personal interviewer, this constraint is not an
issue.
Instruments:
• National Cancer Institute (NCI) Diet History Questionnaire (DHQ). Paper or web-
based instrument in English and Spanish that consists of a list of 134 food items,
including six questions about fish consumption. Reference period is past year or
month. Freely available
(http://riskfactor.cancer.gov/DHQ/webquest/index.html).
• Harvard FFQ. Paper- or web-based instrument that consists of 140+ food items.
Five questions ask about fish consumption Reference period is past year.
Purchase and analysis cost is $15-$20 per questionnaire
(https://regepi.bwh.harvard.edu/health/nutrition.html).
• Block FFQ. Paper- or web-based instrument in English and Spanish that consists
of a list of 127 food items. Five questions ask about fish consumption. Purchase
and analysis cost on reauest (http://www.nutritionauest.com).
Fish
Diet Screener
A short list offish foods (usually 15-30 foods); may include portion size questions.
Mode may be mail, in-person, phone, or web.
Advantages: Low respondent burden. Should be used only to capture information on
a group of specific foods or a single food component.
Limitations: No estimation of upper percentiles of usual intake, substantial amount
of measurement error, literacy required for mail or web administration.
Instruments:
• USDA/NHANES fish frequency questionnaire. Asks frequency of consumption of
29 specific fish. Reference period is past 30 days. Publicly available
(Tittp://www.cdc.gov/nchs/nhanes/nhanes questionnaires.htm).
B-3
-------
Method
Description
• Seafood Assessment Survey. Asks frequency and amount of consumption of
clams and mussels by season.
(http://appliedresearch.cancer.gov/diet/shortreg/instruments/fialkowski shellfis
h assessment.pdf).
B-4
-------
REFERENCES
Carroll, R. J., Midthune, D., Subar, A. F., Shumakovich, M., Freedman, L. S., Thompson, F. I... &
Kipnis, V. (2012). Taking advantage of the strengths of 2 different dietary assessment
instruments to improve intake estimates for nutritional epidemiology. American Journal of
Epidemiology, 175(4):340-347.
Gersovitz, M., Madden, J.P., Smiciklas-Wright, H. (1978). Validity of the 24-hour dietary recall and
seven-day record for group comparisons. Journal of the American Dietetic Association, 73, 48-55.
Kirkpatrick, S., Subar, A., Douglass, D., Zimmerman, T., Thompson, F., Kahle, L., George. S.,
Dodd, K., and Potischman, N. (2014). Performance of the Automated Self-Administered
24-hour Recall relative to a measure of true intakes and to an interviewer-administered
24-h recall. The American Journal of Clinical Nutrition, doi: 10.3945/ajcn.114.083238.
Tooze, J.A., Midthune, D., Dodd, K.W., Freedman, L.S., Krebs-Smith, S.M., Subar, A.F., Guenther,
P.M., Carroll, R.J., Kipnis, V. (2006). A new statistical method for estimating the usual intake
of episodically consumed foods with application to their distribution. Journal of the American
Dietetic Association, Oct;106(10):l 575-87.
B-5
-------
APPENDIX C. APPLICATION OF THE NCI METHOD MACROS
The following information has been extracted from the NCI Method SAS Macros and User Guides
available at http://epi.grants.cancer.gov/diet/usualintakes/macros.html and described based on the
NCI Method macro experience of the authors.
The macro variables to define for use of the MIXTRAN are:
¦ Required
— Data= Specifies the data set to be used.
— Response= Specifies the 24-hour recall variable (in our example, this is
"GramsFish")
— Foodtype= Specifies a name for the analysis, used to identify the output data sets.
This value can be the same as the response variable
— Subject= Specifies the variable that uniquely identifies each subject (in our
example, this is "Participant")
— Repeat= Specifies the variable that indexes repeated for each subject (in our
example, this is "RecallDay")
— Covars_prob= Specifies a list of covariates for the first part of the model that
models the probability of consumption (in our example, these could include
"Season," "Gender," "Age," "Bodyweight," "FishFreq")
— Covars_amt= Specifies a list of covariates for the second part of the model that
models the consumption-day amount (in our example, these could include
"Season," "Gender," "Age," "Bodyweight," "FishFreq")
— Outlib= Specifies a directory where output data sets are stored
— Modeltype= Specifies the model. For the best estimates "corr" should be
specified, though it may not be necessary or feasible. A statistician should be
consulted to determine if an uncorrelated model would be appropriate. The
possible values are:
1. "null string" = fit correlated model
2. "corr" = fit correlated model
3. "nocorr" = fit uncorrelated model
4. "amount" = fit amount-only model
¦ Optional
— Seq= Specifies one or more sequence indicator variables to account for effects
due to the sequence number of a subject's records. This cannot also appear in
covars_prob or covars_amt (in our example this is "Seq")
— Weekend= Specifies the weekend (Fri.-Sun.) indicator to account for a weekend
effect. This variable can NOT also appear in covars_prob or covars_amt (in our
example, this is "Weekend")
— Vargroup= Specifies a variable that groups observations to allow the model to
incorporate a separate residual variance parameter for each of these groups of
C-l
-------
observations. If the output from this macro is to be used in the DISTRIB macro,
then only the weekend variable can be used.
— Numvargroups= Specifies the number of groups defined by the vargroup
variable. If the output from this macro is to be used in the DISTRIB macro and
weekend is the "vargroup" variable, then the number of groups is 2.
— Replicate_var= Specifies the variable to be used in the replicate statement of
PROC NLMIXED or the freq statement of PROC UNIVARIATE. The
specified variable must be integer valued (this is akin to a survey weight)
Note that there are other macro variables that can be defined. Please see the User's Guide to
determine their necessity for your analysis, although in most cases, those listed above are the ones
necessary for best estimates.
After the MIXTRAN macro runs, it produces a number of output data sets, two of which will then
be used by the DISTRIB macro.
"The DISTRIB macro uses results from the MIXTRAN macro and
estimates the distribution of usual intake for episodically foods, foods
consumed every day, and nutrients (Tooze et al., 2006). First, the DISTRIB
macro reads data sets of parameter estimates predicted values output by the
MIXTRAN macro. Then, Monte Carlo simulation of the random effect(s)
is used to estimate the distribution of usual intake." (NCI Method,
DISTRIB macro, version 2.1)
The macro variables to define for use of the DISTRIB macro are:
¦ Required
— Seed= Specifies the seed for the random number used for the Monte Carlo
simulation of the random effects ul and u2
— Nsim_mc= Specifies the number of repetitions to be used in the Monte Carlo
simulation. For each subject, record will be output for each repetition.
— Modeltype= Specifies the model that was used by the MIXTRAN macro to
prepare the data for the DISTRIB macro. The value must be the same as the
model declared for the MIXTRAN macro.
— Pred= Specifies the name of the data set containing values calculated in the
MIXTRAN macro. The data set will be named _pred_XX, where XX =
Foodtype specified in MIXTRAN.
— Param= Specifies the name of the data set containing the parameter estimates
calculated in the MIXTRAN macro. The data set will be named _param_XX,
where XX = Foodtype specified in MIXTRAN.
— Outlib= Specifies the library reference to which the output data set of
distributions will be written.
— Subject= Specifies the variable that uniquely identifies each subject. Required
when "weekend" is used in MIXTRAN. (In our example this is "Participant")
— Food= Specifies a name for the analysis, used to label the output data set.
C-2
-------
— NLoptions= Specifies a list of options to be added to all calls to PROC
NT All X IT) for example: nloptions=qpoints=l gconv=le-12 itdetails. This may
be necessary if the models fail to converge. For example, it may be necessary to
raise or lower the quadrature points (qpoints) or the convergence criteria (gconv).
Please consult the User's Guide and SAS Help for the NLMixed procedure for
more information.
Note that there are other macro variables that can be defined. Please see the NCI Method macro
documentation and user's guide available at
http: / /epi.grants.cancer.gov/diet/usualintakes /macros single.html to determine their necessity for
your analysis, although in most cases, those listed are the ones necessary for best estimates.
The DISTRIB macro outputs a data set "_mcsiml" that can be used to estimate the distribution of
UFCR for the total population and subpopulations that were included as covariates. If you are going
to use this data file, you will need to comment out a statement in the DISTRIB macro where it
deletes the data set. The macro contains a comment to show the user where this needs to be done. It
is in the last procedure call of the macro.
C-3
-------
REFERENCES
Tooze, J.A., Midthune, D., Dodd, K.W., Freedman, L.S., Krebs-Smith, S.M., Subar, A.F., Guenther,
P.M., Carroll, R.J., Kipnis, V. (2006). A new statistical method for estimating the usual intake
of episodically consumed foods with application to their distribution. Journal of the American
Dietetic Association, Oct;106(10):l 575-87.
C-4
-------
APPENDIX D. FCST HARD-COPY QUESTIONNAIRE AND
DOCUMENTATION
D-l
-------
[HARD COPY QUESTIONNAIRE TO BE MODIFIED FOR USE]
INTRODUCTION
We appreciate your willingness to participate in our fish and shellfish consumption survey. The
information given in response to this questionnaire will help [insert name of tribe/ agency/organisation
conducting the study\ to understand the rates of fish and shellfish consumption and the species or types
of fish and shellfish regularly consumed in \insert area ofinterest\.
All of the information you provide to us is confidential. Your responses to the questions will be
combined with those of others so that your answers cannot be identified. If you have any questions,
you are welcome to call [insert individual's name andaffiliation\ at [insert telephone number].
It takes about [insert time to complete to complete this questionnaire. There are four parts. In Part One,
you will be asked how much fish and shellfish you ate yesterday. The second and third parts focus
on particular types of fish and shellfish you eat, and where it was prepared. Finally, you will be asked
for some general information.
Today's Date
Month Day Year
PARTON1-: DIETARY INTAKE - 24 I IOl R RECAEE
Please tell us whether you ate any fish or shellfish yesterday. Think about what you ate for breakfast,
lunch, dinner, and any snacks. Include any fish or shellfish that were part of sandwiches, salads, or
other mixed foods (e.g., seafood pasta).
Ql. Did you eat any fish or shellfish yesterday from the time you woke up until the time you went
to sleep for the night? Please include meals and snacks. Mark the box next to your answer.
~ Yes
~ No [Go to Part Two.]
~ Don't Know [Go to Part Two.]
Q2. Did you eat more than one species or type of fish or shellfish yesterday? Mark the box next to
your answer.
~ Yes
~ No
~ Don't Know
D-2
-------
\Populate this table with species of interest to the study. Keep the other/ unknown categories.]
Q3. Please mark the box
Q4. Was the fish/shellfish prepared at
Q5. How much of each species
next to each species or
home, or at a restaurant or another
did you eat yesterday?
type of fish or
place? (Mark all that apply)
Write the total number of
shellfish you ate
ounces (oz) or pounds
yesterday.
(lbs).
~
Home
ounces (oz)
~
Salmon, canned
~
Restaurant or Other Place
pounds (lbs)
~
Don't Know
[CIRCLE ounces or pounds]
~
Home
ounces (oz)
~
Salmon, fresh
~
Restaurant or Other Place
pounds (lbs)
~
Don't Know
[CIRCLE ounces or pounds]
~
Home
ounces (oz)
~
Trout
~
Restaurant or Other Place
pounds (lbs)
~
Don't Know
[CIRCLE ounces or pounds]
~
Home
ounces (oz)
~
Other finfish
~
Restaurant or Other Place
pounds (lbs)
~
Don't Know
[CIRCLE ounces or pounds]
~
Home
ounces (oz)
~
Unknown finfish
~
Restaurant or Other Place
pounds (lbs)
~
Don't Know
[CIRCLE ounces or pounds]
~
Home
ounces (oz)
~
Clams
~
Restaurant or Other Place
pounds (lbs)
~
Don't Know
[CIRCLE ounces or pounds]
~
Home
ounces (oz)
~
Mussels
~
Restaurant or Other Place
pounds (lbs)
~
Don't Know
[CIRCLE ounces or pounds]
~
Home
ounces (oz)
~
Shrimp
~
Restaurant or Other Place
pounds (lbs)
~
Don't Know
[CIRCLE ounces or pounds]
~
Home
ounces (oz)
~
Other shellfish
~
Restaurant or Other Place
pounds (lbs)
~
Don't Know
[CIRCLE ounces or pounds]
~
Home
ounces (oz)
~
Unknown shellfish
~
Restaurant or Other Place
pounds (lbs)
~
Don't Know
[CIRCLE ounces or pounds]
D-3
-------
PAIH'TWO: I INI ISII CONSUMPTION -SMASON, I Kl-Ql KNCY, PORTIONS
For this part of the survey we will ask about your consumption of fmfish and shellfish over the past
12 months. The first questions are about what species of fmfish you ate, the amount you ate, and
how often you ate each species over a year.
The amount of fish you eat and how often you eat it may depend on the time of year. For example,
if there are seasonal differences in how often you eat fish, you may answer two different ways: when
it is fresh and readily available, and when it is not in season. Or, if you believe there is no difference
in how often you eat the fish, you can tell us how often you eat fish in general without regard to
when it is in season. Please remember to include fish you eat at breakfast, lunch, dinner, and snacks.
Do not include fish you eat at ceremonies or community gatherings and events, as we will ask you
about those in a later section.
Q6. In the past 12 months, did you ever eat \insertfirst species ofinterest^
~ Yes
~ No [Go to Q21.]
~ Don't Know [Go to Q21.]
Q7. Is the amount of [insertfirst species of interest'] you ate different when the fish was in season and
is fresh and available compared to when the fish was not in season? Please keep in mind that
the season is [insert season for pedes of interest].
~ Yes
~ No [Go to Q12.]
Q8. How often in the past 12 months did you eat [insert first spedes ofinterest\ when the fish was in
season?
~
Day
~
Week
times per
~
Month
~
Season
(WRITE number of times on line)
~
Year
(MARK box)
D-4
-------
Q9. How much did you usually eat in a sitting when the fish is in season?
~ Ounces (oz)
~ Pounds (lbs)
(WRITE amount on line) (MARK box)
Q10. How often in the past 12 months did you eat \insertfirst species of interest^ when the fish was not
in season?
~
Day
~
Week
times per
~
Month
~
Season
(WRITE number of times on line)
~
Year
(MARK box)
Qll. How much did you usually eat in a sitting when the fish is not in season?
~
(WRITE amount on line) (MARK box)
Q12. As your consumption did not vary by season, in the past 12 months, how often did you eat
\insertfirst species of interestf| ?
~
Day
~
Week
times per
~
Month
~
Season
(WRITE number of times on line)
~
Year
(MARK box)
D-5
-------
Q13. How much did you usually eat in a sitting?
~ Ounces (oz)
~ Pounds (lbs)
(WRITE amount on line) (MARK box)
Q14. What percent of the time did you eat \insert species of interest] with skin?
~
Never
~
61 to 70% of the time
~
1 to 10%
of the time
~
71 to 80% of the time
~
11 to 20°/
o of the time
~
81 to 90% of the time
~
21 to 30°/
o of the time
~
91 to 99% of the time
~
31 to 40°/
o of the time
~
All of the time (100%)
~
41 to 50°/
o of the time
~
Don't Know
~
51 to 60°/
o of the time
Q15. What percent of the time did you eat [insert species of interest eggs if they are available?
~
Never
~
61 to 70% of the time
~
1 to 10%
of the time
~
71 to 80% of the time
~
11 to 20°/
o of the time
~
81 to 90% of the time
~
21 to 30°/
o of the time
~
91 to 99% of the time
~
31 to 40°/
o of the time
~
All of the time (100%)
~
41 to 50°/
o of the time
~
Don't Know
~
51 to 60°/
o of the time
Q16. What percent of the time did you eat the head, bones, or organs of [insert pedes of interest^
~
Never
~
61 to 70% of the time
~
1 to 10%
of the time
~
71 to 80% of the time
~
11 to 20°/
o of the time
~
81 to 90% of the time
~
21 to 30°/
o of the time
~
91 to 99% of the time
~
31 to 40°/
o of the time
~
All of the time (100%)
~
41 to 50°/
o of the time
~
Don't Know
~
51 to 60°/
o of the time
Q17. What percent of the time was the [insert species of interest] that you eat from the grocery store?
