RTI/1864/50-03F
November 12, 1982
Final Report
POLYCHLORINATED BIPHENYLS IN
HUMAN ADIPOSE TISSUE AND MOTHER'S MILK
by
R. M. Lucas
V. G. lannacchione
D. K. Melroy
Research Triangle Institute
Research Triangle Park
North Carolina 27709
EPA Contract Number: 68-01-5848
Project Officer: Joseph Carra
Task Manager: Richard Levy
U. S. Environmental Protection Agency
Exposure Evaluation Division
Design and Development Branch
401 M Street, S.W.
Washington, D.C. 20460
RESEARCH TRIANGLE PARK, NORTH CAROLINA 27709
-------
This report was prepared under contract to an agency of
the United States Government. Neither the United States
Government nor any of its employees, contractors,
subcontractors, or their employees makes any warranty,
expressed or implied, or assumes any legal liability or
responsibility for any third party's use or the results
of such use of any information, apparatus, product, or
process disclosed in this report, or represents that its
use by such third party would not infringe on privately
owned rights.
Publication of the data in this document does not signify
that the contents necessarily reflect the joint or separate
views and policies of each sponsoring agency. Mention of
trade names or commercial products does not constitute
endorsement or recommendation for use.
-------
RTI/1864/50-03F
November 12, 1982
Approved by:
Final Report
POLYCHLORINATED BIPHENYLS IN
HUMAN ADIPOSE TISSUE AND MOTHER'S MILK
by
R. M. Lucas
V. G. lannacchione
D. K. Melroy
Research Triangle Institute
Research Triangle Park
North Carolina 27709
EPA Contract Number: 68-01-5848
Project Officer: Joseph Carra
Task Manager: Richard Levy
U. S. Environmental Protection Agency
Exposure Evaluation Division
Design and Development Branch
401 M Street, S.W.
Washington, D.C. 20460
Robert M. Lucas, Task Leader Stephen R. Williams, Project Leader
Research Triangle Institute Research Triangle Institute
-------
Table of Contents
Page
1. EXECUTIVE SUMMARY 1
2. INTRODUCTION 2
3. HUMAN DATA SUMMARY 3
3.1 General Description of the National Adipose
Tissue Survey 3
3.1.1 Objectives of the NHMP Adipose Tissue
Survey 3
3.1.2 Overview of Current Survey Design 3
3.1.3 Overview of Weighting 4
3.1.4 Summary of Polychlorinated Biphenyls
in Human Adipose Tissue 6
3.1.5 Limitations of Statistical Inference ... 7
3.2 Polychlorinated Biphenyls in Human Milk 13
3.2.1 General Description of the Study 13
3.2.2 Overview of the PCBs in Human Milk
Study Design 13
3.2.3 Limitations of Statistical Inferences
for PCBs in Human Milk 18
3.2.4 Summary of Polychlorinated Biphenyls
in Human Milk 18
4. REFERENCES 20
APPENDIX A: Analysis Weighting Methodology A-22
APPENDIX B: Statistical Methods Used in Analysis of
Polychlorinated Biphenyls in Adipose Tissue. . B-32
-------
List of Tables
Table Page
1 Weighted Percentage Distribution of PCB Residue
Levels in Adipose Tissue for all Fiscal Years ... 8
2 Significance Levels of Estimated Trends in PCBs
for Fiscal Years 1972 through 1981 16
3 Significance Levels of Estimated Effects of
Demographic and Geographic Factors 17
4 Summary of Polychlorinated Biphenyls in Human
Milk 19
(11)
-------
List of Figures
Page
Sample Stages of the National Human Monitoring
Program 5
2 Estimated Percentage Distribution of U.S. Popula-
tion by Adipose Tissue PCB Residue Category for
Fiscal 1972 through 1981 9
3 Estimated Trends for Percent of PCB Residue Levels
Detected by Age Group for Fiscal 1972 through 1981. 10
4 Comparison of Estimated and Observed Trends for
Percent of PCB Residue Levels Detected for Fiscal
1972 through 1981 10
5 Estimated Trends for Percent of PCB Residue Levels
Above 3 PPM by Age Group for Fiscal 1972 through
1981 11
6 Estimated Trends for Percent of PCB Residue Levels
Above 3 PPM by Race Group for Fiscal 1972 through
1981 11
7 Comparison of Estimated and Observed Trends for
Percent of PCB Residue Levels Above 3 PPM for
Fiscal 1972 through 1981 11
8 Estimated Percentage Distribution of U.S. Population
by Adipose Tissue PCB Residue Category for Fiscal
1976 through 1981 14
9 Comparison of Estimated and Observed Trends for
Percent of PCB Residue Levels Detected for Fiscal
1976 through 1981 15
10 Comparison of Estimated and Observed Trends for
Percent of PCB Residue Levels Above 3 PPM for
Fiscal 1976 through 1981 15
(ill)
-------
ACKNOWLEDGEMENTS
The authors would like to thank the Task Manager, Mr. Richard Levy
for his assistance and patience in compiling the materials necessary to
produce this report, the Project Officer, Mr. Joseph Carra and Project
Director, Mr. Stephen Williams for the helpful comments in revising
the manuscript, and Ms. Martha Clegg for her efforts in typing the
manuscript.
(iv)
-------
1. EXECUTIVE SUMMARY
This report gives a concise summary of the prevalence and levels of
polychlorinated biphenyls (PCBs) in human adipose tissue and mother's
milk. Data collected from the National Adipose Tissue Survey are reviewed
and summarized. Additional material on PCBs in human mother's milk
supplied by the U.S. Environmental Protection Agency is also included.
The percent of persons in the U.S. with detectable levels and with
levels greater than 3 parts per million (ppm) were investigated to
ascertain changes among the fiscal years 1972 to 1979 and 1981. Different
subpopulations defined by the factors age, sex, race, Census Region and
PCB industries were investigated to ascertain their association with the
PCB levels.
The estimated percentage of U.S. population with detectable levels
of PCBs was found to be increased from approximately 85 percent in
fiscal 1972 to nearly 100 percent in fiscal 1979. (The fiscal 1981
estimates based on a subsample of 97 specimens indicates 100 percent of
the U.S. population have detectable levels of PCB). However, the esti-
mated percentage of the U.S. population with greater than 3 parts per
million PCB in their tissue has increased from approximately 2.7 percent
in fiscal 1972 to 8 percent in fiscal 1976 and then has decreased to
less than 1 percent in fiscal 1981.
Age is the only factor associated with the direction and amplitude
of trends. The percentage of persons with detectable levels of PCBs is
increasing in all age groups but is lower in amplitude in young persons
(0-14 years). The trends in the percentage of persons with greater than
3 parts per million PCB in the two older age categories (15-44 years and
45+ years) are similar to the overall trend. The percentage of persons
in the young age category with greater than 3 parts per million PCB is
nearly constant during the decade and generally lower than the other two
age groups. No other factors were found to be associated with the
general shape of trends, but males and nonwhites tended to have a higher
percentage of persons with greater than 3 ppm PCBs than females and
whites, respectively.
A special study that measured the levels of PCB residue in human
mothers' milk in 1977 was not inconsistence with adipose tissue in
percentage of persons with detectable levels but the magnitude of the
levels was lower than that found in adipose tissue.
-1-
-------
2. INTRODUCTION
This report presents concise but comprehensive summary of the
prevalence and levels of polychlorinated biphenyls (PCBs) in humans and
the environment. The primary sources for this report are previously
prepared preliminary reports on the U.S. Environmental Protection Agency's
(US EPA) monitoring program for human adipose tissue and its computer
accessible data files. Additional material provided by the US EPA on
PCBs in human mother's milk is also included.
Chapter three addresses PCBs in humans. The first section concen-
trates on the National Human Monitoring Program's Adipose Tissue Survey.
That survey is an ongoing program that has been measuring PCBs in human
adipose tissue since fiscal 1972. The second section summarizes the PCB
data measured in human mother's milk collected in a special study spon-
sored by the US EPA. Chapter four includes a list of all documents
referenced in the report.
In order to limit the amount of material presented, only brief
descriptions of the monitoring program, the study of human mother's
milk, and their survey designs are included in the text. Additional
information is contained in the appendices. The reader is advised of
potential limitations in making inferences beyond the sample data and
efforts are made to assess the impact of the limitations on the conclu-
sions. References are included to guide the reader interested in more
detail.
Several terms that have rigorous statistical definitions are used
in the report. Below are generic definitions for these terras.
Target Population: The universe to which an investigator wishes to
make inferences. Examples are the civilian population in the U.S.
or all rural soils in the U.S. to a depth of three inches, as of a
specific point in time.
Sampling Frame: A list or inventory of units in the entire universe
to be sampled. Examples are a list of all counties in the U.S. or
a conceptualized list of all 3-inch cubes of top-soil in the U.S.
Probability Sampling: A process by which each unit on a sampling
frame may be selected for study with a known and positive proba-
bility.
-2-
-------
3. HUMAN DATA SUMMARY
3.1 General Description of the National Adipose Tissue Survey
The National Human Monitoring Program (NHMP) was established in
1967 to monitor levels and prevalence of pesticide residues in the
general population of the United States and to assess changes and trends
in these levels. The program was initiated by the United States Public
Health Service and was transferred in 1970 to the then newly created
United States Environmental Protection Agency (EPA).
The National Adipose Tissue Survey (NATS) is one of the major
ongoing components of NHMP. Specimens of adipose tissue are collected
from surgical patients and autopsied cadavers through the cooperation of
local pathologists and medical examiners. The adipose tissue specimens
are chemically analyzed for twenty chlorinated hydrocarbon compounds
including PCBs. These are listed in Lucas RM et al. 1980. This report
also includes a description and evaluation of the chemical analysis
method. This section includes a summary of all PCB data currently
available from the NATS computer accessible data files. This includes
all data for fiscal 1972 through 1979 and approximately a 10 percent
subsample of fiscal 1981.
3.1.1 Objectives of the NHMP Adipose Tissue Survey
The defined primary objectives of the adipose tissue survey
are:
(i) to measure average concentration of pesticides and
other toxic substances in the general U.S. population;
(ii) to measure time trends of these concentrations; and
(iii) to assess the effects of regulatory actions.
As a secondary objective, the adipose tissue survey
(iv) provides baseline data.
The baseline data may help in detecting and evaluating suspected pesti-
cide poisonings. Besides achieving the above objectives on a national
basis, several subpopulations or domains are of interest. These domains
are sex, race, age subgroups, and geographic regions.
Based on these objectives, the following definition seems appro-
priate as a target population.
Target Population: All non-institutionalized persons in the
United States.
3.1.2 Overview of Current Survey Design
A brief description of the current NATS survey design that
was implemented in fiscal 1977 is given below. The survey design was
-3-
-------
modified twice in the past. A description of all survey designs and
their time periods is given in Hartwell et al. (1979). The modifi-
cations have had little impact on the inferential capabilities of the
network or the ability to make comparisons among the fiscal years.
The specimens of adipose tissue are collected through cooperation
of local pathologists and medical examiners. The complex process,
involving several stages of selection, is illustrated in Figure 1. The
48 conterminous States are stratified into nine geographic regions
coinciding with the Census Divisions. The stratification ensures that
the sample is dispersed across Census Divisions, potentially improving
the ability to make regional estimates and the accuracy of national
estimates. Standard Metropolitan Statistical Areas (SMSAs) within each
Census Division are the primary sampling units (PSUs), and a total of 40
are selected with probability proportional to their population based on
the 1970 Census. The number of SMSAs selected in each Census Division
(given in Appendix A) is proportional to the total population of the
Census Division. Within each SMSA, a cooperating pathologist or medical
examiner is solicited for cooperation and asked to select specimens
according to EPA guidelines (described in Lucas et al. 1980). The
guidelines include a quota for each pathologist based on the age, sex,
and race distribution of the Census Division; instructions for exclusion
of "unrepresentative" patients such as ones suspected of pesticide
poisoning or exhibiting cachexia; and instructions on shipping and
storage.
The sample SMSAs are a valid probability sample of SMSAs. Hence,
it is defensible to present inferences about all SMSAs in the U.S. if
one has accurate information for sample SMSAs. The accuracy and precision
of these inferences, however, also depend on the sampling methodology
within each SMSA. Within each SMSA, the hospitals, medical examiners'
offices and specimens are selected in a non-probabilistic manner based
on the EPA guidelines and judgement of the cooperating professionals.
Useful interpretation of the statistical analysis, requires that we
assume that all the tissue specimens within each SMSA, although they
come largely from cadavers, are "representative" of the target population
of non-institutionalized persons living in the corresponding Census
Division. All statements of significance included in the statistical
analysis below are based on this and other assumption. Other necessary
assumptions are stated in the following sections as appropriate.
3.1.3 Overview of Weighting
Because the PSUs constitute a valid probability sample of
SMSAs, a weight is assigned to each specimen based on the selection
protocol (see Appendix A). Adjustments to each weight for fiscal 1972
through 1979 were then performed to assure that the sum of all weights
within each age (0-14, 15-44, 45+), sex (male, female), and race (white,
nonwhite) category equaled the appropriate population count for the
Census Region according to the U.S. Census. The fiscal 1972 survey
design was based on the 1960 Census data, also, the weight adjustments
were based on 1960 Census data. The survey designs for fiscal 1973
through 1979 were based on 1970 Census data. Because only a small
subsample of fiscal 1981 data is currently available, an aggregate
-4-
-------
Figure 1. Sample Stages of the National Human Monitoring Program
Geographic
Stratification
Select Primary
Sampling Unit
(PSD)
Select Secondary
Sampling Unit
(SSU) within PSU
Select Specimen
within SSU for
Specific Age/
Sex/Race Group
Group the 48 conterminous States
into nine geographic areas which
coincide with the nine Census
Divisions.
Select from two to seven SMSAs from
each Census Division for a total of
40.
Select one (or more) hospitals from
each SUSA.
Select adipose tissue specimen from
cadavers or sugical patients.
-5-
-------
adjustment, which did not take into account the age, sex and race cate-
gories, was performed on the basis of 1980 Census data. The subsample
of fiscal 1981 tissue specimens was controlled to assure that the propor-
tions of specimens in the sample in each category were similar to the
proportions of persons in each category given in the 1980 Census. When
these adjustments were completed, the weights for each fiscal year were
inspected to detect extremely large weights. Large weights were reduced.
The reduction in large weights (trimming) is felt to result in more
accurate population estimates. A detailed description of the weighting
procedure is given in Appendix A.
3.1.4 Summary of Polycnlorinated Biphenyls in Human Adipose Tissue
PCB levels in adipose tissue are classified into four cate-
gories by the chemical analysis laboratory: (i) not detected (ND), (ii)
detected but less than 1 part per million (ppm) (DLT1), (iii) between 1
and 3 ppm (1-3) and (iv) greater than 3 ppm (GT3). Because the data are
categorized in the above manner, it is not possible to express the
individual concentrations on a percent lipid basis. However, it is
possible to adjust statistically the estimates of the percent of persons
in each category. The adjustment is statistically significant but the
magnitude of the adjustment is in consequenital. Hence, all data summar-
ies are on a whole specimen (wet weight) basis, not on a lipid adjusted
basis and exclude specimens with less than 10 percent extractable lipid.
The data are first summarized for all years by three demographic factors:
age, sex, and race and by two geographic factors: Census Region and
industry. The PCB industry factor was defined by grouping the tissue
specimens into three categories: (i) those collected from States con-
taining no PCB industries, (ii) those collected from States containing
one PCB industry, and (iii) those collected from States containing two
or more PCB industries. A list of the States in each Census Region and
industry category is given in Appendix B. Table 1 summarizes the esti-
mated percentage of persons falling in the four residue categories by
the three demographic factors, the two geographic factors and over all
factors. Appendix B contains the same summaries for each fiscal year.
Inspection of the tables in Appendix B revealed apparent changes in
the percentage of the U.S. population in each residue category by fiscal
year. Figure 2 presents a chart of the estimated percentage of the U.S.
population falling in the residue categories ND, DLE3, (DLT1 and 1-3
combined) and GTS for fiscal years 1972 through 1979 and 1981. Several
trends become apparent:
( i) the estimated percent of persons in the U.S. without
detectable levels of PCBs decreased (the percent with
detectable levels increased);
( ii) the estimated percent of persons in the U.S. with
detectable levels less than or equal to 3 ppm increased;
and
(iii) the estimated percent of persons in the U.S. with
PCB levels greater than 3 ppm increased until fiscal 1977
and then began to decrease.