~
Never
~
61 to 70% of the time
~
1 to 10%
of the time
~
71 to 80% of the time
~
11 to 20°/
o of the time
~
81 to 90% of the time
~
21 to 30°/
o of the time
~
91 to 99% of the time
~
31 to 40°/
o of the time
~
All of the time (100%)
~
41 to 50°/
o of the time
~
Don't Know
~
51 to 60°/
o of the time
D-6
-------
Q18. What percent of the time was the \insert species of interest] that you eat caught by you, your
family members, or friends?
~ Never
~ 1 to 10% of the time
~ 11 to 20% of the time
~ 21 to 30% of the time
~ 31 to 40% of the time
D 41 to 50% of the time
D 51 to 60% of the time
~ 61 to 70% of the time
~ 71 to 80% of the time
~ 81 to 90% of the time
~ 91 to 99% of the time
~ All of the time (100%)
~ Don't Know
Q19. What percent of the time was the \insert species of interest that you eat from restaurants?
~ Never
~ 1 to 10% of the time
~ 11 to 20% of the time
~ 21 to 30% of the time
~ 31 to 40% of the time
D 41 to 50% of the time
D 51 to 60% of the time
~ 61 to 70% of the time
~ 71 to 80% of the time
~ 81 to 90% of the time
~ 91 to 99% of the time
~ All of the time (100%)
~ Don't Know
Q20. What percent of the time was the \insert species of interest that you eat from other sources?
~ Never
~ 1 to 10% of the time
~ 11 to 20% of the time
~ 21 to 30% of the time
~ 31 to 40% of the time
D 41 to 50% of the time
D 51 to 60% of the time
~ 61 to 70% of the time
~ 71 to 80% of the time
~ 81 to 90% of the time
~ 91 to 99% of the time
~ All of the time (100%)
~ Don't Know
[Repeat questions 6-20 for each finfish species of interest and/ or otherfinfisE\
PAIH'TIIKl-li; SlIliUT'lSI I CONSUMPTION - I'KliQl 1!NCY, PORTIONS
The following questions are about what species of shellfish you eat, the amount you eat, how often
you eat each species, and what parts of the shellfish you consume.
Q21. In the past 12 months, did you ever eat \insertfirst shellfish species of interest^
~ Yes
~ No [Go to Q35.]
~ Don't Know [Go to Q35.]
D-7
-------
Q22. Is the amount of \insertfirst shellfish species of interest\ you ate different when the \insert they were/it
was] in season and is fresh and available compared to when \insert they were/it was] not in
season? Please keep in mind that the season is \insert season for species ofiinteresi\.
~ Yes
~ No [Go to Q27.]
Q23. How often in the past 12 months did you eat \insert first shellfish species of interest when \insert they
were/it was] in season?
~
Day
~
Week
times per
~
Month
~
Season
(WRITE number of times on line)
~
Year
(MARK box)
Q24. How much did you usually eat in a sitting when \insert thej were/it was] in season?
~ Ounces (oz)
~ Pounds (lbs)
(WRITE amount on line) (MARK box)
Q25. How often in the past 12 months did you eat \insert first shellfish species of interest when \insert they
were/it was] not in season?
~
Day
~
Week
times per
~
Month
~
Season
(WRITE number of times on line)
~
Year
(MARK box)
Q26. How much did you usually eat in a sitting when \insert thej were/it was] not in season?
~ Pound" £) |GOT°Q27J
(WRITE amount on line) (MARK box)
D-8
-------
Q27. As your consumption did not vary by season, in the past 12 months, how often did you eat
\insertfirst shellfish species of interest] ?
~ Day
~ Week
times oer ,
1 ~ Month
(WRITE number of times on line) ^ Season
(MARK box)
Q28. How much did you usually eat in a sitting?
~ Ounces (oz)
~ Pounds (lbs)
(WRITE amount on line) (MARK box)
Q29. What percent of the time did you eat the meat/whole body of the [insert species of interestf]?
~
Never
~
61 to 70% of the time
~
1 to 10%
of the time
~
71 to 80% of the time
~
11 to 20°/
o of the time
~
81 to 90% of the time
~
21 to 30°/
o of the time
~
91 to 99% of the time
~
31 to 40°/
o of the time
~
All of the time (100%)
~
41 to 50°/
o of the time
~
Don't Know
~
51 to 60°/
o of the time
Q30. What percent of the time do you eat other parts of the \insert shellfish species of interest^
~
Never
~
61 to 70% of the time
~
1 to 10%
of the time
~
71 to 80% of the time
~
11 to 20°/
o of the time
~
81 to 90% of the time
~
21 to 30°/
o of the time
~
91 to 99% of the time
~
31 to 40°/
o of the time
~
All of the time (100%)
~
41 to 50°/
o of the time
~
Don't Know
~
51 to 60°/
o of the time
Q31. What percent of the time was the \insert shellfish species of interest'] that you ate from the grocery
store?
~
Never
~
1 to 10%
of the time
~
11 to 20°/
o of the time
~
21 to 30°/
o of the time
~
31 to 40°/
o of the time
~
41 to 50°/
o of the time
~
51 to 60°/
o of the time
~ 61 to 70% of the time
~ 71 to 80% of the time
~ 81 to 90% of the time
~ 91 to 99% of the time
~ All of the time (100%)
~ Don't Know
D-9
-------
Q32. What percent of the time was the [,insert shellfish species of interest\ that you ate caught by you,
your family members, or friends?
~ Never
~ 1 to 10% of the time
~ 11 to 20% of the time
~ 21 to 30% of the time
~ 31 to 40% of the time
D 41 to 50% of the time
D 51 to 60% of the time
~ 61 to 70% of the time
~ 71 to 80% of the time
~ 81 to 90% of the time
~ 91 to 99% of the time
~ All of the time (100%)
~ Don't Know
Q33. What percent of the time was the \insert shellfish species of interest that you ate from restaurants?
~ Never
~ 1 to 10% of the time
~ 11 to 20% of the time
~ 21 to 30% of the time
~ 31 to 40% of the time
D 41 to 50% of the time
D 51 to 60% of the time
~ 61 to 70% of the time
~ 71 to 80% of the time
~ 81 to 90% of the time
~ 91 to 99% of the time
~ All of the time (100%)
~ Don't Know
Q34. What percent of the time was the \insert shellfish species of interest that you ate from other
sources?
~
Never
~
61 to 70% of the time
~
1 to 10%
of the time
~
71 to 80% of the time
~
11 to 20°/
o of the time
~
81 to 90% of the time
~
21 to 30°/
o of the time
~
91 to 99% of the time
~
31 to 40°/
o of the time
~
All of the time (100%)
~
41 to 50°/
o of the time
~
Don't Know
~
51 to 60°/
o of the time
[Repeat questions 21-34 for each species of interest and/ or other shellfisE\
D-10
-------
PAIH' I C)l K: GKNlMiAl, QUESTIONS
This is the final section of the questionnaire. It asks you for general information on fish and shellfish
consumption followed by a few questions about you.
Q35. Sometimes people's consumption of fish and shellfish changes over time. Has the amount of
fish and/or shellfish consumed by you or your family changed over the last 20 years?
~ Yes
~ No [Go to Q38.]
~ Don't Know [Go to Q38.]
Q36. How has the amount of fish and/or shellfish you have eaten over the past 20 years changed?
~ Eat more now
~ Eat less now
~ Eat different types now
~ Prefer not to say [Go to Q38.]
~ Don't Know [Go to Q38.]
Q37. Please tell us what you think has caused the change in the amount or type of fish or shellfish
you now eat.
Q38. In most years, how many ceremonies, large gatherings, or other community events do you
attend where fish and shellfish is consumed? \insert examples of events pertinent to the study
population\
Number of events per year [If none or zero, go to Q41.]
D-ll
-------
Q39. At these gatherings, do you ever eat [insertfirst species of interest] ?
~ Yes
~ No [GotoQ41.]
~ Don't Know [GotoQ41.]
Q40. How much [insertfirst species of interest\ do you usually eat in at these gatherings?
~ Ounces (oz)
~ Pounds (lbs)
(WRITE amount on line) (MARK box)
[Repeat questions 39 and 40 for each species of interest and/ or otherfish/shelfisE\
Q41. What is your gender?
~
Male
~
Female
1 low
old are you?
~
18 to 29 years
~
30 to 39 years
~
40 to 49 years
~
50 to 64 years
~
65 to 79 years
~
80 years or older
[Change and insert age groups of interest to study\
Q43. What is your height (in feet and inches) and weight (in pounds)?
Height: feet inches
Weight: pounds
D-12
-------
Q44. What is your race? (Adjust categories to fit the target population)
~ White
~ Black or African American
~ American Indian/Alaska Native
~ Asian
~ Native Hawaiian or other Pacific Islander
~ Two or more races
Q45. Are you of Hispanic origin? (Adjust categories to fit the target population)
~ No, not of Hispanic, Latino, or Spanish origin
~ Yes, Mexican, Mexican American, Chicano
~ Yes, Puerto Rican
~ Yes, Cuban
~ Yes, other Hispanic, Latino, or Spanish origin
Q46. What is your household annual income?
~ $0-$10,000
~ $10,001-120,000
~ $20,001-$30,000
~ $30,001-$40,000
~ $40,001-$50,000
~ $50,001 and over
\Change and insert income groups appropriate for study population\
CONCLUSION
Thank you very much for your cooperation in participating in this survey. Your participation will
significantly contribute to information needed to help protect our natural resources.
As mentioned in the introductory letter, we will be mailing you a brief follow-up survey \insert time
frame as decided on by the study\ that includes only Part One of this survey. [Ifapplicable for study - We will
mail your check for $25.00 soon after we receive your completedfollow-up survey.]
Please call us if you have any questions.
Again, thank you very much.
D-13
-------
Developing the Booklet of Photographs
Preparation of Species Photographs
Survey designers will determine the specific fish and shellfish species to be included in the survey.
For analysis and exposure assessment, it is important to have a comprehensive list of fish species
included in the survey. However, the corresponding respondent burden and the impact of
assembling booklets with a large number of species photographs should be considered.
Sources
Once the species included are defined in the CAPI configuration, the Supervisor searches for images
of specific fish species to include in the booklet. Various sources could be used to obtain
photographs or images. Consultation with the local department of fisheries may be useful.
Government fish and wildlife agencies often have catalogs or posters of local fish species. Local
universities may also have this type of resource. Internet searches provide a wealth of sources for
fish and shellfish images. Using readily available electronic images obtained from the web is cost-
effective and efficient. However, images obtained from web searches should always be verified
against another source. In addition, some websites have limited access to images. They may require
subscriptions or the use of passwords. Also, the format and the quality of on-line images may vary
greatly from one website to another. It may be preferable to use images from websites associated
with academic, government, or commercial institutions. This would help ensure some degree of
uniformity and authenticity in the images. Examples of some of these websites include the
following:
1. The Washington Department of Fish and Wildlife supplies images of Washington state fish
and shellfish on the website. Unless otherwise indicated, the images are owned by the
Washington Department of Fish and Wildlife or have been made available to the
Department for public use. They may be used for non-profit or educational purposes
provided that the Department or copyright holder is properly credited. Available at:
http://wdfw.wa.gov/fishing/salmon/identification.html
2. The U.S Food and Drug Administration's (FDA) Seafood Products Research Center and the
Center for Food Safety & Applied Nutrition have developed the Regulatory Fish
Encyclopedia. The Regulatory Fish Encyclopedia (RFE) is a compilation of data in several
formats that assists with the accurate identification of fish species. Available at:
http:/ /www.fda.gov/food/ foodscienceresearch/rfe/default.htm
3. Other sources may include commercial fishery, hatchery or aquaculture websites. Available
at: http://www7.taosnet.com/platinum/data/light/species/species.html
Preparation of Seafood Portion Photographs
Models of seafood portions are important in determining the amount of each type of seafood
consumed by respondents. In general, the survey attempts to quantify seafood consumption in
terms of uncooked wet weights. Eliciting measurements of uncooked seafood consumption is
D-14
-------
problematic in that consumers often do not evaluate their consumption in terms of uncooked, but
rather cooked seafood portions. Cooking can cause substantial volume changes, and using uncooked
seafood model portions may cause interviewees to underestimate the amount of seafood they
consume. Therefore, EPA decided that cooked seafood portion models should be shown to
interviewees while seafood weights prior to cooking are to be used to quantify consumption.
Consequently, it is important that accurate weights of edible tissue (+/- 0.1 grams) be associated
with each photograph. Again, the CAPI records consumption in terms of raw or uncooked weights.
There is some variability associated with the tissue weights for model types (e.g., shellfish portions,
canned clams and fish). Therefore, replicates of the tissue mass associated with a photograph (e.g., 6
clams, 6 cans of fish, etc.) should be weighed and the mean and standard deviation computed.
Types of Preparation Models
A variety of approaches are available for depicting seafood portions, these include life-like casts of
seafood, rough models giving the interviewee a general idea of portion volume, or life size photos.
Casts accurately depict portion size; however, they can be difficult to prepare, particularly for cooked
seafood. Further, preparation of multiple casts for multiple sets of interview materials is time
consuming. Rough seafood volume models can also be useful, but uncertainty is created due to the
discrepancy between the shape of the model and the actual shape of the seafood portion. Digital
photographs do not communicate volume information to the same degree physical models do.
However, photographs are easier to prepare than physical models, can accurately depict cooked
seafood portions, can be readily organized, are easily replicated to provide multiple sets of interview
materials, are easy to physically transport during the interview process, and can be readily
transmitted via the Internet. For these reasons, EPA chose actual size photographs over other
methods of portraying seafood portion size.
However, using online images for the photographic models is not desirable since the images will
vary greatly from one another depending on the source. Moreover, it is difficult to determine if the
portion size represented in the on-line image is approximately what an individual might consume,
with larger or smaller portions being conceptually accessible multiples or fractions of the proposed
portion. Also, it is not possible to associate an uncooked weight with the portion size represented in
the on-line image.
Therefore, the guidelines presented here outline the best approach for the creation of photographic
preparation and portion size models. The objective is to obtain raw weights, prepare or cook the
items, and photograph the cooked items in a standardized manner.
Preparation and Portion Size Models
The seafood to be photographed and the types of models to be created depend on the specific
CAPI configuration. The supervisor will need to configure the CAPI software by designating a
ModeHD, Model Description, and a Portion Description for each photographic model. The ModeHD
provided will uniquely identify a physical portion model in the database. The Model Description and
Portion Description provided should be helpful to the interviewer and respondent so as to select an
appropriate physical model that best represents the amount consumed. The CAPI software can be
configured so that multiple models can be assigned to a species.
D-15
-------
Survey designers will need to determine how the fish and shellfish species are typically consumed. If
a type of seafood can be prepared in multiple ways and portion sizes of these different preparations
differ in weight, then it is important that models of each preparation type be created. For example,
fish may be consumed as fillets, steaks, canned and in fish soups. Clams may be consumed fresh,
canned, in fritters, and in clam chowders or soups. For shellfish in particular, the number of
organisms depicted per model are selected so that interviewees could describe their consumption in
terms of a reasonably small number of multiples of the model portion. For example, an individual
consuming 24 manila clams at a sitting would do so by indicating that they consumed 4 model
portions, the model portion being 6 clams.
Creating Models
A detailed plan should be prepared for each fish species and each type of preparation model that
needs to be created. The plan should include details about the procurement of seafood items,
different preparation methods, recipes, weights of representative seafood items, photo descriptions,
etc. Organizing this information into a table would be helpful.
Seafood items should be purchased at local markets or ordered from local wholesale retailer. The
items should be representative of what would be available to the target survey population. The raw
materials needed for creating each model type and the preparation of the model are described here.
1. Top loading or triple beam balance: Used to record the weight of raw fish and shellfish
tissue within a weight of +/- 0.1 g.