-6-
-------
(The reader is cautioned to note that the fiscal 1981 estimates are
based on a small subsample of 97 tissue specimens that have been chemi-
cally analyzed at this time.)
To quantify the apparent changes over time, trends were estimated
for percent in the GT3 category and for percent detected (PD), which is
the sum of DLT1, 1-3, and GT3. The trends included a linear term (fiscal
year) and a quadratic term (fiscal year squared). In addition to inves-
tigating trends, the statistical analysis (see Appendix B) allows one to
compare the subpopulations for each factor and their changes over time.
The estimated trends are descriptive in nature and intended to give a
smoothed picture of the underlying trends, which may be obscured by year
to year fluctuations in the estimates.
Figures 3 and 4 illustrate the estimated trends in PD. Figure 3
illustrates the trends^, for the three age groups. These trends were
found to be significant and increasing but different. The youngest age
group had the lowest PD in fiscal 1972 but increased more quickly and
approached the two older age groups by fiscal 1981. The trends in the
subpopulations of J:he other factors were statistically tested and found
to be significant , increasing, and equal in both rate of change and
amplitude. Hence, the trend illustrated in Figure 4 adequately describes
the trends for both sexes, both racial groups, the four Census Regions,
and the three industrial categories.
Figures 5, 6, and 7 illustrate the estimated trends in GT3. Figure
5 illustrates the treads for the three age groups. These trends were
found to be significant but different. The two older age groups increase
from fiscal 1972 to 1976 and decrease from fiscal 1976 to 1981 with the
oldest group highest. The youngest age group remains relatively constant.
Figure 6 illustrates the trends for the two. racial groups. The two
racial groups were found to have significant trends of the same shape
but differing in amplitude. The trends in the subpopulations for. the
factors sex and Census Region were also found to be significant but
differing in amplitude like the racial groups. The^trends in the three
industrial categories were found to be significant also but equal in
amplitude when adjusted for the factor Census Region. Because there is
overlap between the two geographic factors, the amplitudes of the trends
for the industrial subpopulations were also tested and found to be
different when not adjusted for the factor Census Region. Figure 7
illustrates the overall trend for GT3.
3.1.5 Limitations of Statistical Inference
The NATS survey samples a surrogate population of surgical
patients and autopsied cadavers instead of the general U.S. population.
This is done for sound ethical reasons. By obtaining specimens from the
sample population no additional risk is assigned to persons selected in
the sample. However, because the sampled population is different from
the target population, one must be willing to assume that the prevalence
*
Assuming no measurement or sampling bias exist.
-7-
-------
Table 1 Weighted Percentage Distribution of PCS Residue Levels
oo
in Adipose Tissues for all
Domain
Sample
Size
Population
Estimate
(in thous . )
Age (years):
0 -14
15-44
45+
Sex:
Male
Female
Race:
White
Non-white
Census Region:
Northeast
North Central
South
West
Number of PCB
Industries per
State:
0
1
2+
Overall
1554
2894
4145
4403
4190
7165
1428
2124
2364
2969
1136
2332
2408
3853
8593
56
85
60
101
101
176
27
48
56
63
34
57
58
86
203
,942
,718
,487
,665
,482
,141
,006
,567
,276
,321
,983
,912
,386
,849
,147
Fiscal
Years
•
PCB Residue Levels
Detected
Not but Less Between I...
Detected Than 1 PPM and 3 PPM"
percent (standard error)
10.5
3.0
3.6
5.1
5.5
4.9
7.7
5.8
5.0
5.5
4.4
5.1
4.4
6.0
5.3
(
(
(
(
(
(
(
(
(
(
(
(
(
(
(
1-6)
0.6)
0.7)
0.8)
0.7)
0.6)
1.9)
1.5)
1.2)
1.1)
1.2)
1.1)
1.1)
1.0)
0.6)
76.7 1
59.0 1
51.11
57.4 1
65.8 1
63.3 1
50.4 1
51.6 1
65.8 1
63.2 1
65.8 1
66.4 I
67.7 1
54.3 1
61.6 1
[ 2.6)
[ 2.2)
( 2.3)
[ 2.3)
[ 2.1)
[ 1.8)
[ 3.9)
( 3.5)
[ 3.1)
[ 3.5)
[ 4.2)
( 3.8)
[ 3.0)
[ 2.6)
[ 1.8)
11.3
31.9
36.1
30.8
24.0
26.5
33.3
33.3
24.7
25.9
26.0
23.8
23.7
32.2
27.4
(
(
(
(
(
(
(
(
(
(
(
(
(
(
(
1.7)
2.0)
2.0)
1-9)
1.8)
1.5)
4.1)
3.3)
2.7)
3.0)
4.1)
3.1)
2.7)
2.4)
1.6)
Greater than
3 PPM»
1.4 1
6.2 1
9.3 1
6.8 I
4.7 1
5.3 1
8.7 1
9.2 1
4.5 1
5.3 1
3.7 1
4.7 1
4.2 1
7.5 1
5.8 1
; 0.5)
; 0.7)
: i.o)
: 0.7)
: 0.6)
[ 0.5)
[ 1.3)
( i.o)
[ 0.9)
( 1.0)
: i.D
[ i.o)
[ 0.8)
[ 0.9)
[ 0.5)
Source: Calculated by the Research Triangle Institute from the NATS computer accessible data files.
* PPM denote parts per million.
-------
Figure 2.
Estimated Percentage Distribution of U. S. Population
by Adipose Tissue PCB Residue Category for Fiscal 1972 Through 1981
* 1981 Is a sub-sample of 97 observations.
LEGEND: ND •= Not Detected DLE3 - Detected and Less Than or Equal to 3 PPM GT3 « Above 3 PPM
-------
Figure 3.
133-
Q
t
e
c
t
e
95-1
P
e
r
c
n 90
t
85-
80-
75-
Eatimoted Trends for Percent of PCS Residua Levels Detected
by Ago Group for Fiscal 1972 Through 1981
72 73 74
LEGEND: AGE
75
76 77 78 79 80
F i sea I Year
0-14 15-44 45-1-
Flgure 4. Comparison of Estimated and Observed Trends for Percent of PCS Residue Levels Detected
for Fiscal 1972 Through 1981
75
78
76 77
F I sea I Y
-------
rigur« 5.
15-
a-
Ljtimotia Trtnoi lor Pircinl of PCS R.irtuo Urtil *boy» 1 FPU
by Ago Croup lor Fiscal 1972 Througn 1981
\
72
73
7<
LEGEND. AGE
75
76 77
Fi.col Y.or
78 79 88 81
Ftguri 6.
ClHmotM Tr.noi lor P«r«nt of PCS R.rtu. Lonli *bo». 3 PPU
by Roe* Croup for Tbcfll 1972 Througn 1981
\
73
7S
LEGEND. RACE
78 77
F i .en I Y.
-------
and levels of the substances of interest (chlorinated hydrocarbon com-
pounds) are the same in both populations (age, sex, race adjusted) to
make inferences from the sampled population to the target population.
-Other limitations result from the survey design and its implementa-
tion. Because cities with populations greater than 25,000 persons or
SMSAs form the sampling frame, areas that are rural in character are
excluded from the sample. A sizable percent (38.3 percent in 1970
(Lucas RM et al. 1980) of the U.S. population resides in these rural
areas. The real percentage of omission is most likely less, however,
because the service areas of hospitals in cities -or SMSAs extend into
neighboring rural areas. Obviously, the urban nature of the sample
results in an under coverage of the population of interest.
Limitations also result from the purposeful selection of cooperating
professionals and the lack of probability sampling used by them in
selecting tissue specimens within the PSUs. The purposeful selection
methods may result in biases in estimates derived from the sample. In
an effort to minimize potential biases, the data collected in the network
were assigned weights to be used in the statistical analysis. The
weights can be thought of as the number of persons in the target popula-
tion that each sample specimen represents. The reader is cautioned that
even though the weighting is felt to reduce bias; due to the lack of
probability sampling within the SMSAs, this bias is not necessarily
overcome by the weighting procedure. Even the estimates based on the
weights may be substantially biased.