2. Kitchen facilities for preparing seafood
3. Exacto knives: Used to scrape edible tissue out of shells.
4. Chef s knife: Used to pare down fillet models.
5. Weighing trays or beakers: Used to hold tissue during the weighing process.
In addition to these items, additional equipment needed may include pans, knives, steamers, or other
utensils.
Obtaining weights
The raw seafood is weighed using a standard food scale and the weight recorded. Where an exact
weight is not associated with a portion size (e.g., six shrimp vs. 8 oz fillet), the weight of six
representative raw portions are measured to determine the mean standard deviation of the
preparation. For soups and chowder preparations, it is important to note that the weights of interest
are for the fish or clam tissue and not all other materials in the preparation (i.e. onions, potatoes,
cream, etc.). The mean weight is entered into the database while individual portion weight is
associated with the photographic model.
Seafood Preparation
If specific recipes or preparation instructions are needed, they should be obtained prior to the photo
shoot. Again, the models of the various preparation types should be representative of what is
available and consumed by Tribal members. A shopping list of additional ingredients (e.g., butter,
D-16
-------
lemon, breadcrumbs, flour, eggs, and oil) needed to prepare the seafood items according to the
recipes or preparation is needed.
Taking Photographs
It is recommended that photos be taken in two photo sessions over 2-3 days. The number of
photographic models and the resources available will determine the logistics for the photo shoot.
The required items needed for the photo shoot are:
1.
Digital camera
2.
Tape measure
3.
Tripod
4.
Neutral colored plates
5.
White poster board
6.
Color printer
Neutral colored plates (such as neutral blue) are used to stage the prepared seafood items for the
photographs. White plates are not recommended since white seafood items might not be as visible.
The same size plates and bowls are used consistently throughout the shoot. The plates are staged
against a white poster board backdrop and on top of white poster board or tablecloth. To aid in
judging the size of the portion it is helpful to have standard utensils in the picture as well.
The use of a digital camera is recommended since this allows for the immediate confirmation of
photo quality photos, and eliminates the need for scanning traditional photographs. The digital
camera is mounted on a tripod and remains in the same position throughout the photo shoot for
consistency in angle, lighting, and ease of setup. If possible, the lighting is adjusted to eliminate most
shadows and provide maximum illumination. A tape measure is included in photographs of seafood
portions so that photographs can be scaled to actual size.
After photographs of the preparation models are taken, the digital images are resized and color
corrected using standard photo publishing software such as PhotoShop, Corel Paint, MS Paint,
iPhoto, etc.) To resize the photos, the supervisor should set up (define) a crop that best defines the
area of the largest plate and the tape measure with a little white space surrounding it. This crop area
should be used to crop all images. Keeping the same crop dimensions will force the smaller servings
to be proportioned correctly. Color correction is done as necessary. Most applications have an
AUTOADJUST feature to simplify things. The images should be saved as medium resolution .jpg
files (150dpi). The file images should be named with brief titles that relate to content (ex. Clam_raw,
Clam_cooked). The images are then imported into a Microsoft Power Point presentation or other
some other layout program. The photos can then be scaled to actual size in Power Point, by drawing
a 1" line segment on the same slide as a photo, followed by scaling the photo until a 1" increment on
the photographed tape measure was equal in length to the 1" line segment.
Booklet Assembly
To facilitate editing and rearranging of images, it is recommended that each fish species and each
preparation model is formatted onto its own individual PowerPoint slide. All images are printed on
its own page. Pages should be formatted for printing in a landscape format since this allows the
D-17
-------
most space for the image. Images should be printed on a laser color printer. Printing from an inkjet
color printer will not produce sufficiently clear images. For durability, lamination of the pages is
recommended. Depending on the available resources the images can be produced on glossy photo
quality paper. However, this increases the cost and overall weight of the product. Pages can be
three-hole punched for assembly into a three-ringed binder or bound in spiral notebooks.
Labeling and Arranging Photos
In formatting and assembling the booklet, it is important to consider the ease of administration and
the facilitation of the flow of the CAPI interview. It is recommended that the images be arranged in
the sort order designated by the supervisor during the CAPI configuration. Supervisors can use the
GroupID and the SpeaesID in the CAPI configuration to assemble the Species Identification section
of the booklet. They may choose to put captions with the Species Description defined in the CAPI
or codes on the photographs such that they are easily identifiable and match up with the Library.
Supervisors can use label the photographic models with the ModeHD or the Model Description.
Since the CAPI interview has several sections that refer to same photographs throughout the
interview, interviewers will have to flip back and forth between pictures. Therefore, it is necessary to
prepare two booklets: one for species identification and one for the preparation models.
D-18
-------
The Fish Consumption Survey Tool (FCST)
CAPI Survey
Description of the Supervisor Software
Overview of Survey Structure
The computer assisted personal interview (CAPI) software is based on a hard-copy consumption
survey developed for the Suquamish Nation.
The CAPI asks questions about consumption of individual species of fish and shellfish as well as
consumption of groups of fish and shellfish. Fish and shellfish are grouped on the basis of
similarities in behavior and habitat (e.g., migratory species, marine bottom dwelling fish, freshwater
predatory fish, saltwater shellfish). Questions that capture information at the group level minimize
Respondent burden and avoid the collection of data with little variance and added value during
analysis that will contribute to an assessment of health risks. For example, Respondents are asked to
consider all types of salmon consumed when providing catch area locations. They are not asked to
provide locations for each species of salmon consumed since salmon are grouped together due to
similar behaviors and habitats and thus are most likely be found in similar areas that would share
contaminant concentrations. At the end of different survey sections, an opportunity is given to
record consumption of seafood species that were not included in any of the groups.
The survey consists of several section types:
1. A 24-hour recall section records a Respondent's seafood consumption during the day prior
to the survey.
2. The seasonal consumption section records consumption of seafood over the course of a
year. The seasonal consumption rate section of the survey proceeds by asking questions to
establish yearly consumption rates for individual species within a group prepared in various
forms. Questions on individual species are then followed with general summary questions
about all of the species consumed within that group (e.g., Where are species in that group
harvested from; What parts of organisms from the group are consumed?).
3. The children's consumption section allows male and female Respondents with children less
than six years of age to record their children's seafood consumption. Information captured
includes a child's age, weight, breast feeding history as well as seafood consumption patterns.
Questions about children's finfish consumption are asked on the basis of seafood groups
rather than individual species. Shellfish consumption is asked for individual species.
4. The general section of the survey records information on how a Respondent's seafood
consumption has changed over time, consumption of seafood at special gatherings or
celebrations, and general information such as approximate age, income, body weight, and
height.
D-19
-------
Introduction to the CAPI
The Seafood Consumption Survey is administered by computer assisted personal interview (CAPI)
software developed in MS Access 2010. Most of the screen shots included in this Guide reflect the
survey as configured by EPA for the Quinault Nation in March 2006.
The software has been designed so users can specify their own version of the CAPI. Versions may
vary due to definition of seafood categories or groups, species included in a group, allowable harvest
and catch locations, as well as text that further clarifies a question asked by an Interviewer to a
Respondent. A user can have multiple interviewers conducting the same version of the CAPI in the
field.
Each study will have one designated Supervisor's PC. This Guide will refer to the user or users of
this PC as the Supervisor. The Supervisor Module allows a Supervisor to configure the CAPI
software specifically for his/her study. The configured version of the CAPI is then deployed to a
USB device (also known as a flash-drive or thumb-drive). The USB device is intended to be used for
CAPI installation on Interviewer PCs. An Interviewer PC is referred to as a site in this Guide. The
Supervisor Module also allows for the accumulation of data collected at multiple sites, once the
interviewing phase is complete, into a single database on the Supervisor's PC. The functionality and
use of the Supervisor Module is described in this Guide.
Computer Requirements
The designated Supervisor computer must have a Ventium III or hipher processor. It must have at least 128
MB of memoiy thowh 256 MB is preferable. The hard disk should have at least 200 MB available pace. The
computer's operating system should be:
Windows 7. MS Access 2010 or higher should be installed on the computer prior to installing the
Supervisor Module software.
Setup
To install the Supervisor Module software, follow these steps:
1) Download the Fish Consumption Survey Tool (FCST) zip file available on EPA's website at
https:/ /www.epa.gov/fish-tech/epa-guidance-developing-fish-advisories.
2) Once downloaded, click on Supervisor Seafood.exe and follow the steps to install the
"Welcome to the Supervisor Seafood Setup Wizard."
The installation will copy the necessary software files to the local folder C:\SupervisorSeafood
unless otherwise specified during installation. The files copied are: four Access 2010 databases of
which two (Supervisorseafood_dat.accdb and SeafoodSurvej_dat.accdb) contain the structure of the data
collected during interviews and lookup tables (i.e. tables that contain possible answers to interview
questions) and the other two (Supervisorseafood_app.accdb and SeafoodSurvej_app.mdb) contain
application code and objects such as queries, forms, and reports. Also copied is the file and
Install.msi that will be used by the Supervisor Module for deploying to and accumulating data from
D-20
-------
the sites. A shortcut icon is copied to the Supervisor PC's desktop that will be used to initiate the
module. Under the folder C:\ \Supervisor Seafood, the installation procedure will create four
subfolders. The Export folder will contain a zipped copy of the Ssafood_dat.accdb file for each
deployed site. The Import folder will contain an unzipped copy of the S'eafood_dat,accdb file for each
imported site. The Translate folder will contain a copy of the Seafood_dat.accdb file for each site after
the "Other Specified" seafood items have been translated to an item found in the library (more on
this later in this appendix). Items not asked about during an interview but reported by the
Respondent must be translated before integrated with other site data. These databases in the
Translate folder are integrated into the master database of all sites with completed interviews in
preparation for analysis.
Once the installation program has completed the files shown in Figure 1 will be in the
C:\SupervisorSeafood folder.
Figure 1. Folder structure after installation
The Supervisor Module
During installation an icon was copied to the Supervisor's PC desktop (Figure 2).
g
s
s
IM
iupervisor
ieafood
Figure 2. Desktop icon
Click on the icon to initiate the software. Click on "Yes" if prompted to allow the executable to
install the software. When the Supervisor's Module is started for the first time, a prompt to "Enable
content" will appear (Figure 3).
D-21
-------
! Security Warning Some active content has been disabled. Click for more details. Enable Content
Macro Single Step
I; ? || S3 ]
Macro Name:
Step
AutoExec
Condition:
Stop All Macros
Continue
Action Name:
Error Number:
RunCode
[2001
Arguments:
SetGlobalDB 0
Figure 3: Enable Content prompt
Select the Stop All Macros button. Then select the Enable Content button in the yellow header.
This step should not have to be followed after the first instance of starting the Supervisor software.
You will then be prompted for the Study number and name(Figure 4). The Study number allows
for more than one study to utilize the same CAPI version. For example, a user may be planning on
two studies, one of licensed anglers and one of the general population. Provide the name of the
organization conducting the study (e.g., the State of Maryland, the Quinault Nation). Then provide
the geographic region, area, or water body for which the study is interested in identifying harvest and
catch areas (e.g., Puget Sound).
~D Study Information . ^ .
Study Number:
r
Organization Name:
1
Interest Catch Area:
1
OK Cancel
/
Figure 4. Study number and name prompt
D-22
-------
Once this screen has been completed and the OK button pressed, a screen with a menu at the top
will be displayed (Figure 5).
SEAFOOD CONSUMPTION
SUPERVISOR MODULE
Figure 5. Supervisor Module's main menu
There are two tabs at the top: 1. File and 2. SCSM Menu. The file tab is the main functional menu
for Access 2010. The working tab is the SCSM Menu tab.
The SCSM Menu has six items: File, Library; Configure, Integrate, Reports, and About. Exit is
found under the File tab as well as under the File submenu.
To exit the application, choose the Exit item on the File tab or the File submenu. The Windows
exit control button L^l that appears in the top right hand corner of the screen has been disabled to
prevent abnormal shutdown that might affect the integrity of the database.
Question Library
Select the Library menu item and then select Questions submenu to add or modify the library of
questions (Figure 6). The library has been populated with records defined by EPA. Again, the
screen shots of the library included in this Guide contain seafood used in the Quinault Tribal
Consumption Survey.
y Seafood Consumption Supervisor Module
Exit
Library Configure
Integrate Reports About
Questions
Figure 6. Library menu item
This menu item is only available if the CAPI configuration has not been finalized by the Supervisor.
The library is the master list of species that can be selected during configuration to be included in
the CAPI (Figure 7). Additions or modifications should be done by an experienced Access
programmer.
D-23
-------
Select the Questions menu item to view the library.
(A field can be sorted by clicking on its column header)
SpeciesID | Species Description | IncludeText | Migratory | Latin Name
]
King Salmon
Chinook
r
2
Sockeye
Red
r
3
Coho
Silver
r
4
Churn
Dog
r
5
Pink
r
6
Steelhead
r
7
King Salmon
Sockeye, Coho, Chum, King, Pink,
r
8
Smelt
r
9
Herring
r
10
Cod
Rock, Tom
r
11
Perch
r
12
Pollock
imitation crab and other imitation se
r
Search Fields
Select Field to Search on: (* Species ID Species Description C Include Text
Enter a Search: Search Reset
Add
Edit
Record: H \ i 11 T ~ I H [~*! of 62
Figure 7. Species question library
Each record in the library has four fields associated with it that make the record unique:
1) SpeciesID is a unique numeric identifier assigned to each record. When adding a new record to the
library, the system will calculate the next available number. This number, however, can be modified
though it cannot duplicate an existing number. This ability to modify the SpeciesID allows a
Supervisor to add a new record with a number (and presumably text) so that identical records can be
maintained across studies.
Each CAPI consumption question or sets of questions are linked to a SpeciesID record selected by
the Supervisor during configuration.
2) SpeciesDescription is the text that describes the species to be asked about in the CAPI. It is the text
that is incorporated into the Interviewer Prompt in the CAPI. "King or Chinook Salmon" is an
example of a SpeciesDescription.
3) IncludeText is text that further explains the species to be asked about in the CAPI. The
SpeciesDescription "Salmon" may be asked about by itself or may be asked so that the Respondent
considers specific types of salmon as well when answering the one question (Figure 8). This field
might also be used to refer to common names of the SpeciesDescription that may be more recognizable
to the survey population.
D-24
-------
(A field can be sorted by clicking on its column header)
SpeciesID
Species Description
IncludeText
King or Chinook Salmon
Chum or Dog Salmon
Pink Salmon
Sockeye or Red Salmon
Coho Salmon
Salmon
Migratory | Latin Name
r
Sockeye, Coho, Chum, King, Pink,
Select Field to Search on:
-Search Fields
C SpeciesID (* Species Description f IncludeText
Enter a Search: salmon
Search
Reset
Add
Edit
Record: H |
1 ~ I H I I of 6
Figure 8. Salmon records in question library
While the columns are fixed length, the full text can be seen using the arrow keys when the field is
selected.
Typically questions with identical SpedesDesmption values but different Incln0Text values are selected
during configuration to be included in different seafood frequency sections of the CAPI. The
Salmon record, SpeciesID "I" if chosen by the Supervisor during configuration to be asked the 24-
Hour Recall section will appear as in Figure 9.
3) PART ONE; DIETARY INTAKE - 24 HOUR RECALL
S3
Subject IP 1 [574-00-12345-1
Interviewer Prompt
Did you eat any King Salmon such as Chinook yesterday from the time you woke up until the time you went to sleep for
the night? Please include at meals and as snacks.
|Interviewer Instructions
Display the image of King Salmon.
[rT
Eat King Salmon?
Previous
"0
Yes
NO
-7 (Prefer not to say)
-8 (Pont know)
Figure 9. SpeciesID "1" was chosen to be included in the Recall section
D-25
-------
The Salmon record, SpeciesID "1", if chosen by the Supervisor during configuration to be asked in
the seasonal rates section will appear as in Figure 10.
JlPART TWO: FINRSH CONSUMPTION - SEASON, FREQUENCY, PORTIONS
|Subject ID: ~| |574 00-12345 1
Interviewer Prompt
Interviewer Instructions!
In the past 12 months, did you ever eat King Salmon such as Chinook?
Display the species image of King Salmon.
|T1A | |Eat King Salmon?
Yes
No
-7 (Prefer not to say)
-8 (Don't know)
Figure 10. SpeciesID "1" was chosen to be included in the Rates section
4) Migratory specifies whether a species is considered migratory or resident of the region. The
SpeciesDescription and InclndeText could be identical across two records but one species could be
migratory and the second one not. This allows for a species, such as "Salmon," that would be
considered migratory to one geographic region but in a different location might be lake locked and
unable to migrate. The differentiation between records allows for more specific statistical analyses
leading to better estimates.
As explained previously, when the Library sub-menu item is chosen, a library form window
Questions will be displayed (Figure 11). As is indicated in the title of the library window, you can
sort one column in ascending order by clicking on the column header of a text field. The default sort
order is by SpeciesID. Any one of the first three columns can be searched by completing the bottom
third of the form. To find questions that reference the species "salmon", for example, choose the
SpeciesDescription field and type "salmon" in the text field underneath. The search is not sensitive to
case.
(A field ca
be sorted by clicking on
ts column header)
| SpecieslD|
Species Description
IncludeText | Migratory!
Latin Name
LJ
D
King or Chinook Salmon
r
_l
2
Chum or Dog Salmon
r
3
Pink Salmon
r
4_
Sockeye or Red Salmon
r
I 5
Coho Salmon
r
6
Migratory Finfish
Salmon V
7
Salmon
Sockeye, Coho, Chum, King, Pink, 1
8
Steelhead
r
9
Surf Perch
r
10
Anchovies, Smelt, Sardines, He
r
11
Pelagic Finfish
S urf Perch, Anchovies, S melt, S arc V
12
Sole/Flounder
r
o ur u
| Select Field to Search on:
Species ID C Species Description
C IncludeText
| Enter a Search: |
S„,ch 1
Reset |
1 Add H Edit ¦ Close |
¦ J
|| Record:
li i ~ 1 1
J of 62
Figure 11. Searching for Salmon questions
D-26
-------
Press the Search button to start the search. Resulting records containing the word "salmon" will be
displayed (Figure 12).
(A field can be sorted by clicking on its column header)
1
|| SpeciesID| Species Description
| IncludeText
Migratory | Category
Preparation |
D
King Salmon
Chinook
17
7
King Salmon
Sockeye, Coho, Chum, King, Pink,
r
44
Salmon and steelhead
r
57
Salmon, including Steelhead
r
G3
King Salmon
r
64
Dried Salmon
Chinhook
r
Dried
Select Field to Search on:
-Search Fields
Species ID Species Description Include Text
Enter a Search: salmon
Search
Reset
Add
Edit
Record: H
1 ~ I H I I of 6
Figure 12. Search results for "salmon"
To display all the records again or to begin a new search, press the Reset button.
If the Supervisor has not published a final version of the CAPI, he/she may add a new record to the
Species Question Library. To add a record, press the Add button to bring up the Add Species
Question window (Figure 13).
Species Question
SpeciesID | S"
Species Description |
IncludeText |
Migratory I-
Latin Name |
Species Comments:
1 Save and Close
Cancel
Figure 13. Adding a new species question to the library
D-27
-------
The system will assign the next SpeciesID number though this can be modified to a number that does
not already exist in the library. An error message will appear if you try to save a record with a
SpeciesID that already exists. Press the Save and Close button to save the new record or press the
Cancel button to not save the record and return to the library form (Figure 14).
Species Question
SpeciesID
Species Description
IncludeText
Migratory
Latin Name
Species Comments:
63
|A new fish
| include text]
r
Save and Close
Cancel
Figure 14. Adding a new record to the Species Question Library - a completed form
If the Supervisor has not published a final version of the CAPI, he/she may edit an existing record
to modify the Species Description, IncludeText., Migratory, Category, and Preparation fields. Use the mouse
to highlight a record. Press the Edit button to open up the edit window (Figure 15).
Species Question
SpeciesID
Species Description
IncludeText
Migratory
Latin Name
Species Comments:
7
|Surf Perch
r
I
Save and Close
Cancel
Figure 15. Editing an existing library record
D-28
-------
Press the Save and Close button to save the edited record or press the Cancel button to not save
the edit and return to the library form.
On the library form, press the Close button to return to the main menu.
Configuration
Select the Configure menu item and the Datasheets submenu item to configure your version of
the CAPI (Figure 16).
Configure Integrate
Publish
Figurel6. Configure menu
To configure the software to your specific survey specifications, you will be taken through seven
screens to customize the survey species questions. Use the Next and Back navigation buttons at the
bottom of each of screen to move to the next screen or move back to the previous screen.
You will be asked to modify seven data tables that will define and control the questions that will be
asked during the CAPI. For each of the tables you will be presented with a datasheet that is made up
of rows and columns. Similar to a standard spreadsheet, each column represents a field of one type
of information and each row represents an entry. Use the mouse or tab and arrow keys to navigate
within the datasheet both vertically and horizontally. The right-facing arrowhead in the first column
will indicate the row on which you are currently focused on. To delete a row, position the mouse in
the left most column of the row (the whole row should be highlighted) and press the delete key.
Sometimes, the text may appear to be truncated. This is because of the column width on the form.
To increase the column width, position the mouse on the right box of the column header and drag it
to the right to increase the display area. Note, however, increasing a column width will affect how
many fields can appear on a page.
Other relevant navigation controls and indicators are indicated on the screen shown in Figure 17.
D-29
-------
e STEP TWO: Define Species
SEAFOOD CONSUMPTION CONFIGURATION WIZARD
Column
Species Description
he udeText
Migrator,
SpeciesID
Group ID
Number of
ays In Seas
jj King or Chinook Salmon
Chum or Dog Salmon
Pink Salmon
Sockeye or Red Salmon
Coho Salmon
Migratory Finfish
Salmon
Steelhead
Surf Perch
Anchovies Smelt
Pelagic Finfish
Sole/Flounder
Sturgeon
Row or
Record
Salmon
Sockeye. Coho Chum. K
Sardin
Surf Perch Anchovies Si
Record
Number
Record:
Exit
Configuration
Move to
Move to
First
Previous
Record
Record
Move to
Move to
Insert
Next
Last
New
Record
Record
Record
Horizontal
Scroll Bar
Figure 17. Datasheet layout
The first datasheet screen (Figure 18) asks you to specify the groups for the fish and shellfish
sections. Group assignment provides a means to organize fish according to similar characteristics.
The group identifier is used by the CAPI to determine the order in which questions are asked within
a section. (See the "Overview of Survey Structure" for further discussion of groups and how they
are used.)
01 STEP ONE: Define Groups
SEAFOOD CONSUMPTION CONFIGURATION WIZARD
Group ID
Group Description
Rate Section
~
Record: M | [
1 I H | I of 1
Cancel
Back
Figure 18. Group description
D-30
-------
Groups should contain similar seafood items.
Group Description is limited to 20 characters. Under the Rate Section column, choose which Rate
Section of the Survey (Finfish, Shellfish, Other) applies to the species assigned to each group ID
(Figure 19).
efine broups
SEAFOOD CONSUMPTION CONFIGURATION WIZARD
Group ID
Group Description
Rate Section |
~
A
Migratory
Finfish^^^^^H
*
Finfish
Shellfish
Other
Record: M
Exit
Back
Next
Figure 19. Rate Section designation
Respondents are asked to provide seasonal rates for Finfish but not for Shellfish. The body parts
eaten are also asked differently for Finfish and Shellfish; Finfish parts are asked at the group level
while Shellfish parts are asked at the species level. Finfish parts are pre-defined in the CAPI while
Shellfish parts to be asked about are specified during this configuration process in a later datasheet.
Finfish and Shellfish species types should always be mutually exclusive. For the example survey, the
groups were defined as shown in Figure 20.
D-31
-------
Si STEP ONE: Define Groups
SEATOOD CONSUMPTION CONFIGURATION WIZARD
Exit
Record: H | j
I
Group ID
Group Description
Rate Section
~
s
Migratory Finfish
Finfish
B
Pelagic Finfish
Finfish
c
Elttin Feeding Finfish
Finfish
D
Bttm Pelagic Finfish
Finfish
E
Freshwater Finfish
Finfish
F
Freshwater Shellfish
Shellfish
G
Marine Shellfish
Shellfish
H
Or Marine Finfish
Finfish
1
Or Frshwater Finfish
Finfish
J
Or Marine Shellfish
Shellfish
*
1 ~ H ~* of 10
Back
Next
Figure 20. Definition of groups in Quinauit Survey
Again, the Group ID determines the order in which species belonging to a group are asked in die
CAPL Within the Rate Section specified groups are asked in alphabetical order. In Figure 20 species
questions belonging to Group "A" will be asked before "B" which will be asked before "C" and so
i >n. In the "Other" Rate Section Group "P" questions will precede Group "G" questions.
Press the Next button to advance to the next datasheet screen.
The second datasheet screen (Figure 21) will ask you to define which species will be asked in the
four different sections of the CAPI (i.e. 24-hour recall, seasonal tin fish and annual shellfish, child,
and gatherings or special events) that ask about consumption.
In die SpetiesDesciiption column press the downward facing arrowhead to the right of the column to
display records from the question library. They are sorted alphabetically. The drop down box
displays both SpetiesDesmption and InckdeText fields.
D-32
-------
"
S^STEP TWO: Define Species
SEAFOOD CONSUMPTION CONFIGURATION WIZARD
1
SpeciesID Group ID Species Description
IncludeText
Migratory Number of Days In Seas |
Anchovies, Smelt, Sardi
Black Crappie
Bottom Dwelling Finfish
Bottom Feeding Finfish
Bull Cod or Cabezon
Butter Clam
Chiton
| Chum or Dog Salmon
Cod, Greenlings, Rockfish, Pollack
Sole, Flounder, Sturgeon, Halibut
Record: M ^
60 ~ I H MH of 60
J I
Exit
Back
Next
Figure 21. Species defined in the Question Library appear in drop down box
You can type ahead to find the species. For example, to find "Steelhead", typing "st" will bring you
to "Steelhead" line. To search the SpeciesDescription and IncludeText by keywords, press the F5 key. A
screen similar to the Question Library search screen (Figure 12) will be opened. Use the search
functionality explained in the Question Library section, select the record containing the desired
species, and press OK to return to the second configuration datasheet, "Define Species". The
selected species will be filled in. Or press Cancel from the Question Library form to return to the
second datasheet without selecting an item.
Note, any changes made to the corresponding record in the Question Library will not be reflected in
a SpeciesID already selected on this datasheet. For example, if after you select Speciesld "60", as in
Figure 22, you decide you want to designate that fish as migratory. You must edit the record for
Speciesld" 1" in the Question Library, come back to the second datasheet, delete the old Speciesld "1"
record, and add the modified Speciesld "1" to the datasheet to pick up the new value of migratory.
Once you select the species, the fields SpeciesID, Species Description, IncludeText, and Migratory are filled
in by the system. These cannot be modified. Next, select the appropriate group in the Grotp ID
column by clicking your mouse on the downward facing arrowhead to the right of the column.
D-33
-------
STEP TWO: Define Species
SEAFOOD CONSUMPTION CONFIGURATION WIZARD
SpeciesID
Group ID
Species Description
IncludeText
Migratory
Number of Days In Seas
-
57 J
Sea Anemone
~
58 J
Sea Urchin
~
59 J
Sea Cucumber
~
J
60
I
Other Shellfish
~
*
A
Migr
ajJ&ry Finfish
H
B
Pelagic Finfish
C
Bttm Feeding Finfish
D
Bttm Pelagic Finfish
E
Freshwater Finfish
F
Freshwater Shellfish
G
Marine Shellfish
H
Or Marine Finfish
J
-
Record: H | i
||
M ~ 1 M !~*! of 59
-------
r
P STEP TWO:
Define Species
¦
1 SEAFOOD CONSUMPTION CONFIGURATION WIZARD
1
SpeciesID
Group ID
Species Description
Recall
Recall Sort Order
Rate Tables
Rate Tables Sort Ord
1
A
King or Chinook Salmon
~
0
2
A
Chum or Dog Salmon
~
0
3
A
Pink Salmon
~
0
4
A
Sockeye or Red Salmon
~
0
5
A
Coho Salmon
~
0
6
A
Migratory Finfish
~
~
7
A
Salmon
0
1
~
8
A
Steelhead
~
0
9
B
Surf Perch
0
1
0
10
B
Anchovies. Smelt Sardin
0
2
0
11
B
Pelagic Finfish
~
~
12
c
Sole/Flounder
0
1
0
13
c
Sturgeon
0
2'
0
zi
1 | Record: M ] 4 | \
IT ~ I M !~*! of 59 < | | [\ ~
Exit
Back
Next
Figure 23. Horizontally scrolling to the right reveals Recall fields
Use your mouse key or tab to move to the box and hit the space bar so a checkmark appears in the
checkbox. The checkmark will indicate that the item will be asked in the corresponding section of
the survey. After each section's corresponding checkbox there is a field where you specify a number
that will determine the order in which an item will be asked in each section of the survey within the
GroupID field.
a
0 STEP TWO:
Define Species
SEAFOOD CONSUMPTION CONFIGURATION WIZARD
1
SpeciesID |
Group ID
Species Description
Recall
| Recall Sort Order
Rate Tables |
Rate Tables Sort Ord -*¦
4
A.
Sockeye or Red Salmon
~
0
|
5
A
Coho Salmon
~
0
6
A
Migratory Finfish
~
~
_l
7|
A
Salmon
0
1
~
8
A
Steelhead
~
0
~
9|
B
Surf Perch
0
1
0
10
B
Anchovies Smelt Sardin
0
2
ra
11
B
Pelagic Finfish
~
~
12
C
Sole/Flounder
0
1
0
13
C
Sturgeon
0
2
0
14
C
Halibut
0
3
0
15
C
Bottom Feeding Finfish
~
~
16
D
Cod
0
1
0
'1
Record: M | 4 || § ~ | M !~&( of 59 < | ~
Exit
Back
Next
Figure 24. Selected questions to be asked in Recall Section
For example, in the Quinault Survey, Halibut, a bottom-feeding fish (Group "C"), is asked about in
the 24-hour Recall (Figure 24). One food item entry or row in the datasheet will denote this. By
checking off the Recall checkbox, the CAPI software will ask about eating Halibut in the section. A
value of 3 in the Recall Sort Order column directs the software to ask about eating Halibut in die past
D-35
-------
24-hours after Sturgeon, the second Group "C" fish asked about. Sole/Flounder is the first fish in
Group "C" asked about in the 24-hour Recall.
If you scroll to the right, the first three columns again are frozen (Figure 25). Now, you can see
Sole/Flounder is first Group "C" fish asked in the rate consumption table, Sturgeon, the second and
Halibut the third.
BE STEP TWO: Define Species
SEAFOOD CONSUMPTION CONFIGURATION WIZARD
SpeciesID | Group ID
Species Description
Rate Tables
Rate Tables Sort Order
Gatherings
Gatherings *
4 A
Sockeye or Red Salmon
0
4
~
5 A
Coho Salmon
0
5
~
6 A
Migratory Finfish
~
~
_l
7 A
Salmon
~
0
8 A
Steelhead
0
6
~
9 B
Surf Perch
0
1
~
10 B
Anchovies. Smelt. Sardin
0
2
~
11 B
Pelagic Finfish
~
~
12 C
Sole/Flounder
0
1
~
13 C
Sturgeon
0
2
~
14 C
Halibut
0
3
0
15
C
Bottom Feeding Finfish
~
~
16
D
Cod
0
1
~
¦J
Record: M |
1 ~ I M !~*! of 59 <1 1
Exit
Back
Next
Figure 25. Halibut will be the third bottom-feeding fish asked in Consumption Rate section
Scrolling horizontally reveals that consumption at special events and gatherings of only Halibut, not
Sole/Flounder and Sturgeon, will be asked about in the last section of the CAPI since the box it is
the only one checked off in the Gatherings column (Figure 26).