The chemical analysis methods are another potential source of bias.
A detailed description of the current chemical analysis methods for PCBs
is given in Lucas R M et al. (1980). The report also gives a comprehen-
sive evaluation of the current analytical method which includes extraction
according to the modified Mills-Olney-Gaither procedures (USEPA 1974
Section 5A[1]) and analyzed by gas chromatography with electron capture
detection (GC/ECD) on two columns. The report (Lucas R M et al. 1980)
also contains a brief assessment of the analytical method, thin layer
chromatography (TLC) (USEPA 1977 and Mulhern et al. 1971) used prior to
November 1974. The measurement errors for the techniques are quite
large. For the TLC method the error is estimated to be ±50 percent
(Lucas R M et al. 1980) and for the GC/ECD method ± 50 percent is the
best-case estimate and ±100 percent is felt to be a reasonable estimate
of the measurement error in this particular application. The measurement
errors can be partitioned into two components, systematic errors that
bias estimates and random errors that affect the precision of estimates.
The statistical analysis techniques properly incorporate the random
component in the estimates of precision. However, any systematic errors
that may exist could not be accounted for in the analysis.
To obtain some insight into how biases may affect the conclusions
with regard to trends and differences among demographic or geographic
groups, additional material is presented below. First the trends are
investigated for fiscal 1976 through 1981 to assess the effect of the
change in chemical analysis.
-12-
-------
To investigate the impact on trends resulting from the change in
chemical analysis methods, trends were estimated for the categories PD
and GT3 using only fiscal 1976 to 1978 and the 1981 subsample. Figure 8
presents the estimated percentage of the U.S. population falling in the
three residue categories ND, DLT3 and GT3.
Figures 9 and 10 present the estimated trends for the residue
categories PD and GTS respectively. The trends illustrated in these
figures are consistent with the trends estimated using all nine fiscal
years of data. Hence, the change in chemical analysis method does not
have a noticable impact on the estimated trends for fiscal 1976 through
1981. Because the analytical technique did not change during the last
six years, the minimal detectable level should not substantially change,
and thus, should have little effect on the PD trend.
J.
The trends are highly significant (the probability of no trend iSj.
very small) and the two racial groups and sexes are highly significantly
different for GTS. Tables 2 and 3 summarize the results of the statis-
tical tests. Because of the relatively high levels of significance
here, it is unlikely that actual biases in the population estimates
would change the conclusion about trends or race and sex group differ-
ences. Biases are more likely to have an effect on the conclusion about
the different trends for the age groups in PD and for differences among
the Census Regions. A table is given in Appendix B illustrating the
effect of bias on the significance level of statistical test.
The potential for bias is a two-edged sword. Biases may mislead
the investigator into making false conclusions, but biases in the data
may also hide real differences in the population and mislead the investi-
gator into believing that no differences exist when in fact they do. It
is impossible to estimate the magnitude or direction of biases in the
present situation, if they exist, with the available information.
3.2 Polychlorinated Biphenyls in Human Milk
3.2.1 General Description of the Study
This section summarizes the data present in a special pro-
ject report on PCBs in human milk (SPR 1977). The survey design is
described in Savage et al. (1981) and was originally targeted at five
chlorinated hydrocarbon insecticide residues. Approximately 72 percent
of the milk specimens had sufficient volume to permit a second chemical
analysis for PCBs. The chemical analysis method is described in SPR
1977.
3.2.2 Overview of the PCBs in Human Milk Study Design
A total of 163 hospitals was randomly selected from a list
of approximately 7,000 general care hospitals in the United States. The
hospitals were grouped into five geographic regions: (i) Northeast,
(ii) Southeast, (iii) Midwest, (iv) Southwest, and (v) Northwest. From
the selected hospitals, a subsample of nursing mothers who had given
*
Assuming no measurement or sampling bias.
-13-
-------
Figure 8. Estimated Percentage Distribution of U.S. Population
by Adipose Tissue PCB Residue Category for Fiscal 1976 Through 1981
BLOCK CHART OF PERCENT
Fiscal Y«or
*
1981
£17
a
1878
1878
1977
1976
99
90
89
ie
90
NO
PLE3 6T3
L.v.l of PCB
* 1981 Is a sub-sample of 97 observations.
LEGEND: ND ~ Not Detected DLE3 •» Detected and Less Than or Equal to 3 PPM GT3 - Above 3 PPM
-------
Figure 9. Comparison of Estimated and Observed Trends for Percent of PCS Residue Levels Defected
for Fiscal 1976 Through 1981
I 00.0-
77 78 79 80
FI sea I Year
LEGEND: • Observed Estimated
81
Figure 10. Comparison of Estimated and Observed Trends for Percent of PCB Residue Levels Above 3 PPM
for Fiscal 1976 Through 1981
10.0-J
76
77
78 79
F I sea I Year
LEGEND: • Observed - Estimated
-15-
80
-------
Table 2. Significance Levels of Estimated Trends in PCBs
for Fiscal Years 1972 through 1981**
Probability of No Trend
Trend Percent Percent Greater
Detected Than 3 ppm
Linear Term .0000 .9519
Quadratic Terra .0017 .0000
*
Probability of Equal Trends
Among Age Groups
Linear Terra .0307 .3753
Quadratic Term .0857 .0000
* Assuming no measurement or sampling bias.
** Source: Calculated by the Research Triangle Institute from the
National Adipose Tissue Survey sponsored by the U.S.
Environmental Protection Agency.
"V
NS Not significant , probability > .05.
-16-
-------
-.V
Table 3. Significance Levels of Estimated Effects of
Demographic and Geographic Factors**
*
Probability that the Subpopulations within
Each Factor are Equal
Factor
Age Group
Race
Sex
Census Region
Industry
Percent Detected
NA
.1362
.8267
.8725
.1157
Percent Greater than 3 ppm
NA
.0005
.0003
.0397
.4373
* Assuming no measurement or sampling bias and adjusted for the o.ther
factors.
** Source: Calculated by the Research Triangle Institute from the
National Adipose Tissue Survey sponsored by the U.S.
Environmental Protection Agency.
NA Because the difference among age groups changes by fiscal year, it
is not appropriate to test for this effect.
-17-
-------
birth in the selected hospitals were solicited for cooperation. Milk
specimens were collected from 1,436 participants.
3.2.3 Limitations of Statistical Inferences for PCBs in Human Milk
The available material is not sufficiently detailed to fully
evaluate the survey design or to incorporate the design into the statis-
tical analysis. Also, the statistical method used to summarize the PCB
data presented in SPR 1977 are not described. Hence, there is insuffi-
cient information to fully evaluate the limitations of making inferences
from the sample data to the target population of nursing mothers giving
birth in U.S. general care hospitals during the study period. However,
one important limiting factor is the fact that a nonrandom subsample
(only specimens with sufficient volume for the second chemical analysis)
of 1,033 milk specimens were analyzed for PCBs from the original 1,436.
This admits the possibility of biases which might result in making
inferences to the intended target population. Hence, statements that
the levels measured in the sample are indicative of those one would find
in the target population may be misleading.
3.2.4 Summary of Polychlorinated Biphenyls in Human Milk Data
To summarize the levels of PCBs in human milk, the measured
values are grouped into the three categories: (i) not detected (NTJ),
(ii) trace (detected but less than 0.05 parts per million (ppm)), and
(iii) positive (greater than 0.05 ppm). Table 4 contains these summar-
ies by the five geographic regions (not Census Regions) and overall.
The Northeast and Southeast have the highest percent positive (38.2 and
37.7 percent respectively) and the Southwest region the lowest (21.5
percent). The maximum positive value (not adjusted for percent lipid)
was 0.563 ppm in Texas in the Southwest region. In comparing this human
milk data with the adipose tissue data, both similarities and differences
are noticable. The percent of specimens containing detectable^ levels
(99.1 percent) are similar, but the levels in milk are much lower than
the levels in adipose tissue.
The data given in SPR (1977) also included the positive amounts
adjusted for percent lipid. These adjusted data were not included in
this report because of noncomparability with the adipose tissue data
that could not be directly adjusted for percent lipid.