B§ STEP TWO: Define Species
SEAFOOD CONSUMPTION CONFIGURATION
SpeciesID
Group ID
Species Description
Rate Tables Sort Order
Gatherings
Gatherings Sort Order
o
4 A
Sockeye or Red Salmon
4
~
I
5 A
Coho Salmon
5
~
6
A
Migratory Finfish
~
_l
7
A
Salmon
0
1
8
A
Steelhead
6
~
9
B
Surf Perch
1
~
10
B
Anchovies Smelt Sardin
2
~
11
B
Pelagic Finfish
~
12
C
Sole/Flounder
1
~
13
C
Sturgeon
2
~
J
14
c
Halibut
3
m
I 2
15
c
Bottom Feeding Finfish
~
16
D
Cod
1
~
Record: M | i 11 14 ~ | M |>*| of 59 < | | ~ |
Exit
Back
Next
Figure 26. Halibut asked about during the Gathering and Special Events section
D-36
-------
The four sections of the survey that ask about consumption address consumption of species in
different ways. Looking at Figure 27 at salmon for example, you can see the 24-hour recall portion
of the survey ascertains consumption of similar species as a collection, while the seasonal
consumption portion of the survey generally ascertains consumption at the individual species level.
In the 24-hour recall portion of the survey, collective salmon consumption as indicated by SpedesID
"7" is asked about. See Figure 9 to view the resulting Interviewer prompt. While in the seasonal
consumption portion of the survey, salmon consumption is recorded on a species specific basis.
There will be six sets of seasonal consumption questions, one for SpedesIDs "1" - "6".
eg.
>TEP TWO: Define Species
¦
| SEAFOOD CONSUMPTION CONFIGURATION WIZARD
SpeciesID |
Group ID
Species Description
Recall
Recall Sort Order
Rate Tables
Rate Tables Sort Ord *
1
a d
King or Chinook Salmon
~
0
2
A
Chum or Dog Salmon
~
0
3
A
Pink Salmon
~
0
4
A
Sockeye or Red Salmon
~
0
5
A
Coho Salmon
~
0
6
A
Migratory Finfish
~
~
7
A
Salmon
0
1
~
8
A
Steelhead
~
0
9
B
Surf Perch
0
1
0
10
B
Anchovies Smelt Sardin
0T"
2
0
11
B
Pelagic Finfish
~
~
12
C
Sole/Flounder
0
1
0
13
C
Sturgeon
0
2
0
^r1
1 | Record: 14 I i
1 ~ | H |>*| of 59
4
I
Exit
Back
Next
Figure 27. Different Salmon SpecieslDs
Please note that finfish is asked at the group level in the child section and therefore should be:
configured appropriately in this datasheet. This design matches the Suquamish Nation survey of
which this CAPI automated. To represent this in the table, check one box under the Child column
for each collective group description (e.g., Migratory Finfish for Group ID A) to be asked as shown
in Figure 28. Only Groups designated to be asked in the Finfish or Shellfish Rate Sections can be
asked in the child section of the CAPI.
D-37
-------
'
Si STEP TWO: Define Species
1 SEAFOOD CONSUMPTION CONFIGURATION WIZARD
L
SpeciesID
Group ID
Species Description
Child
Child Sort Order|
3
A
Pink Salmon
~
4
A
Sockeye or Red Salmon
U
5
A
Coho Salmon
u
6
A
Migratory Finfish
0
1
7
A
Salmon
~
8
A
Steelhead
~
9
B
Surf Perch
~
10
B
Anchovies Smelt Sardin
~
11
B
Pelagic Finfish
0
1
12
C
Sole/Flounder
~
13
C
Sturgeon
~
14
c
Halibut
u
15
c
Bottom Feeding Finfish
0
1
Record: M | 1 ~ | H !~*) of 59
aJ
Exit
Back
Next
Figure 28. Selected finfish items asked in child section
Shellfish items asked in the Child Section are identified in Figure 29. There is no requirement to
limit shellfish species in the Child Section to one Species ID per Group ID.
HI STEP TWO: Define Species
SEAFOOD CONSUMPTION CONFIGURATION WIZARD
SpeciesID
Group ID
Species Description
Child
Child Sort Order|
24
F
Freshwater Shellfish
0
1
25
G
Dungeness Crab
0
1
26
G
Red Rock Crab
0
2
27
G
Mussels
0
3
28
G
Clams
0
4
29
G
Razor Clams
0
5
30
G
Pacific Oyster
0
6
31
G
Crab
~
32
H
Bull Cod or Cabezon
~
33
H
Mackerel
~
34
H
Tuna canned
~
35
H
Tuna fresh
~
36
H
Shark
~
Record: M I
II
l ~ | M |>*| of 59
Exit
Back
Next
Figure 29. Shellfish asked in child section
Press the Next button to proceed to the next configuration screen.
On the third datasheet screen (Figure 30) you itemize the photographs that will be used when
conducting the interview to capture portion serving size of the item prepared a certain way.
D-38
-------
Photographs will be included in a booklet for the interview. Each photograph represents a standard
portion size of a seafood item presented in a certain form such as a filet, as part of a soup, or fritters.
The Respondent reports the amount he or she normally eats relative to the serving size represented
in the picture (e.g., half the displayed amount, the displayed amount, twice the displayed amount).
Each survey needs to develop a set of photographs for use in their survey. Guidance for developing
the booklet is included with the FCST. Examples are included in the supplemental Microsoft
PowerPoint file "FCST Example Species and Portion Size Photographs.ppt."
Enter one row for each unique photograph to be used during an interview. A serving size
photograph can be used for multiple species. A species may be associated multiple photographs as
defined on the next configuration datasheet.
The Photo ID provided will uniquely identify a photograph of a portion of a seafood item prepared in a
certain form in the database. The Preparation Form Type and Portion Description provided should be
helpful to the Interviewer and Respondent so as to select an appropriate photograph that best
represents the amount and form consumed. The Raw Gram Wt will be used in the statistical reports to
describe consumption amounts. Note the Cooked Weight is reserved for future use.
r-
81^5
>TEP THREE: Define Portion Preparation Forms
~
SEAFOOD CONSUMPTION CONFIGURATION WIZARD
1
Photo ID
Preparation Form Type
Portion Description
I Raw Gram Wt
Cooked W1
A
~
>:t:a Container offish Half pint jar 326
CRAY
Crayfish
6 crayfish tails
101.64
CRBD
Cooked meat Dungeness Crab
8 oz cooked meat
226 796
CRBM
Cooked meat from one crab
1 crab
382
CUPC
Cup of clam chowder
1 Cup
45 3
CUFF
Cup offish stew
1 Cup
136
-
FLT
Fillet
3 oz baked fillet
226 796
FRTR
Fritters
3 Fritters
52.569
MAN C
Steamed manila clams
6 steamed clams
98 2
MUSS
Steamed Mussels
6 steamed mussels
103 916
OYST
Raw Oysters
6 raw oysters
41 139
R2CF
Fried razor clams
6 fried razor clams
85.932
SCLP
Sauteed scallops
6 sauteed scallops
238.674
Record: M | | | 1 ~ | H |>*| of 20 i ~
Exit Back Next
Figure 30. Defined portion models for Quinault CAPI
Press the Next button to proceed to the next configuration screen.
The fourth datasheet screen (Figure 31) is where you select which of the photographs defined in the
previous datasheet screen are valid for the species questions included in the CAPI. In the first
column, select the species for which you would like to associate a photograph. The drop down box
will display the SpeciesDescription and SpeciesID.
D-39
-------
HI STEP FOUR: Assign Preparation Forms to Species
SEAFOOD CONSUMPTION CONFIGURATION WIZARD
Species ID
Photo D
Kina or Chinook Salmon
Chum or Dog Salmon
Pink Salmon
Sockeye or Red Salmon
Coho Salmon
Migratory Finfish
Salmon
Steelhead
58 1 H I 1 of 53
Record: MM
Next
Figure 31. Selection of Species ID in first column
In flu: second column, select the portion photograph to be assigned to each species (Figure 32).
You can assign multiple photographs to a species; each Species ID/Photo ID combination will be on
its own row. Preparation Farm Type, Photo ID and Portion Description are displayed in the drop down box.
STEP FOUR: Assign Preparation Forms to Species
SEAFOOD CONSUMPTION CONFIGURATION WIZARD
Species ID
King or Chinook Salmon
Record: \i
Photo ID
Container offish
COFI
Half pint jar1®
Crayfish
CRAY
6 crayfish tails
Cooked meat Dungeness Crab
CRBD
8 oz cooked meat
Cooked meat from one crab
CRBIvl
1 crab
Cup of clam chowder
CUPC
1 Cup
Cup offish stew
CUFF
1 Cup
Fillet
FLT
8 oz baked fillet
Fritters
FRTR
3 Fritters
53 ~ I H !~*! of 58
Exit
Back
Next
Figure 32. Select photograph in second column
In Figure 33, three photographs are assigned to Chum or Dog Salmon. The Respondent will be
asked three questions if they respond that they eat Chum or Dog Salmon: if he or she eats Chum or
Dog Salmon from 1) a container, 2) as a fillet, and 3) as a steak. The three photographs represent a:
standard portion size of these three prepared forms. During the CAPI, the interviewer will be able
to record answers that reflect the Respondent's inability or unwillingness to select a portion model.
D-40
-------
81 STEP FOUR: Assign Preparation Forms to Species
1 SEAFOOD CONSUMPTION CONFIGURATION WIZARD
Species ID
Photo ID
King or Chinook Salmon
Container offish
King or Chinook Salmon
Fillet
King or Chinook Salmon
Steak
Chum or Dog Salmon
Container offish
Chum or Dog Salmon
Fillet
Chum or Dog Salmon
Steak
Pink Salmon
Container of fish
Pink Salmon
Fillet
Pink Salmon
Steak
-I
| | Record: M 1 ^ | | 53 ~ [ ~! 1^1 of 58
Exit | Back
Next
Figure 33. Photographs assigned to species in Quinault CAPI
Press the Next button to proceed to the next configuration screen.
On the fifth datasheet screen (Figure 34) you define unique shellfish parts that will be asked for at
least one shellfish food item. The system will assign a unique Part ID for each Part Description defined
as shown in Figure 35. Note that the part "Other Part" is pre-defined in the datasheet and will
appear in the first row when you first open this datasheet.
a STEP FIVE: Define Shellfish Parts
1 SEAFOOD CONSUMPTION INSTALLATION WIZARD
Part Description
Other Part
H
Record: H | i [ | 2 | H | I of 2
Back Next
Figure 34. "Other Part" is defined by the application
I >41
-------
m STEP FIVE: Define Shellfish Parts
SEAFOOD CONSUMPTION CONFIGURATION WIZARD
Part Description
Other Part
Whole body
Whole body less gut
Roe y
Roe and mpat
Neck
Strap
Gut
_ Meat
_ Butter
Ink
_ Mantle
Tentacles
Record: M I 4
5 ~ | H IN*! of 19
id
Exit
Back
Next
Figure 35. Completed parts datasheet for Quinault CAPI
Press the Next button to proceed to the next configuration screen.
The sixth datasheet screen (Figure 36) is where you specify which shellfish parts are asked about for
each species of shellfish. Shellfish parts are asked about in the consumption rate sections of the
survey for adults and children. Choose a SpeciesID from the drop down box in the left field and an
associated Patt ID (Figure 37) in the right field's drop down box.
B1 STEP SIX: Assign ShellFish Parts to Species
Species ID
Part ID
1
I Crayfish
Freshwater Shellfish
Dungeness Crab
Red Rock Crab
Mussels
Clams
Razor Clams
| Pacific Oyster
Record: MM
50
23
24
25
26
27
28
29
30
J of 50
Exit
Back
Next
Figure 36. Select shellfish Species ID in first column
D-42
-------
Si STEP SIX: Assign ShellFish Parts to Species
SEAFOOD CONSUMPTION CONFIGURATION WIZARD
J Clams
Other Part
Whole body
Whole body less gut
Roe
Roe and meat
Neck
Strap
Gut
Record: H I < 11 50 ~ I H !~*! of 50
Figure 37, Select parts in second column
In the Figure 38, five parts are asked about for Razor Clams in the CAPI. The Respondent will also:
have an opportunity to provide percentage eaten information for another part not specified by
choosing "Other Part", though the Supervisor does not have to make that association on tins
datasheet.
5 STEP SIX: Assign ShellFish Parts to Species
SEAFOOD CONSUMPTION CONFIGURATION WIZARD
J
Razor Clams
Razor Clams
Razor Clams
Razor Clams
Razor Clams
Chiton
Chiton
Chiton
Cockles
Record: MM
Exit
Species ID
Part ID
Whole body
Whole body less gut
Neck
Strap
Boot
Gut
Mantle
Foot
Whole body
50 ~ M ~* of 50
Back
Next
Figure 38. Completed parts to ask for Quinault CAPI
Press the Next button to proceed to the next configuration screen.
On the seventh and last datasheet you will define the Shellfish harvest and Finfish catch locations in
your geographic area (Figure 39). These values will be used to populate the drop down box for
location questions in the CAPI. In the CAPI, if the Respondent reports a location not included on
the list, the Interviewer will be able to choose location "OTHER" and be prompted to specify the
name of the location provided by the Respondent.
D-43
-------
B1 STEP SEVEN: Define Catch and Harvest Locations
SEAFOOD CONSUMPTION CONFIGURATION WIZARD
Catch Area Code
Catch Area Name
Display Order
~
BB135
Chehalis River
14
COPB
Copalis Beach
10
COPR
Copalis River
9
GRYH
Grays Hahor
12
HMPR
Humptulips River
11
KALB
Kalaloch Beach
1
MOCR
Moclips River
7
OFSM
Offshore Marine Catch Area
13
—
OTHER
Other
99
Record: M
1 ~ | H !~*! of 16
Finish
Back
Figure 39. Locations of interest for Quinauit Nation
Press the Finish button at the bottom of the screen to exit the configuration screens. You may
modify the configuration datasheets as many times as you would like provided you have not
published a final version of the CAPI.
CAPI Draft Version
Once the Supervisor has completed the seven configuration datasheets and would like to view the
configured CAPI, he/she will choose the Publish Draft from the Configure menu item (Figure
40).
Configure Integrate Reports
Datasheets
Abou
Publish ~
Draft .
Deploy
Final
Figure 40. Publishing a draft version of the CAPI
Because this a new database again, you will have to "Stop All Macros" as instructed on pages 3-4.
Choosing this menu option updates a system table that tracks the versions of the CAPI. The draft
version of the CAPI will then be started. Any previous draft version, including test interview data,
will be overwritten. Exit from the CAPI software to return to the Supervisor Module. See the
Interviewer's Guide for instructions on how to run the CAPI and exit.
You may publish and run the draft CAPI as many times as you like until you publish a final version.
If after publishing a final version you try to publish a draft again you will see the error message
shown in Figure 41.
1) II
-------
Seafood Consumption Supervisor Module
0
The CAPI configuration has been finalized.
The CAPI cannot be run in draft mode mode when finalized.
OK
Figure 41. Error citing the Draft Publish options are no longer available
CAPI Final Version
Once the Supervisor is satisfied with the CAPI version, he/she will choose to publish a final version.
It is up to the Supervisor to ensure the configuration is complete, accurate, and logical so that the
CAPI performs effectively and as intended. Choose the Publish Final menu options (Figure 42).
Integrate Reports
Configure
Pub ish
Figure 42. Publishing a final version of the CAPI
Choosing this menu option updates a system table that tracks the versions of the CAPI and makes
note that the CAPI has been finalized and is ready to be deployed to sites. You will be asked if you
want to review the CAPI software (Figure 43).
Figure 43. After final publication, Supervisor is asked to review the CAP
software
Yes
Seafood Consumption Supervisor Module
Do you want to review the CAPI survey based on the finalized configuration?
No
Press the Yes button to initiate the CAPI and press the No button to return to the menu.
This menu option creates a Supervisor's version of the final CAPI. The last draft version is now the
same as the final draft. Should you want to run the final version subsequently, choose the Choose
the Publish Final menu options. Test data can be saved in the final draft version. This Supervisor's
version of the CAPI database will not be the version used to compile data from the sites upon
import. The Supervisor, therefore, may run the final version and add test data without affecting the
database that is deployed to the sites.
Once you publish a final version, you may no longer add new records to the Species Question
Library through the main menu and modify text to existing records in the Species Question Library,
and you may no longer access the configuration datasheets.
D-45
-------
Deployment
Once a final version of the CAPI has been published, the Supervisor can deploy the software to the
sites. Choose the menu option Deploy to initiate this function (Figure 44). You should have ready
a USB storage device, also known as a thumb-drive or flash-drive.
Configure Integrate
Publish
Figure 44. Deployment menu item
A window will be displayed asking you to provide the name of the site to receive the software
(Figure 45).
Si Site Information
Site Name:
OK Cancel
Figure 45. Prompt for site information
Pressing the Cancel button will return you to the main menu.
If the OK button is pressed, the module will create one zip file containing the software and database
needed by the sites to install the CAPI locally. In addition, the Install.exe file to guide the interviewer
through the installation process at the site will be copied to the USB device. The device can then be
given to an interviewer to install the CAPI software on their PC. After entering the Site Name, you
will be prompted to enter the location of where you would like the deployment files will be written
(Figure 46).
D-46
-------
Save in:
Local Disk (C:
Recent
Desktop
My Documents
My Computet
^ J File name: X.
[TNetworkSave as type: |Zip Files ("Zip)
Places
Cancel
Save
Type
CAPI
here
Change to
USB drive
designation,
in this case
JBoneMarrow
E^bsi
Q CHIPSET
l^| Documents and Settings
Gl INTEL
;"'i mks
_j My Downloads
PilMvLocalData
Qnic
t^OfficeScan NT
l£| Program Files
RECVCLER
QRM
SEAFOOD
£| SOUND
System Volume Information
jj^lTEMP
Q VIDEO
£3 WINDOWS
QWNT483E
—J Diskl. Zip
JjDisk2.zip
Figure 46. Prompt for deployment file location
Insert the USB device into a port (in Figure 47 it is the "B*8 drive) to write the files to the device in
the Save In field. Type in the file name CAPI and press the Save button.
Save as: f^fx]
Save in: | Removable Disk (E:] £i d* n-
I
My Recent
Documents
0
Desktop
&
My Documents
•Si
My Computer
Save
Cancel
File name: |bbei 3j
My Network Save as type: |Zip Files (".Zip)
Places
Figure 47. Prompt for Deployment file location
D-47
-------
You will be presented with a warning that any installation files previously installed on the device in
any folder will be deleted before the CAPI installation files will be written. Once completed you will
see a prompt verifying success (Figure 48).
Seafood Consumption Supervisor Module
Deployment files for thi
s site have bee
OK
n successfully completed
Figure 48. Prompt to determine success
Accumulation of CAPI Data from Sites
The Supervisor Module has the functionality to integrate interview data collected at the deployed
sites. Accumulation of a site's data into the Supervisor's master version (which differs from the
Supervisor's final version) is a three step process. First, the site specific SeafoodSurvey_dat.mdb file has
to be imported, then the "Other Specifieds" recorded by Interviewers need to be translated to a
SpeciesID that is part of the Species Question Library, and finally the data is added into the master
database.
Import
Once an interviewing site has completed its interviewing, it will export its data to a USB device,
possibly more depending upon the number of completed interviews. This process is explained in the
Interviewer's User Guide. This device should be sent to the Supervisor with care.
Insert the USB device received from the site into the Supervisor's machine USB drive. The
Supervisor will then import the data on the diskette(s) to his/her PC by choosing the Import menu
item (Figure 49).
Integrate
Reports
Import
J
Translate
Append
Figure 49. Import menu item
D-48
-------
A window listing deployed sites is opened (Figure 50).
HI Import Site
Please Choose a Site for Import
Site s
Site Home
1
Bairibridge Island
2
Indianola
3
Port Madison
OK
Cancel
Figure 50. Choose a site to import
Highlight the site you would like to import. Press the OK button to start the import or press Cancel
to return to the menu. If you continue to import, you will be asked to specify the location of the
data to be imported (Figure 51).
Please Select a file
Look in:
EE
E & H-
a
Recent
&
Desktop
J
My Documents
&
My Computer
Ir^BoneMarrow
^)bsi
O CHIPSET
ir^t Documents and Settings
INTEL
¦i ~'irnks
. )My Downloads
i£|MyLocalData
Iconic
_jOffice5can NT
Q Program Files
RECYCLER
Orm
Q SEAFOOD
QSOUND
System Volume Information
l£)TEMP
23 VIDEO
l£D WINDOWS
£)WNT483E
^Diskl.Zip
33 Disk2.zip
My Network
Places
File name:
Files of type:
r
jZip Files (x.Zip)
r~ Open as read-only
"3
"3]
Open
Cancel
Change to
USB port
designation,
Figure 51. Specify the drive letter where the site's USB device has been inserted
Press the Open button. You should see a file with the name of transs/Ye/JO.zip, where siteno is the
site number assigned by the system during deployment (Figure 52).
D-49
-------
Please Select a file
Look in:
Removable Disk (E:
[2)
My Recent
Documents
Desktop
My Documents
My Computer
File name:
My Network Files of type:
Places
-Jtrans3,2ip
' $= B Cje m-
k
|Zip Files (" Zip)
Open as read-only
Open
Cancel
Figure 52. Highlight the file name with your mouse
The site number should match the number that corresponds to the site name: chosen in Figure 50.
Highlight the file with your mouse. Press the Open button. If the site number of the file name on
the diskette does not match the site number of the site chosen in Figure 50, the error message in
Figure 53 will be displayed.
Seafood Consumption Supervisor Module
Incorrect transfer file from Interviewer CAF'I System. Please verify that you have the correct file and re-import
Figure 53. Error when file name site number does not match selected site number
Press OK to return to site selection screen in Figure 50.
If the site numbers match, the system will copy the file in compressed format to C:\
SupemsorSeafood\Import and write an uncompressed version to C:\ SnpenmorSeafood\Translate. The
uncompressed version will have a file name of transs/Veno.mdb, where siteno is the site number
assigned by the system during deployment. You will be notified when the process has completed
(Figure 54),
D-50
-------
Seafood Consumption Supervisor Module X
File has been succe
ssfully Imported.
OK
Figure 54. Import has been successful
The selected site will no longer be available for import and will be absent from any future site
selections list displayed in Figure 50.
The site's CAPI data is then ready to be prepared for accumulation.
Translate
During the interview at the end of the 24-hour Recall and following annual Shellfish consumption,
the Respondent is asked if he/she ate or eats any other seafood that was not asked about during the
CAPI. These are referred to as "Other Specifieds" and have been categorized to Group "X" in the
site database. If a site's database contains "Other Specifieds," the site database cannot be added to
the master database.
The Translate menu item will allow you to handle the other seafood items reported by a
Respondent (Figure 55).
Integrate
Reports
Import
Translate .
Append
Figure 55. Translate menu item
Other reported seafood items specified are assigned a temporary SpeciesID during the administration
of a CAPI. They must be translated to a SpeciesID that exists in the Species Question Library. Having
the same SpecJesIDs across sites facilitates analysis. Note, translated items remain in Group ID "X"
and do not appear on any of the pre-defined reports.
Choose Translate to display the names of the sites that have had their data imported (Figure 56).
D-51
-------
SLOther Species Translation (from Site Data)
Please Choose a Site for Translation
Site a
Site Name
Bainbridge Island
iOKi
Cancel
Figure 56. Listing of sites ready for translation
Use the mouse to highlight a site name and press the OK button. The Cancel button will return you
to the menu. If OK is pressed, a list of other specified seafood will be displayed (Figure 57).
Translation Screen
Other Specifieds Needing Translation
Green Crab (Recall Section)
Candlefish (Recall Section)
Surf smelt (Rates Section)
Cutthroat trout (Recall Section)
_Ll 1 jJ
| r Add to Library
Search Fields for Translating Species
<* Species Description (Include Tent) Search
Enter Search:
Reset
King Salmon (Chinook)
Sockeye (Red)
Coho (Silver)
Chum (Dog)
Pink
Steelhead
King Salmon (Sockeye, Coho, Chum, King, Pink, and : i
<—• I j. 1
Save Translation
Other Specifieds
Translated To
Figure 57. Other seafood items to be translated
The species list on the right hand side provides a list of species from the Species Question Library.
Search functionality has been provided to more easily find a species from the master library for the
"translated to" value. To translate "Surf Smelt" use your mouse to highlight the item in the upper
left pane. To see what "smelt" species currently exist in the Species Question Library you can search
the library by entering in "smelt" in the field after the "Enter Search" label (Figure 58).
D-52
-------
Translation Screen
Other Specifieds Needing Translation |
Green Crab (Recall Section)
Candlefish (Recall Section)
Surf smelt (Rates Section)
Cutthroat trout (Recall Section)
| r Add to Library
Search Fields for Translating Species
Species Description f (Include Text) Search |
Enter Search: IsmeltJ Reset I
King Salmon (Chinook)
Sockeye (Red)
Coho (Silver)
Chum (Dog)
Pink
Steelhead
King Salmon (Sockeye, Coho, Chum, King, Pink, and :
Save Translation
Other Specifieds
Translated To
Figure 58. Search for "smelt" in Species Question Library.
Press the Search button. The results are shown in Figure 59.
Translation Screen
Other Specifieds Needing Translation
Green Crab (Recall Section)
Candlefish (Recall Section)
Surf smelt (Rates Section)
Cutthroat trout (Recall Section)
<
r Add to Library
Search Fields for Translating Species
<•" Species Description C (Include Text) Search |
Enter Search: Ismell] Reset
King Salmon (Chinook)
Sockeye (Red)
Coho (Silver)
Chum (pog)
Pink
Steelhead
King Salmon (Sockeye, Coho, Chum, King, Pink, and : ^
Save Translation
Other Specifieds
Translated To
Figure 59. Search results for "smelt" in Species Question Library.
If you choose to translate "Surf smelt" to "Smelt", highlight "Smelt" in the upper right pane and
press the Save Translation button. The translation will appear in the lower half of the window and
an "x" will appear to the right of "Surf smelt" (Figure 60).
D-53
-------
Translation Screen
Other Specifieds Needing Translation
Green Crab (Recall Section)
Candlefish (Recall Section)
x Surf smelt (Rates Section)
Cutthroat trout (Recall Section)
< ml _>
Remove Translation l~~ Add I
Search Fields for Translating Species
f* Species Description C (Include Text) Search
Enter Search: Ismelt Reset
Smelt, Herring
Save Translation
Other Specifieds
1
Translated To
Surf smelt
Smelt
Close
Figure 60. "Surf smelt" has been translated to "Smelt"
Highlight any translated "Other Specified" in the upper left to display the translation in the lower
half of the window. If you would like to delete the translation, highlight the record in the bottom
half of window and press the Remove Translation button. The record at the bottom will disappear
as will the "x" to the left of the item in upper left box.
You must translate each item. You cannot translate two "Other Specifieds" to the same SpeciesID so
as not to comprise database integrity. If you cannot find an appropriate translation in the existing
Question Library, you can add a new entry to the library. Highlight the "other specified" to be
translated, click the "Add to Library" checkbox (Figure 61). Press the Save Translation button.
The text will be added to the Species Question Library under SpeciesDescription and a unique SpeciesID
will be assigned by the module. The item will be translated to itself and appear at the bottom of the
screen under items translated (Figure 62).
yfiUiiiiiiim
Other Specifieds Needing Translation |
Green Crab (Recall Section)
Candlefish (Recall Section)
x Surf smelt (Rates Section)
Cutthroat trout (Recall Section)
Search Fields for Translating Species
(* SpeciesDescription C (Include Text) Search
Enter Search: Ismelt Reset
Smelt, Herring
W iAdd to Library
Save Translation
Other Specifieds
Translated To
Figure 61. Add Cutthroat trout to the Species Question Library
D-54
-------
Translation Screen
Search Fields for Translating Species
f* Species Description
Enler Search: Ismelt
(Include Text) Search
Resel
Smelt
Green Crab (Recall Section)
Candlefish (Recall Section)
Surf smelt (Rates Section)
Cutthroat trout (Recall Section
Other Specifieds Needing Translation
Remove Translation I- Add i
Other Specifieds
1
Ti.in&l.ited To
Cutthroat trout
Cutthroat trout
Close |
Figure 62. Cutthroat trout has been added to the library
You can choose the Translate menu option multiple times as long the site's data has not been
appended that site's data into the master database. You will not be able to append site data if there
are any other species that have not been translated.
Append
Once there are no items to be translated, the site data can be added to the master database. When
you choose Append (Figure 63), the module will display the names of the sites that have had their
data imported (Figure 64).
Integrate
Reports
Import
Translate
Append
Figure 63. Append menu option
D-55
-------
Hi Appending Site Data
Please Choose a Site for Appending Data
Site =
Site I Lime
Bainbridge Island
I OK i
Cancel
Figure 64. List of sites ready to be appended to master database
Highlight a site name and press the OK button. The module will then append all site data from the
selected site to the master database and display a successfully completed message as shown in
Figure 65.
firry
Seafood Consumption Supervisor Module X
Appending of Site D.
ata has been si
OK
jccessfully completed.
Figure 65. Notification of successful integration
Reports
In the Supervisor Module, there are six statistical reports defined in the application software. These
reports can be found under the Reports main menu item (Figure 66). Descriptive statistics can be
generated for 24-hour recall and seasonal/annual rate data at the species item level, at the group
level, and overall (Figure 67). Overall and group statistics by gender for 24-hour recall and rate data
are also available.
Reports About
Overall and Group ~
Species ~
Gender and Group ~
Query
Figure 66. Reports menu options
D-56
-------
Reports About
Overall and Group ~
Recall
Rates
Rates Percentiles
Species ~
Gender and Group ~
Query
Figure 67. Reports sub-menu options
Query
Under the Reports menu item, there is a Query option that offers a structured method of querying
the adult consumption rates ascertained in the Rates section of the CAPI on pre-defined selection
criteria. These rates are stored in TblRates table. You will be taken through a series of seven screens
that control what rates data will be presented in the results. You can choose to export the result to a
comma delimited file or to execute a report using the results as the basis. The report includes the
count of the number of subjects included in the subset (summed across seasonal records), mean of
the annual gram amount consumed, the standard deviation and variance of the amounts as well as
percentile values.
Choosing this menu item will bring up the first window in a series that asks you to select whether
you want to look at species asked in the Finfish rates section or whether you want to look species
asked in the Shellfish rates section of the CAPI (Figure 68). Once, you have made your selection,
the relevant groups are populated in the bottom left window.
U GROUP SELECTION
^ Finfish r Shellfish
Finfish Group
Migratory
Pelagic
Benthic
Bottom Feeding
Cancel | Back | Next | Finish |
Figure 68. First query filter screen
Use your mouse to highlight a group or groups of interest and move over to the right box using the
operators (Figure 69).
D-57
-------
Finfish
r Shellfish
Finfish Group
Migratory
Benthic
Bottom Feeding
Pelagic
M GROUP SELECTION
Figure 69. Selection of group of interest
You can press Finish at any time to move to the output option screen where you can either export
the queried data or preview the statistical results in a report format.
Press the Next button to move to the next query screen or choose the Cancel button to return to
the main menu.
This next screen will ask you if you would like to see only species questions related to migratory fish,
resident fish, or both (Figure 70). The default is both.
C Migratory Fish
C Resident Fish
® MIGRATORY SELECTION
Figure 70. Second query filter screen
Press the Next button to move to the next query screen or choose the Cancel button to return to
the main menu.
This next screen will display all of the locations defined during the configuration, for the group(s)
that you have chosen on the first screen (Figure 71). Select locations of interest from the left pane
and move to the right pane.