-18-
-------
Table 4. Summary of Polychlorinated Biphenyls in Human Milk
Region
Levels
(parts per
million)
Not Detected
Trace
Positive
Sample Size
Northeast
percent
2.2
59.7
38.2
186
Southeast
percent
0.0
62.3
37.7
159
Midwest
percent
1.0
70.6
28.4
289
Southwest
percent
0.7
78.5
20.9
297
Northwest
percent
0.0
67.6
32.4
102
Total
0.9
69.3
29.8
1033
Source: Calculated by the Research Triangle Institute from data provided
by the U.S. EPA (SPR 1977).
_i_i_
Greater than the minimum quantifiable level of 0.05 parts per million
in whole milk.
-19-
-------
4. REFERENCES
Hartwell T, Piserchia P, White SB et al. 1979. Analysis of EPA Pesti-
cides Monitoring Networks. Prepared by the Research Triangle Institute
for the Office of Toxic Substances. U.S. Environmental Protection
Agency. Washington, DC 20460. EPA-560/13-79-014.
Inglis A 1982. Personal communication with Anthony Inglis of the Expo-
sure Evaluation Division, Office of Toxic Substances, U.S. Environmental
Protection Agency. Washington, DC 20460.
Lucas D, Mason RE, Rosenzweig M et al. 1980. Recommendations for the
National Surface Water Monitoring Program for Pesticides. Research
Triangle Institute, Research Triangle Park, NC 27709.
RTI/1864/14/01-01I.
Lucas RM, Rosenzweig MS, Williams SR et al. 1980. Evaluation of and
Alternate Designs for the National Human Monitoring Program's Adipose
Tissue Survey. Research Triangle Institute, Research Triangle Park, NC
27709. RTI/1864/14/02-02I.
Lucas, RM, Erickson, MD and Williams, SR 1980B. PCB Residue Levels in
Human Adipose Tissue: A Statistical Evaluation By Racial Grouping.
Prepared by the Research Triangle Institute for the Office of Toxic
Substances, U. S. Environmental Protection Agency. Washington, DC.
EPA/13-79-015.
McDonald M, Hardy M, Drummond D 1981. National Soil Monitoring Program:
Urban Soils Monitoring Network. Research Triangle Park, NC: Research
Triangle Institute. RTI/1864/14/03-02I.
Mulhern BM, Cromartie E, Riechel WE, Belisle AM. 1971. JAOAC 54:
548-550.
SPR 1977. Special Project Report: Polychlorinated Biphenyls in Human
Milk. (Unpublished document describing analysis of PCB from a subsample
of the milk specimens collected in Savage et al. 1981) Provided to the
Research Triangle Institute by the U.S. Environmental Protection Agency.
Savage EP, Keefe TJ, Tessari JTJ et al. 1981. National Study of Chlori-
nated Hydrocarbon Insecticide Residues in Human Milk, U.S.A. Am J
Epidemiol 1981; 113:1413-22.
Schmitt, CJ, Ludke, JI and Walsh, D (1981). Organochlorine residues in
fish, 1970-1974, National Pesticide Monitoring Program. Pestic. Monit.
J. 14(4): 136-206.
USDA 1966. National Handbook for Updating the Conservation Needs Inven-
tory United States Department of Agriculture.
USEPA 1977. U.S. Environmental Protection Agency. Analysis of pesti-
cide residue in human and environmental samples, section 9D. Research
Triangle Park, NC. U.S. Environmental Protection Agency.
-20-
-------
USEPA 1974. U.S. Environmental Protection Agency. Analysis of pesticide
residue in human and environmental samples. Research Triangle Park, NC:
U.S. Environmental Protection Agency.
USGA 1980. National Water Data Exchange. U.S. Geological Survey.
Reston, Virginia.
Whitmore R, Rosenzweig M, and Hines J 1981. National Soil Monitoring
Program. Research Triangle Park, NC: Research Triangle Institute.
RTI/1864/14/03-01I.
-21-
-------
Appendix A
Sampling Weighting Methodology
-------
Appendix A.
Sample Weighting Methodology
A.1 Introduction
The true probability of selection for specimens collected in the
adipose tissue network could not be calculated because some stages of
the sample selection involved nonprobability sampling. Even if the
selection probabilities of a specimen within sample cities were assumed
to be equal, the sample design of the NHMP adipose tissue network did
not give a self-weighting sample with regard to the living population
because of discrepancies between samples and the quotas. Therefore, a
weight was calculated for each specimen that reflects its approximate
•(V
probability of selection . Including these weights in the analysis may
reduce bias in estimating means or proportions. The following para-
graphs describe the procedures used for computing approximate weights
for the NHMP data.
A.2 Initial Sample Weights
The calculation of the initial sample weights considered two stages
of selection. In the first stage, primarily sampling units (PSUs) were
selected within each stratum with probability approximately proportional
to their population. Table A.I gives the number of PSUs specified by
the survey design for the fiscal years. In the second stage, samples of
cadavers and surgical patients were selected in a nonprobabilistic
manner from a hospital(s) or other facility located in the sample cities.
*
Equal probabilities of selection were assumed for this stage. Table
A.2 gives the number of tissue specimens included in the statistical
•
analysis by fiscal year.
Probabilities of selection at the final stage are with reference to a
sampling rate for the living population rather than for the population
of cadavers and surgical patients.
A-23
-------
Table A.I. Number of Primary Sampling Units Per Stratum
by Fiscal Year as Specified by Survey Design*""
Fiscal 1972
Census Region
Northeast
North Central
South
Westt
Total
Fiscal 1973-1981
Census Division
Number
11
12
9
_7
39
Fiscal 1973-1976
Number
Percent
Percent
28.2
30.8
23.1
17.9
100.0
Fiscal 1977-1981
Number
Percent
New England
Middle Atlantic
East North Central
West North Central
South Atlantic
East South Central
West South Central
Mountain
Pacific!
Total
4
14
15
6
11
5
7
3
10
75
5.3
18.7
20.0
8.0
14.7
6.7
9.3
4.0
13.3
100.0
2
7
8
3
6
3
4
2
5
40
5.0
17.5
20.0
7.5
15.0
7.5
10.0
5.0
12.5
100.0
* For fiscal years 1972 through 1976 the primary sampling units were
cities with populations over 25,000 persons and from fiscal 1977
to 1981 the PSUs were Standard Metropolitan Statistical Areas
(SMSAs).
** Source: Hartwell et al, 1979.
t Excludes Alaska and Hawaii.
A-24
-------
Table A.2. Number of Tissue Specimens Used in
Statistical Analysis by Fiscal Year
Fiscal Year Number
1972 1826
1973 1183
1974 1025
1975 895
1976 779
1977 927
1978 951
1979 910
1981 97
Total 8593
Source: Calculated by the Research Triangle Institute from the
National Adipose Tissue Survey's computer accessible data files
sponsored by the U.S. Environmental Protection Agency.
A-25
-------
The initial sample weight for specimen (j) in PSU (i) and stratum
(h) was calculated as the inverse product of the approximate probabilities
*
of selection for each stage and is expressed as
W1(hiJ)= [PhiPj|hir1'
where P denotes the probability of selecting the i-th PSU in the h-th
stratum (for large PSUs selected with probability 1 this is more accu-
rately described as the expected number of times the PSU is selected in
the sample); and P.i. . denotes the sampling rate of specimens to the
living population given PSU (i) in stratum (h).
The PSUs were selected independently within each stratum. For each
stratum, the cumulative total population of eligible PSUs was divided by
the number of PSUs to be selected in the h-th stratum. This calculation
gave the sample selection interval. A random start was obtained by
selecting a random number between "1" and the value of the selection
interval. The method then involved listing the cities in a random
order, calculating the cumulative totals and selecting cities by matching
the cumulative totals to the random start and integer multiples of the
selection interval plus the random start (USEPA 1973).
The above can be expressed algebraically in the following manner.
For each stratum, the selection interval is
:h = Vv
where I, , N, , and m, denote the selection interval, cumulative pop-
h h h
ulation total, and number of PSUs to be selected for the h-th stratum,
respectively. The PSUs selected in the sample are those for which the
cumulative totals match r + K x I for K=0, mh~1( wnere rh is tne
random start for the h-th stratum.
y-
Probabilities of selection at the final stage are with reference to a
sampling rate for the living population rather than for the population of
cadavers and surgical patients.