D-58
-------
m LOCATION SELECTION
AVAILABLE
SELECTED
Beach from Pt. Brown to Copalis River
Quinault River
Beach from Grenviiie to Queets River i
Queets River
Beach from Queets River to Kalaloch Beach > 1
l~~ Don't filter on Locations
Cancel Back Next Finish
— i — i —
i - i
i
i
Figure 71. Selecting location on third query filter screen
Press the Next button to move to the next query screen or choose the Cancel button to return to
the main menu. If you do not wish to choose a location but would like to keep going through the
query steps, choose the "Don't filter on locations" checkbox and press the Next button.
This next screen will display all of the defined sources of seafood, for groups satisfying the criteria
specified in the first two screens. Select the source or sources of interest from the left pane and
move to the right pane (Figure 72).
m SOURCES SELECTION
IX
AVAILABLE
SELECTED
Inside Area
iGrocery
Outside Area .
Restaurant
Other Sources >
iiJ
I~~ Don't filter on Sources
Cancel Back Next Finish
- i - i —
i - i
1
I
Figure 72. Selection of sources on fourth query filter screen
Press the Next button to move to the next query screen or choose the Cancel button to return to
the main menu. If you do not wish to choose a source but would like to keep going through the
query steps, choose the "Don't filter on sources" checkbox and press the Next button.
This next screen will display the species questions that match the criteria specified in the previous
four query screens (Figure 73). Use your mouse to highlight a species question or species questions
of interest and move over to the right box using the operators.
D-59
-------
SB SPECIES QUESTIONS SELECTION
AVAILABLE
SELECTED
Sockeye (Red)
Smelt
Sturgeon
Sole/Flounder
jJ
id
«l
King Salmon (Chinook)
Herring
Pollock (imitation crab and other imitation seafood)
Cancel
Back
Next
Finish
Figure 73. Selection of species of interest on fifth query filter screen
Press the Next button to move to the sixth query screen or choose the Cancel button to return to
the main menu.
This screen asks whether you want to include in your subset subjects that consumed each of the
species selected in the previous screen or subjects that consumed at least one of the selected species
(Figure 74). Subjects that consumed all species are called All Consumers while subjects that
consumed at least one species are called Ever Consumers.
E CONSUMER TYPE SELECTION
r
IALL Species Consumed j
c
At Least One Species Consumed
Cancel |
Back | Next | Finish
I I
Figure 74. Selecting consumer type on the last query filter screen
Press the Next button to move to the last query screen or choose the Cancel button to return to
the main menu. Select what you would like to do with the results (Figure 75).
rn OUTPUT SELECT ON
^ Report
Export
Cancel
Figure 75. Selecting output format for query results
D-60
-------
The Export option will save the results to a comma delimited file that can then be imported into
software such as Excel or SAS. There will be one record for each seafood item reported consumed
and stored in TblKates that meets the subsetting criteria provided. For Finfish species seasonal rates
are asked for in the CAPI. If the Respondent was able to provide both in and out of season rates,
there will be two records, one for each time period, in the file for that species. In addition to the
fields in TblKates, the export file will contain a calculated annual consumption amount based on
reported frequency and raw portion size. You will be prompted for a location and file name (Figure
76).
Save in:
P
Recent
0
Desktop
My Documents
s*
My Computer
My Network
Places
Local Disk (C:)
«- (*-l HI-
Ir^AccFiles
Ir^Inetoub
Ir^l Pictures
l^)APP5-OLD
l^| INTEL
15) Pico
Ir^lBoneMarrow
Ij^Kccs
IrTiPlcoPath
QCData
lc3LABs
O Polish
CHIPSET
£HMABRY
|£|PolUtility
Documents and Settings
[5l rinks
1^1 Program Files
Ir^ Election
|^| My Download Files
|£|RECYCLER
OF2I5.0
IrTiMv Downloads
fi^Scte
Ir^lFamBrain
Ir^iMv Games
IrTi Seafood
l£)Formsii
(£|Nese
l£|SnagIt6
l£)Greentree
|£|NIC
|£| SOUND
O icons
IcHNLSTSystem
1^1 Source
Ir^i Ideas
(^OfficeScan NT
System Volume Inforrric
i
-------
M Consumption Rates For Data Subset
Consumption Rates For Data Subset
N (Ever) N (Never) MEAN STD
VAR
| 1 | 0 | 21
1.67 |
I
Consumption Rates Percentiles
1
21.
67
5
21.
67
10
21.
67
15
21.
67
20
21.
67
25
21.
67
30
21.
67
35
21.
67
40
21.
67
45
21.
.67
50
21.
.67
55
21.
.67
60
21.
67
65
21.
67
Page: I I 11
T_U J-J
I
Figure 77. Report results from query search
D-62
-------
The Fish Consumption Survey Tool (FCST)
CAPI Survey
Description of the Interviewer Software
The Fish Consumption Survey Tool (FCST) is administered by computer assisted personal interview
(CAPI) software developed in MS Access. The Suquamish Nation Seafood Consumption Survey, a
previously used, hardcopy, interview administered questionnaire, is the basis of the design of the
CAPI instrument. The screen shots included in this section reflect the survey as configured for the
Quinault Nation in March 2006.
The software consists of two Access 2010 databases: one database (SeafoodSurvej_dat.accdb) contains
the data collected during interviews and lookup tables (i.e. tables that contain the various possible
answers to interview questions) while the second database (SeafoodSurvej_app.accdb) contains the
application code and objects such as the queries, forms, and reports. These databases will be copied
to the local folder C:\Seafood during installation.
Computer Requirements
The designated Interviewer computer must have a Pentium III or higher processor. It must have at
least 128 MB of memory though 256 MB is preferable. The hard disk should have at least 60 MB
available space. The computer's operating system should be Windows 7. Access 2010 or higher
should be installed on the computer prior to installing the CAPI software.
Setup
The CAPI software has been customized for your study so that seafood typically consumed by the
population are included in the survey. This customization has been completed on the Supervisor PC
and passed on to sites on a USB storage device (also known as a flash-drive or thumb-drive).
Interviewers are responsible for installing the CAPI on their designated PCs referred to as Sites.
To install the CAPI, insert the USB device into your laptop's USB port. For these instructions it is
assumed the port is in the "E:" drive. Open Windows Explorer. Windows Explorer may be found
in many places on your PC and it varies from PC to PC. In Windows 7, typically, you can find it by
pressing Start, choosing Programs, choosing Accessories, and choosing Windows Explorer. Use
Windows Explorer to expand the "E:" drive (or the USB port/drive used on your PC). You will see
a window as shown in Figure 1.
D-63
-------
)¦<=> 1 '1 i
j*>\ « GeneralUse (Wwestat,comVDFS) (G:)
~ For Rajiv ~
Search For Rajiv
P
1 Organize ~ Burn
New folder
=» a ®
~*f Favorites
Name
Date modified
Type
Size
E Desktop
Jy CAPLzip
2/2/2015 9:27 AM
Compressed (zipp...
840 KB
Downloads
'Ml RecentPlaces
jjfj' Install.msi
2/1/20154:28 PM
Windows Installer...
284 KB
3 Libraries
H) Documents
qfr Music
Sal Pictures
8 Videos
Computer
& OSDisk (CO
^ GeneralUse (Wwe
S? TAYAL.R (\\west
INSTALL (Wwest,
5* NCS (Wwestat.co
DFS (Wwestat.co
PATHHMSTEST
5? WROC (Wncsfs)
C& Docs (WTESTWEE
>. . ¦
I 2 items Offline status: Online
Offline availability: Not available
Figure 1. Opening Windows Explorer
Double-click on the file Install.msi. The installation procedure will copy the necessary software
files to the local folder C:\Program Files fx86)\Seafood Survey unless otherwise specified during
installation.
To begin the second part of the installation, double click on the file In stall App. n ccdh in the
C:\Program Files (x86)\Seafood Survey folder as shown in Figure 2.
OSDisk (C:) ~ Program Files (x86) ~ Seafood Survey
Organize *
VT Favorites
¦ Desktop
£ Downloads
% RecentPlaces
1^1 Libraries
i_£l Documents
J} Music
B Pictures
0 Videos
Computer
SL OSDisk (CO
1-3* GeneralUse (\\we
4k* TAYAL_R (\\west
tX* INSTALL (\\west<
^ NCS (Wwestat.co
DFS (Wwestat.
PATHHMSTEST (
V WROC (Wncsfs) (L
1Docs (WTESTWEf
Include in library ~ Share with
Name
®] InstallApp.accdb
New folder §= w 3 ^
Date modified Type Size
1/30/201512:15 PM Microsoft Access... 736 KB
Figure 2. Double click on InstallApp.accdb
D-64
-------
Click on the Cancel button then Click on the Enable Contents button. Select the Run Install
button to continue installation (Figure 3).
[Aj^^
PflX
Figure 3. Select Cancel
Press OK to continue at the nest prompt (Figure 4)
Seafood Survey Installation Module
P ease choose the fo der where the Insta ation hi e CAPI.ZIP exists.
Figure 4. Press OK
Once the screen in Figure 5 is presented, change the drive designation to the USB port (drive "E:"
in the figure).
D-65
-------
Specify folder and filename
Change to
USB drive
designation,
typically E:
MEMORY (E:J
nt
Documents
fr
My Computer
My Network
Places
Name ^
£0 CAPI.zip
9
Desktop
3
My Documents
File name:
Files of type:
Size Type
2,647 KB WinZip File
Zip Files (".Zip)
Open as read-only
Date Modified
6/7/2006 2:27 PM
u
E
Figure 5. Specify the USB drive
Highlight the file CAPI.zip and select the Open button.
You will be notified of the successful installation (Figure 6).
00
Open
Cancel
Seafood Survey Installation Module
Files have been successfully installed
Figure 6. Notification of successful installation
You will need to copy a shortcut file to your desktop to allow for easy access to the CAPI software.
While there are a few different ways of doing this, this guide will describe one way. Using Windows
Explorer, go to the C:\Seafood folder. Right click on Shortcut to Seafood Survey file (Figure 7)
and choose "Copy" from the menu.
D-66
-------
Comput
OSDisk (C:)
.¦ic.-r '
library ~
Share with
New folder
Date modified
Mame
hT Favorites
Fish.
3] SeafoodSurvey_app.accdb
SeafoodSurvey_dat.accdb
Shortcut to Seafood Survey
D 65 Hep
2A/2015 4:28 PM
2/2/2015 9:20 AM
2/2/2015 9:27 AM
1/29/2015 3:02 PM
Icon
Microsoft Access
Microsoft Access
Shortcut
:¦ KB
Downloads
RecentPlaces
: >:0 1 •
jit Libraries
13 Documents
Jl Music
Q Pictures
B Videos
Computer
& OSDisk (CO
5? GeneralUse (\\we
V TAYAL.R (\\west
^ INSTALL (\\west<
NCS (Wwestat.co
*-3* DFS (Wwestat.coi
4* PATHHMSTEST (
^ WROC (Wncsfs) (
<3* Docs (WTESTWEl
Figure 7. Shortcut file in the folder
Minimize all opened windows by choosing the minimize window control at the top right cornet of a
window IS. You should then be on the desktop. Right click anywhere on the desktop to display a
tunction menu and choose "Paste." A shortcut icon will be pasted ott he desktop (Figure 8). If you
prefer, you can use your mouse to drag the icon to anywhere on the desktop.
Figure 8. Shortcut file pasted to the desktop
When start the survey for the first time a prompt to "Enable content" will appear (Figure 9)
Confumptson Surrt
"1
3?
Macro S ngte Step
Figure 9. Enable Content prompt
D-67
-------
Select the Stop All Macros button. Then select the Enable Content button in the yellow header.
You will be presented with the screen shown in Figure 10.
Figure 10. Interviewer prompt
Enter the Interviewer name that has been assigned to you and select Ok.
Groups and Species
The CAPI structure and flow are in large part based on the definition of seafood groups and the
species assigned to those groups of interest to the study. The Supervisor defines the groups when
customizing the CAPI. Examples of groups are: 1) Migratory Finfish 2) Bottom-feeding Finfish and
3) Shellfish. Specific seafood types are assigned to these groups by the Supervisor. For example,
Sole, Flounder, and Rockfish might be assigned to the group Bottom-feeding Finfish while Clams
and Crabs would be assigned to Shellfish.
Interview Structure/Flow
The CAPI is organized into the following sections:
Startup In this section, the Interviewer will introduce the survey to the Respondent, record the date
and start time of the interview, record the Respondent's gender, and determine if the Respondent
has previously completed a survey within the designated time period. The survey is designed to allow
for multiple administrations to a Respondent. A survey is uniquely identified by a Subject ID. A
Subject ID is made up of the Study number, Interviewer name, Person ID, and Interview number.
The Subject ID appears at the top of every page. Person ID uniquely identifies the Respondent and
is assigned outside the survey. The Interviewer enters the Respondent ID. The system will calculate
the survey interview number by incrementing the last survey number for that Person ID. If it is the
first survey, the interview number is 1. The survey system will construct the full Subject ID. For the
Subject ID 574-00-12345-1, the Study number is "574", the Interviewer name is "00", the Person
ID is "12345" and the survey interview number is "1".
24-Hour Recall In this section, the Respondent is asked if certain seafood items were eaten
yesterday, the day before the interview. The Respondent is also given the opportunity to report on
other seafood items consumed yesterday but not already asked about.
D-68
-------
Finfish In this section, the Respondent is asked about fin fish consumption over the course of a year,
or in and out of season. Information included in this section of the survey includes: consumption
rates and amounts, fish sources (e.g., grocery store, restaurant, and catch area), and parts of the fish
eaten. This section is divided into multiple subsections, one for each group of finfish defined by the
Supervisor. Starting with the first group, the Respondent is asked about consumption of species
when the fish are in season and when the fish are out of season. If the Respondent is unable to
report by season, the Respondent is asked about annual consumption patterns. Quantifying seafood
consumption is done through the use of photographs depicting typical portion sizes of seafood
prepared in different forms such as steak or in chowder. Once the Respondent provides
consumption information for the first group ot fish, the Respondent is asked to report on what
percentage of the time he/she eats specific parts, considering all the fish trom the first group
typically eaten. The Respondent is also asked to report on from where the fish consumed was
generally obtained. The locations have been predefined by the Supervisor though the Respondent
has the option to report a location not defined by the Supervisor. If the fish was reported to have
been caught in or outside the area of interest, the Respondent is further prompted to provide the
location of where the fish was caught. A study may choose not to gather information about harvest
or catch locations outside the area of interest. The Interviewers can skip the prompts for outside
area information by answering "No" at the prompt shown in Figure 11.
21 FINFISH CATCH OUTSIDE AREA - PART 1
S3
[Subject ID: ~| |574 00-12345 1
Interview Prompt
We are interested in knowing where outside the Potomac you, your family members, the Westat Nation, and/or friends
catch the fish you eat. If you know where the fish is caught and want to show me by referring to these display maps,
we would appreciate having the information. If you prefer not to tell me, we can go on to the next question.
Interview Instructions
If your Nation has opted not to capture sources located outside of the Potomac, answer NO to skip the next series of
questions. Refer to area maps in the booklet to highlight inside and outside areas of interest.
Q6
Able to provide outside of Potomac information?
1 Id
flext Previous
Yes
No
-7 (Prefer not to say)
-8 (Don't know)
Figure 11. Interviewer should answer "No" to avoid prompting Respondent
for seafood sources located outside the tribal area of interest
Shellfish In this section, the Respondent is asked about shellfish consumption rates and amounts,
shellfish sources (e.g., grocery store, restaurant, and harvest area), and parts of the shellfish eaten.
This section is divided into multiple subsections, one for each of the group of shellfish defined by
the Supervisor. After providing portion and frequency values for a specific shellfish, the Respondent
is asked what percent of the time specified parts of that shellfish are eaten. As in the finfish groups,
the Respondent is asked to consider all consumed shellfish when reporting sources. If the shellfish
was harvested in or outside the area of interest, the Respondent is prompted to select locations from
a pre-defined list for each shellfish item reported consumed. Again, some studies may choose not to
D-69
-------
collect information about seafood sources located outside the main area of interest. To skip
questions that ask the respondent for outside locations answer "No" to this question (see Figure 9).