A-26
-------
This method assigned a probability of selection to each PSU equal
to the population of the PSU divided by the selection interval. This
can be expressed algebraically as
P.. = N../L.
hi hi h'
where N, . denotes the population of the PSU (i) in stratum (h) and P..
and I, are defined above.
h
Under the assumption made above for selection of samples within a
PSU, the calculation is straightforward. The pseudo probability of
selection is given by dividing the number of specimens selected in each
PSU by the population of the PSU. This can be expressed algebraically
as
P . ., . = n, . /N, .
j|hi hi hi
for j=l, ..., n, ., were n, . denotes the number of samples collected in
the i-th city in the h-th stratra and P.,, . and N, . are defined above.
3 jI hi hi
The initial weight can be written as
Wl(hij) = [N^xn^/N^r1,
which simplifies to
Wl(hij) = yn^
Hence, the approximate initial sampling weight for the PSU (i) in the
stratum (h) is the same for all samples collected in the city and is
calculated by dividing the selection interval for the h-th stratum by
the number of specimens collected in the city.
The method described above does yield selection probabilities
proportional to the population of the PSUs when certain precautions are
followed. Adequate methods for selecting alternate (substitute) cities
were not taken. Hence, the procedures followed in the study alter the
probability of selection for some PSUs, but the degree, although probably
A-27
-------
minor, is undetermined. For the purpose of calculating approximate
weights, these special situations were treated the same as all others.
A.3 Missing Data Adjustments and Poststratification
Survey estimates of means and proportions may be biased if specimens
that were identified for selection, but not collected, had different
levels of PCB residue than those actually collected. Furthermore,
estimates of population totals will be underestimated unless some allow-
ance is made for the loss of the missing data. Therefore, the initial
sample weights were adjusted by stratum so that the sum of the weights
equaled national Census population counts in each stratum. This included
only those data that were suitable for statistical analysis. Data was
classified as suitable for analysis by percent extractable lipid (> 10
percent) and valid chemical analysis confidence codes.
The missing data adjustment process began with the definition of a
response indicator for specimen (k) of the form
RESP(k) =
1 if specimen (k) had data suitable
for analysis
0 otherwise.
Missing data occurred at different rates among demographic domains.
Therefore, ratio adjustments to the initial weights were made within
demographic domains referred to as poststrata. These poststrata were
formed by crossing some or all of the following classification vari-
ables :
Fiscal year: 1972,...,1979,1981
Census Region: North East, North Central, South, West
Age Group: 0-14, 15-44, 45+
Race: White, Non-White
Sex: Male, Female
A-28
-------
Once a poststratum (£) was formed, N(£), the Census population count for
that poststratum, was obtained. The Census counts used depended on the
fiscal year the sample data were obtained. Fiscal 1972 used 1960 Census
data (US Census 1960), while fiscal 1973 through fiscal 1979 used 1970
Census data (US Census 1970). The fiscal 1981 sample was partitioned
into poststrata formed by the nine Census Divisions and used 1980 Census
data (US Census 1980).
The Census counts for each poststratum were divided by the weighted
number of tissue specimens to form a poststratum adjustment factor:
ADJU) = N (2) / [I Wl(k) • RESP(k)],
ke£
where ke£ indicates that the summation is over all specimens belonging
to poststratum (£).
Each responding specimen (k) received an adjusted sample weight,
W2 (k), which was its poststratum adjustment factor times its initial
sample weight:
W2(k) = ADJ(£) ' Wl(k).
Hence, within a poststratum, the sum of the adjusted sample weights
equaled the corresponding Census population count.
Excessive sampling variances may be caused by unequal weighting ef-
fects. Therefore, a maximum increase of fifty percent over the standard
deviation of an equally weighted sample was imposed on each of the nine
fiscal year samples. Because fiscal 1972, 1973, and 1976 had unacceptable
unequal weighting effects under this rule, the following calculations
were undertaken to reduce the extreme values among the weights for these
fiscal years.
First, the poststratum containing the largest adjusted weight in
the fiscal years was found. Next, this largest adjusted weight was set
A-29
-------
equal to the second largest adjusted weight within the poststratum.
Then, the difference between these two largest weights was spread equally
over the remaining adjusted weights in the poststratum. This weight
trimming process reduced the unequal weighting effect for fiscal 1973
and 1976 to below 50 percent. The unequal weighting effect for fiscal
1972 was still 57 percent after trimming, but was considered acceptable
since there were several large weights. The adjusted weight sums for
the affected poststrata remained equal to their Census population counts
and the resulting mean square errors of estimation should be reduced
using the trimmed weights.
A-30
-------
A. 4 References
USCENSUS. 1980. U.S. Census of Population, Supplementary Reports.
PC80-51-1, Table 2.
USCENSUS. 1970. U.S. Census of Population, U.S. Summary PC(1) IB, Table 57.
USCENSUS. 1960. US Census of Population, U.S. Summary PC (1) IB, Table 52.
USEPA. 1973. U.S. Environmental Protection Agency. Listing of eligible
cities with cumulative populations, selection intervals, and random
start numbers. Washington, D.C.: Office of Pesticides and Toxic Sub-
stances, U.S. Environmental Protection Agency.
A-31
-------
APPENDIX B
Statistical Methods Used in Analysis of Polychlorinated
Biphenyls in Adipose Tissue
-------
Appendix B
Statistical Methods in Analysis of Polychlorinated
Biphenyls in Adipose Tissue
B.1 Statistical Analysis of Polycfalorinated Biphenyls in Human
Adipose Tissue
The statistical analysis that was performed on PCB residue levels
used the NATS computer accessible data files and software that was
specifically developed to take into account the survey design and weights.
Estimates of the percentage of the U.S. population in the PCB residue
categories for geographic, demographic, and overall categories are
produced. These estimates were computed using a software package called
SESUDAAN (Shah 1979). Tables B.I through B.9 summarize these estimates
by fiscal year. Table 2 in chapter three summarizes these estimates
over all fiscal years.
To investigate differences among the categories for PD and percent
GTS and their changes over time, statistical models were developed using
a software package called SURREGR (Holt 1977). Before the models are
discussed, it is important to note that because of the limitations of
the survey design, implementation and chemical analysis, all statements
of statistical significance are asterisked. The purpose is to remind
the reader that unverified assumptions about the survey design were
required to assign weights to the specimens collected in NATS and to
estimate the precision of the estimates. Additional reservations about
the statistical significance result from the potential bias in population
estimates that may result from the chemical analysis limitations.
B-33
-------
Table B.I Weighted Percentage Distribution of PCB Residue Levels
in Adipose
Domain
Sample
Size
Tissues
for Fiscal Year 1972
Population
PCB Residue Levels
Estimate
(in thous . )
Age (years):
0 -14
15-44
45+
Sex:
Male
Female
Race:
White
Nonwhite
Census Region:
Northeast
North Central
South
West
Number of PCB
Industries per
State:
0
1
2+
Overall
223
446
1157
1029
797
1501
325
630
476
486
234
462
419
945
1826
55
70
52
89
89
159
20
44
51
54
28
69
32
77
179
,786
,920
,626
,990
,342
,081
,252
,678
,628
,973
,053
,863
,337
,133
,332
Not Less than
Detected 1 PPMt
percent (standard
11.7
7.0
9.8
8.0
10.6
9.0
11.5
11.2
13.1
6.1
5.3
6.2
10.4
11.6
9.3
( 3.3)
( 2.1)
( 2.7)
( 1.5)
( 2.0)
( 1.6)
( 3.9)
( 3.5)
( 3.6)
( 2.5)
( 1.7)
( 2.4)
( 3.0)
( 2.5)
( 1-6)
42.2
24.6
22.0
27.6
31.0
29.5
27.4
15.9
35.8
28.0
41.3
33.8
30.0
24.9
29.3
( 6.6)
( 4.1)
( 3.2)
( 4.3)
( 3.5)
( 3.2)
( 6.3)
( 6.1)
( 6.0)
( 5.0)
(12.4)
( 5.0)
( 9.4)
( 5.0)
( 3.3)
Between 1
and 3 PPMt
error)
44.2
63.3
63.4
58.3
56.5
57.6
55.8
67.7
48.1
60.1
52.8
55.4
55.9
59.8
57.4
( 4.9)
( 4.0)
( 3.8)
( 3.6)
( 3.9)
( 3.1)
( 6.1)
( 8.0)
( 3.0)
( 3.4)
(12.6)
( 3.0)
( 7.8)
( 6.0)
( 3.1)
Greater than
3 PPMt
2.0 I
5.1 1
4.8 {
6.2 1
1.9 I
3.9 1
5.3 1
5.3 1
2.9 1
5.8 1
.6 1
4.6 1
3.8 1
3.6 1
4.0 I
[ 1-4)
[ 1.7)
[ 0.9)
[ 1-9)
[ 0.6)
[ 1-1)
[ 2.0)
[ 2.8)
[ 1.2)
[ 2.4)
[ 0.3)
[ 2.0)
[ 1.5)
[ 1.7)
( l.D
t parts per million
^Standard error not estimable.