Other Seafood In this section, the Respondent is asked to report the consumption of other seafood
as defined by the Supervisor. The Respondent is also given an opportunity to provide names of
other seafood, not previously asked, and report on consumption.
Child In this section, Respondents are asked to report on children living in their households and
seafood consumption patterns for each reported child under six years of age. If the study is not
interested in fish consumption of children, do not ask respondent, just select "No" to Q31, "Are
there any children under six years old living in your household?". Any household member can
answer questions about a child, including a parent or grandparent. However, if any other household
member has participated in the CAPI survey, the current Respondent should not answer these
questions. Skip these questions by answering "Yes" to the prompt shown in Figure 12.
3) FISH and shellfish eating patterns of children
£3
Subject ID: |
574 00 12345 1
Interview Prompt
Have any other household members already participated in this survey during this time period?
Interview Instructions
Time period is indicated by the number to the right of the hyphen within the SubjectlD.
Other member
k
Next
Previoi
Yes
No
-7 (Prefer not to say)
-8 (Don't know)
Figure 12. Interviewer should answer "Yes" to skip Children section if
other household members previously responded to the CAPI Survey
For Finfish, the child questions have been configured to be asked at the group level and not at a
specific species level. For each finfish group reported consumed, the Respondent is asked to report
on the percentage of parts consumed. Shellfish group specific items, such as butter clams and horse
clams, are then asked about, as are the percent of specific parts eaten for each item.
General and Special Events Questions In this section, the Respondent is prompted for general
information on seafood consumption, demographic information, and consumption of certain
seafood items at special events and gatherings. Also, in this section, you, the Interviewer, record that
the Respondent signed the Participation Verification Form as well as the interview stop time and
duration.
Interviewer Notes In this section, the Interviewer records any comments about how well the
interview went and rates the reliability of the answers.
D-70
-------
Starting the CAPI Application
During installation an icon was copied to your desktop (Figure 13).
Figure 13. Desktop icon
Double click on this icon to start the software. The Interviewer will be prompted for his/her name
that will be used by the software to create a personalized introduction.
Main Menu Functions The main menu has four items: Exit, Data Entry, Export, and About
(Figure 14).
¦|]»| SFCS Menu
X Exit Data Entry T Export About
Figure 14. Main menu
Exit To exit the application, you must choose the Exit item under the File tab or under the SFCS
Menu tab. The Windows close button that appears in the top right hand corner of the screen
has been disabled to prevent abnormal shutdown that might affect the integrity of the database.
Data Entry When this item is selected, three sub-menu items will appear: New Inteniew, Edit Interview,
and Delete Inteniew. Select New Inteniew when starting a new interview. The software will assign a
unique identifier to each subject's interview. Choose Edit Inteniew to open an interview that was
previously completed or started. You will be prompted to provide the subject interview ID. Choose
Delete Inteniew to delete an existing interview and remove all interview data from all data tables.
Export This function will prepare the database on your desktop for transfer to the Supervisor's PC
where it will be integrated with CAPI data collected by other Interviewers. You will only be allowed
to export your data once. It should be done ONLY when you have completed all interviews.
About The selection of this item causes a window to open that reports the version of the software.
Hit the OK button to return to the main menu.
To begin an interview, select the Data Entry Tab and select "New Interview".
D-71
-------
Screen Layout
Most of the CAPI screens are designed to ask one question per screen as shown in Figure 15.
Below the Subject ID, there if an Interviewer Prompt (Figure 15). This is the text the Interviewer
should read to the Respondent. It should be read exactly as it appears.
Next on the screen (Figure 15), Interviewer Instructions appear. This text should not be read to the
Respondent but should be read silently by the Interviewer. This text should help the Interviewer
better guide the Respondent and manage the interview. If there are no relevant instructions, the
word will appear in the box.
Under the Interviewer Instructions text box, the field name, a brief field description, and finally, the
response field appear. The cursor will be focused on the response field where you will key the
Respondent's answer. If there is a predefined codelist (Figure 15), a downward facing arrowhead
will appear to the right of the field. Clicking on the arrowhead will display the relevant codelist. For
numeric fields and non-codelist fields, the Interviewer must type in a value or text response. At the
bottom of the screen (Figure 15), the active control buttons are displayed. These buttons will allow
the Interviewer to navigate within the application.
3] PART ONE: DIETARY INTAKE - 24 HOUR RECALL
S3
Subject ID ]|574 00 12345 1
Interviewer Prompt
Did you eat any King Salmon such as Chinook yesterday from the time you woke up until the time you went to sleep for
the night? Please include at meals and as snacks.
Interviewer Instructions
Display the image of King Salmon.
R1 Eat King Salmon?
Next
Previous
"0
Yes
No
-7 (Prefer not to say)
-8 (Don't know)
Figure 15. Typical screen layout
D-72
-------
There are a few screens in flu: CAPI that contain two or more fields that logically belong together.
For example, frequency and unit (Figure 16), height, feet, and inches (Figure 17), and percents that
must add to 100 (Figure 18).
jg] PART ONE; DIETARY INTAKE - 24 HOUR RECALL
S3
Subject ID |
574 00-12345-1
Interviewer Prompt
Thinking about all the King Salmon you ate yesterday, how much total King Salmon did you eat? You can either give me
that amount in pounds or ounces.
Interviewer Instructions
Respondent should provide total amount in pounds or ounces. If Respondent prefers not to provide the amount, enter
-7 in the first field. If Respondent does not know the amount, enter -8 in the first field.
|R3
Amount of King Salmon eaten?
Next Species
Previous
1 H
R4
Ounces
Pounds
-7 (Prefer not to say)
-8 (Don't Know)
Unit
Figure 16. Screen with two fields
lH GATHERING QUESTIONS - PART2
S3
Subject ID |
574-00-12345-1
|lnterviewer Prompt
What is your height?
Interviewer Instructions
If Respondent prefers not to tell you, enter -7 in the first field. If does not know, enter -8.
Q52a Height - feet
Q52b Height - inches
Next
Previous
Figure 17. Another screen with two fields
D-73
-------
fU PARTS CONSUMED AND PREPARATION
S3
Subject ID: |57400-12345 1
Interviewer Prompt
What percent of the time do you eat the fillet with and without skin? Your two answers should equal 100%. Provide me
the "fillet v^ith skin percentage" value first.
Interviewer Instructions
Enter -7 if Respondent prefers not to provide a percentage. Enter -8 if Respondent does not know the percentage.
Press the Tab key to update percent total.
Q2A1
Fillet with skin percentage
llOO %
Q2A2
Fillet without skin percentage
|3
(Total
|100 ¦ %
Next
Previous
Figure 18. Screen with multiple fields
Comments
The Interviewer can record comments or notes at any time during an interview (Figure 19). When
the F3 key is hit, the Comments Window is opened.
P Participant Comments
When entering Comments, please indicate question number, followed by a ;, species
name, followed by a ;, and then a comment, followed by a return
Enter Notes/Comments:
Q5B;Group B; respondent was somewhat unsure if inside area was code 876 or 877 so I recorded 877 because
ne felt more strongly sbout 877
TIB; Perch; respondent providing a best guess for seasonality responses!
Cancel
OK
Figure 19. Comments window
As stated in the top of the Comments window, the Interviewer should record comments in a
standard format to facilitate any future analyses. In Figure 19, the Interviewer has two comments
thus far indicating for "Group B" finfish, the Respondent was somewhat unsure about a catch area
D-74
-------
but felt strong enough to provide an answer and for Perch was providing a best guess for
seasonality.
The Interviewer can minimize the window by hitting the button 1131 in the top right hand corner of
the Comments window. The Comments window will then appear minimized in the bottom left hand
corner of the Access screen. To maximize this window, the Interviewer can click on the window's
restore control or hit F3 again. Any additions and changes have not yet been saved.
The Interviewer can also close the Comments window and save the changes by hitting the OK
button. Again, F3 will open the window.
The Interviewer can hit the Cancel button to not save the most recently entered comments, that is,
since the OK button was hit last.
Missing Values A value is required for most every CAPI question field, otherwise the software will
not allow the interview to proceed. For most of the questions, the Interviewer is allowed to record
"missing" responses. A value of -7 should be recorded in a non-codelist or numeric field when the
Respondent refuses to answer. A value of -8 should be recorded in a non-codelist or numeric field
when the Respondent does not know the answer. Codelists accessed for fields with the downward
facing arrowhead, contain these two missing response codes.
Edits and Skips When the Interviewer attempts to move off a screen, pre-defined edit checks are
run on the entered value(s). If the value falls outside a prescribed range or is inconsistent, given
another response(s), the Interviewer will be presented with a message window that provides
information so that the value can be verified.
If there is a range or
(Figure 20).
yic error, the message will tell the Interviewer why the value is suspect
Seafood Consumption Survey
o
Numbers of portions per day should not be more than 5.
Do you still want to Continue?
Yes
No
Figure 20. Example of warning message
These types of errors can be ignored and over-ridden by the Interviewer by hitting the Yes button
to continue. Or the Interviewer can choose to remain on the field if the No button is selected, so as
to review the response and make any necessary corrections.
If a value that was provided or changed conflicts with other information, the message will inform
the Interviewer of the conflict and, if appropriate, provide a warning that the application will clear
out any now irrelevant data (Figure 21).
D-75
-------
Yes
Seafood Consumption Survey
T1A has to be Ves when TIB is filled.
Do you want to clear all data from TIB and all relevant records?
No
Figure 21. Example of warning message
Some questions will be skipped if a previous response or lead-in type question is answered a certain
way. For example, it the Respondent is unable to answer questions about her child's diet, all the
dietary questions will be skipped and the interview will continue with the general questions.
Navigation There are two basic ways to navigate within an ongoing interview: the control buttons
at the bottom of each screen and the GoTo Menu item at the top of the screen. The buttons at the
bottom of the screen will vary depending on the CAPI section and active question. The buttons
allow the Interviewer to navigate within one of the CAPI sections. If the Interviewer wants to move
to a previously completed section, the Interviewer must use the GoTo menu item. Buttons will be
grayed out and rendered inactive if not appropriate for the species and/or question in focus (Figure
22).
PART TWO: FIN FISH CONSUMPTION - SEASON. FREQUENCY. PORTIONS
S3
Subject ID: |574 00 12345 1
Interviewer Prompt
In the past 12 months, did you ever eat King Salmon such as Chinook?
Interviewer Instructions
Display the species image of King Salmon.
T1A Eat King Salmon?
"0
Next
Yes
No
-7 (Prefer not to say)
-8 (Don't know)
Figure 22. Typical screen view
The Next button will advance the Interviewer to the next screen in the interview sequence. Any
prescribed edits will be run before the next screen is displayed. Prescribed skipped patterns will be
followed. Eor example, if the Respondent answered that his consumption pattern had not changed
over the past 20 years, the Next button will skip to a question about frequency of special events
attendance. If the Respondent had answered that his diet had changed, the Next button would
advance the Interviewer to the question about how it has changed.
D-76
-------
When on the last question for a food item, the Next button will be inactive. The Previous button
will take the Interviewer to the previous screen or lead-in question that controlled a series of
questions. For example, if the Interviewer was asking about Out of season consumption of King or
Chinook Salmon (Figure 23) and hit the Previous button, the Interviewer would be taken back to
questions about in season consumption (Figure 24).
^ PART TWO: FIN FISH CONSUMPTION - SEASON, FREQUENCY, PORTIONS
£3
^SulqectlD^^J
574-00-12345-1
Interviewer Instructions
Display prepared Form photo. Portions can be decimal or whole number
multiples. Enter -7 if Respondent prefers not to tell you frequency or
portion amount. Enter -8 if does not know frequency or portion amount.
Th roug hout Yea r
Prompt
Frequency
TIE
Per Unit
TIF
Prompt
Portion
T1D
How often do you eat King Salmon
(as/in) Container of Fish throughout
the year?
1
Day
-
Looking at the photo, about how
much do you usually eat at a sitting?
l|
I Record: 1 of l
Next Species
Previous
Previous Species
Figure 23. Example of active Previous button
iH PART TWO: FIN FISH CONSUMPTION - SEASON, FREQUENCY, PORTIONS
£3
|SubjectID: jj
574-00-12345-1
Interviewer Instructions
Display prepared Form photo. Portions can be decimal or whole number
multiples. Enter -7 if Respondent prefers not to tell you frequency or
portion amount. Enter -8 if does not know frequency or portion amount.
Throughout Year
Prompt
Frequency
TIE
Per Unit
TIF
Prompt
Portion
T1D
How often do you eat King Salmon
(as/in) Container of Fish throughout
the year?
1
Day
T
Looking at the photo, about how
much do you usually eat at a sitting?
l|
I Record: l of l
Next Species
Previous
Previous Species
Figure 24. Preceding screen to Figure 23
D-77
-------
If the active question is the first in a section or series of questions, the Previous button will be gray
and inactive. In Figure 25, the question Do you mt King Salmon such as Chinook? is the first in the
series of frequency questions for this food item, thus the Previous button is gray and inactive.
In the past 12 months, did you ever eat King Salmon such as Chinook?
Display the species image of King Salmon.
[Subject ID: [574 00 12345 1
Interviewer Prompt
Interviewer Instructions
m PART TWO: FINFISH CONSUMPTION - SEASON, FREQUENCY, PORTIONS
Figure 25. First question for cod
The Next Species button will be active on the last question of a series of questions asked for each
species item. Selecting this button will advance the Interviewer to the next species item in that
group. If the current species is the last item in the group, the interview will proceed to the next
question type.
The Previous Species button will be active for all but the first food item in a group. At any point
during a question, selecting this button will return the user to the first question in the frequency
series for the most recently asked food item.
GoTo The GoTo menu item is accessible during an interview if it becomes necessary to return to a
previously completed section. When this item is selected, a window of completed sections (Figure
26) will be displayed so that the Interviewer can choose a section to move to.
D-78
-------
ED OTHER FIN FISH CONSUMPTION - FREQUENCY, PORTIONS
23
[subject ID: [j =1 MAIN SECTIONS
S3
Interviewer Prompt
Did you eat any oth<
Interviewer Instruct
None.
Ofl Other?
|Subject ID
574-00 12345 1
Please select Survery Section:
Recall Group
Finfish Section
General Section
Conclusion
Next
Cancel
¦ut?
Figure 26. GoTo brings up previously completed sections
Break If it becomes necessary to stop an interview in mid-stream, the Interviewer can select the
Break menu item at the top of the screen. This will cleanly shut down the interview and return the
Interviewer to the main menu. To re-enter an interview that has been "broken off' or completed,
select Edit Interview under the Data Entry menu item and provide the subject interview ID. A
window of previously touched sections will appear. The Interviewer can then select where to begin
again.
Export: Once the interviewing phase is over, you will need to send your data to the Supervisor for
integration and analysis. You should have ready and inserted into a port, a USB device (also known
as a thumb-drive or flash-drive) before you choose this function. The Export function (Figure 27)
will compress the database containing the interview data and copy the "zipped" file to the USB
drive.
Seafood Consumption Survey
Exit Data Entry Export About
Figure 27. Export menu item
You will be prompted to provide the location where the files will be written (Figure 28).
Please choose a file location for exporting the CAPI Data
File Location:
C ose
Figure 28. Prompt to write to USB drive
D-79
-------
Choose the Browse button to help navigate to the USB port. Specify the drive letter that
corresponds to the USB device which in Figure 29 is drive "E:".
Save in: h* USB MEMORY (E:)
My Recent
Documents
Desktop
My Computer
My Network
Places
Name
My Documents
File name:
Save as type: [zip Files ("Zip]
"3 <*= LJ HE-
Size Type
"71
"3]
Date Mc
Save
Cancel
Figure 29. Specify the filename as your site name
Choose Save to export the interview data in compressed format to the USB device. The Export
function will not delete any files already on the device. Send the device to the Supervisor for
processing.
You will only be allowed to export once.
D-80
------- |