Source: Calculated by the Research Triangle Institute from the National Adipose
Tissue Survey's computer accessible data files.
-------
Table B.2 Weighted Percentage Distribution of PCB Residue Levels
at
I
u>
Ul
in Adipose
Domain
Sample
Size
Ti ssues
for Fiscal Year 1973
Population
PCB Residue Levels
Estimate
(in thous . )
Age (years):
0 -14
15-44
45+
Sex:
Male
Female
Race:
White
Nonwhite
Census Region:
Northeast
North Central
South
West
Number of PCB
Industries per
State:
0
1
2+
Overall
188
419
576
549
634
1026
157
204
359
439
181
282
390
511
1183
57
83
61
101
102
176
26
49
56
62
34
52
61
89
203
,900
,A37
,875
,12A
,088
,3A5
,867
,041
,572
,795
,804
,AA8
,080
,684
,212
Not Less than Between 1
Detected 1 PPMt and 3 PPMt
percent (standard error)
26.8
13.6
15.4
18.0
17.9
15.8
31.8
19.0
11.1
24.1
16.5
21.5
13.2
19.1
17.9
( 5.9)
( 3.7)
( 3.8)
( 4.0)
( 3.7)
( 2.9)
(10.3)
(11.1)
( 3.7)
( 5.8)
( 7.3)
( 6.7)
( 3.5)
( 6.6)
( 3.6)
61.3
45.9
33.6
38.7
54.3
49.1
29.7
34.5
51.9
47.4
53.2
49:5
53.2
40.3
46.5
( 6.4)
( 4.3)
( 3.8)
( 3.9)
( 4.1)
( 3.6)
( 7.0)
(11.2)
( 5.7)
( 4.7)
( 6.0)
( 6.7)
( A. 3)
( 6.3)
( 3.5)
9.7
36.4
41.8
37.1
23.8
30.8
27.7
39.7
32.1
23.6
27.0
26.2
28.1
34.5
30.4
( 2.8)
( A. 2)
( 3.7)
( A.I)
( 2.9)
( 3.2)
( 5.3)
( 9.2)
( 6.3)
( 3.0)
( A. 6)
( A. A)
( 5.5)
( 5.2)
( 3.0)
Greater than
3 PPMt
2.1 1
A.I {
9.2 1
6.2 1
A.O (
A. 2 I
10.8 (
6.8 I
5.0 1
A. 9 1
3.3 1
2.9 1
5.5 1
6.1 1
5.1 1
: 1.4)
: 1.3)
: 1.6)
: 1.3)
: i.o)
: 0.9)
: 2.8)
: 2. A)
: 2.3)
; i.D
: 2.1)
: I.D
: 2.1)
[ 1.5)
( 1.0)
t parts per million
"•^Standard error not estimable.
Source: Calculated by the Research Triangle Institute from the National Adipose
Tissue Survey's computer accessible data files.
-------
Table B.3 Weighted Percentage Pi stribut ion of PCB Residue Levels
td
I
a.
in Adipose Tissue
Domain
Sample
Size
for Fiscal Year 1974
Population
PCB Residue Levels
Estimate
(in thous.)
Age (years):
0 -14
15-44
45+
Sex:
Male
Female
Race:
White
Nonwhite
Census Region:
Northeast
North Central
South
West
Number of PCB
Industries per
State:
0
1
2+
Overall
207
362
456
521
504
868
157
225
304
374
122
242
306
477
1025
57
83
61
98
104
177
26
49
56
62
34
49
64
89
203
,900
,437
,875
,912
,300
,066
,146
,041
,572
,795
,804
,439
,057
,716
,212
Not Less than Between 1
Detected 1 PPM t and 3 PPM t
percent (standard error)
19.1 I
5.5 I
6.3 1
9.8 1
9.5 1
9.2 1
12.8 1
13.5 1
13.4 1
6.5 1
3.9 1
12.0 1
8.0 I
9.5 I
9.6 I
( 5.6)
[ 1.9)
[ 2.6)
[ 2.9)
( 2.7)
[ 2.2)
[ 5.0)
[ 5.4)
[ 5.3)
( 2.4)
( 2.2)
( 5.3)
( 2.9)
( 3.4)
( 2.2)
72.9
50.7
41.7
49.3
59.1
55.5
46.0
40.7
55.0
57.4
66.7
60.3
57.2
48.9
54.3
( 6.5)
( 4.2)
( 4.3)
( 3.5)
( 3.3)
( 2.8)
( 7.3)
( 4.9)
( 4.8)
( 3.6)
( 8.9)
( 6.7)
( 2.9)
( 4.5)
( 2.6)
8.0
38.6
43.8
36.9
26.3
30.9
35.4
37.8
28.7
31.2
27.4
22. 4
32.3
35.8
31.4
(1.9)
(4.2)
(3.9)
(4.1)
(2.9)
(2.9)
(5.7)
(6.5)
(4.5)
(3.5)
(9.2)
(4.3)
(3.9)
(5.1)
(2.8)
Greater than
3 PPMT
0.0
5.2
8.1
4.0
5.2
4.5
5.7
8.0
3.0
4.9
2.0
5.3
2.4
5.8
4.6
( * )
( 1-6)
( 2.5)
( 1.0)
( 1.9)
( 1-3)
( 2.0)
( 4.1)
( 1.4)
( 1.7)
( 1.1)
( 2.3)
( 0.8)
( 2.3)
( 1.2)
t parts per million
"'Standard error not estimable.
Source: Calculated by the Research Triangle Institute from the National Adipose
Tissue Survey's computer accessible data files.
-------
Table B.5 Weighted Percentage Distribution of PCB Residue Levels
in Adipose Tissue
Domain
Sample
Size
for Fiscal Year 1976
Population
PCB Residue Levels
Estimate
(in thous.)
Age (years):
0 -14
15-44
45+
Sex:
Male
Female
Race:
White
Non-white
Census Region:
Northeast
North Central
South
West
Number of PCB
Industries per
State:
0
1
2+
Overall
165
248
366
397
382
642
137
229
205
254
91
181
216
382
779
57
83
61
97
105
177
25
49
56
62
34
42
63
96
203
,900
,437
,875
,391
,821
,887
,325
,041
,572
,795
,804
,997
,566
,649
,212
Not Less than Between 1
Detected 1 PPMt and 3 PPMt
percent (standard error)
4.5
0.5
0.0
1.8
1.3
1.2
3.7
0.1
2.1
1.4
2.6
0.3
1.7
1.9
1.5
(
(
(
(
(
(
(
(
(
(
(
(
(
(
(
1.8)
0.5)
* )
0.8)
0.8)
0.7)
2.5)
0.1)
1.9)
1.0)
1.0)
0.3)
1.0)
1.1)
0.6)
86.6
56.3
52.4
56.5
70.4
64.2
60.3
47.4
77. 1
62.5
67.2
81.8
69.3
52.0
63.7
( 3.4)
( 2.6)
( 5.1)
( 4.2)
( 3.7)
( 2.8)
( 6.5)
( 3.0)
( 6.1)
( 5.4)
(10.1)
( 3.6)
( 6.1)
( 3.4)
( 2.8)
7.7
35.7
34.1
32.4
22.5
27.8
23.4
37.0
19.5
29.5
22.2
15.5
25.2
33.9
27.3
( 2.6)
( 2.9)
( 4.7)
( 3.6)
( 3.6)
( 2.6)
( 5.1)
( 5.0)
( 5.7)
( 4.6)
( 5.6)
( 3.2)
( 5.3)
( 3.4)
( 2.5)
Greater than
3 PPMt
1.2
7.4
13.4
9.3
5.8
6.8
12.6
15.5
1.3
6.5
8.0
2.4
3.8
12.1
7.5
( 0.8)
( 2.2)
( 3.0)
( 2.1)
( 2.0)
( 1.5)
( 3.6)
( 3.4)
( 0.6)
( 2.5)
( 5.2)
( 1.8)
( 2.2)
( 2.5)
( 1.5)
t parts per million
"^Standard error not estimable.
Source: Calculated by the Research Triangle Institute from the National Adipose
Tissue Survey's computer accessible data files.
-------
Table B.4 Weighted Percentage Distribution of PCB Residue Levels
w
u>
00
in Adipose Tissue
Domain
Sample
Size
for Fiscal Year 1975
Population
PCB Residue Levels
Estimate
(in thous . )
Not
Detected
Less than
1 PPM t
Between 1
and 3 PPM T
Greater than
3 PPMt
percent (standard error)
Age (years):
0 -14
15-44
45+
Sex:
Male '
Female
Race:
White
Nonwhite
Census Region:
Northeast
North Central
South
West
Number of PCB
Industries per
State:
0
1
2+
Overall
163
304
428
445
450
747
148
206
339
231
119
222
256
417
895
57
83
61
98
104
177
26
49
56
62
34
46
69
86
203
,900
,437
,875
,912
,300
,067
,145
,041
,572
,795
,804
,815
,481
,916
,212
19.7 1
1.5 1
0.7 1
6.9 1
6.0 1
6.5 1
6.0 1
7.0 I
4.6 1
7.1 1
7.5 1
3.5 1
6.8 1
7.7 1
6.4 1
[ 6.8)
[ 0.7)
; o.4)
[ 3.1)
[ 2.0)
[ 2.6)
[ 2.3)
[ 3.9)
[ 2.2)
[ 6.1)
[ 3.4)
: 1.8)
[ 5.4)
( 2.6)
[ 2.3)
71.1
59.5
57.0
63.0
61.1
62.9
56.4
54.9
62.6
63.0
69.4
66.2
66.3
56.3
62.0
( 7.9)
( 4.1)
( 5.0)
( 4.4)
( 4.7)
( 4.1)
( 6.8)
(10.6)
( 6.9)
( 6.1)
( 9.0)
( 7.4)
( 5.6)
( 7.0)
( 4.0)
8.6
30.6
26.9
21.4
25.0
22.9
25.2
21.9
25.1
23.8
21.1
21.7
22.9
24.3
23.2
(2.6)
(3.8)
(3.4)
(3.2)
(3.3)
(2.6)
(4.8)
(5.9)
(4.6)
(4.2)
(6.3)
(5.1)
(4.0)
(4.3)
(2.6)
0.5
8.4
15.4
8.7
7.9
7.7
12.4
16.3
7.8
6.1
2.0
8.6
4.0
11.6
8.3
( 0.4)
( 1.9)
( 3.6)
( 2.3)
( 1.6)
( 1-4)
( 3.3)
( 3.0)
( 2.9)
( 3.1)
( 2.1)
( 3.4)
( 2.4)
( 2.2)
( 1.5)
t parts per million
'-Standard error not estimable.
Source: Calculated by the Research Triangle Institute from the National Adipose
Tissue Survey's computer accessible data files.
-------
Table B.6 Weighted Percentage Distribution ot PCB Residue Levels
U)
VO
in Adipose Tissue for Fiscal Year 1977
Domain
Sample
Size
Population
PCB Residue Levels
Estimate
(in thous . )
Age (years):
0 -14
15-44
45+
Sex:
Male
Female
Race:
White
Nonwhi te
Census Region:
Northeast
North Central
South
West
Number of PCB
Industries per
State:
0
1
2+
Overall
174
357
396
457
470
794
133
206
257
341
123
312
262
353
927
57
83
61
98
104
178
24
49
56
62
34
59
60
83
203
,900
,437
,875
,912
,300
,972
,240
,041
,572
,795
,804
,410
,641
,161
,212
Not Less than Between 1
Detected 1 PPMt and 3 PPMt
percent (standard error)
2.0 {
0.0 (
0.0 {
0.3 1
0.8 I
0.7 1
0.0 1
0.7 1
0.0 I
1.3 1
0.0 1
0.0 1
0.0 {
1.4 1
0.6 1
: 1-6)
: * )
: * )
; 0.3)
: o.s)
: 0.5)
: * )
[ 0.7)
r ,•. \
^ ** /
: 1.3)
[ * )
[ * )
[ * )
[ 1.0)
[ 0.4)
90.1 I
66.8 I
59.8 I
66.7 1
75.6 I
71.5 1
69.7 1
69.5 1
68.6 I
75.2 I
71.2 1
77.6 1
67.9 1
69.3 I
71.3 I
[ 2.9)
: 3.6)
[ A.I)
[ 3.4)
[ 2.6)
[ 2.3)
: 5.1)
[ 3.4)
[ 4.8)
[ 5.2)
[ A. 2)
( A.O)
( A. 8)
[ 2.7)
[ 2.4)
3.7
24.9
23.6
22.4
14.7
18.8
15.9
19.2
21.7
13.8
20.6
14. A
21.6
19.1
18.5
(1.6)
(2.8)
(2.6)
(2.3)
(1.9)
(1.6)-
(A. 2)
(3.3)
(3.1)
(2. A)
(3.6)
(2.6)
(2.8)
(2.1)
(1.5)
Greater than
3 PPMt
A. 2
8.3
16.6
10.5
8.9
9.0
1A.A
10.7
9.7
9.6
8.2
8.0
10.5
10.3
9.7
( 2.3)
( 2.1)
( 3.5)
( 2.5)
( 1.8)
( 1-7)
( A. 6)
( A. 5)
( 2.4)
( 3.2)
( 3.7)
( 2.0)
( 3.5)
( 3.0)
( 1.7)
t parts per million
"'Standard error not estimable.
Source: Calculated by the Research Triangle Institute from the National Adipose
Tissue Survey's computer accessible data files.
-------
Table B.7 Weighted Percentage Distribution of PCB Residue Levels
CO
in Adipose Tissue
Doma i n
Sample
Size
for Fiscal Year 1978
Population
PCB Residue Levels
Estimate
(in thous . )
Age (years):
0 -14
15-44
45+
Sex:
Male
Female
Race:
White
Non-white
Census Region:
Northeast
North Central
South
West
Number of PCB
Industries per
State:
0
1
2+
Overall
205
365
381
475
476
771
180
216
211
372
152
316
270
365
951
57
83
61
98
104
177
25
49
56
62
34
74
48
80
203
,900
,437
,875
,912
,300
,749
,463
,041
,572
,795
,804
,356
,062
,794
,212
Not Less than Between 1
Detected 1 PPM t and 3 PPM t
percent (standard error)
5.7
0.3
0.3
1.0
2.6
1.5
4.4
0.6
2.0
1.6
3.6
2.4
0.8
1.9
1.8
(1-7)
(0.3)
(0.2)
(0.5)
(0.9)
(0.6)
(3.0)
(0.5)
(1.1)
(1.1)
(2.0)
(1.1)
(0.6)
(0.9)
(0.6)
85.1 1
67.9 I
64.7 1
67.0 1
76.4 1
72.8 1
65.2 I
69.4 I
70.9 1
74.0 I
72.8 I
77.5 1
74.5 I
65.0 1
71.8 I
[ 4.3)
( 4.4)
[ 5.1)
[ 3.9)
[ 3.6)
[ 3.8)
( 5.6)
[ 8.9)
[ 5.4)
[ 7.5)
[ 3.9)
[ 5.2)
[ 4.4)
C 6.2)
( 3.5)
6.5
19.6
27.0
20.3
16.0
17.8
20.3
22.6
19.6
13.8
17.2
12.9
17.9
23.1
18.1
(2.4)
(2.9)
(4.4)
(2.9)
(2.9)
(2.7)
(4.3)
(7.0)
(5.0)
(3.7)
(3.5)
(3.5)
(3.9)
(4.3)
(2.5)
Greater than
3 PPMt
2.7
12.2
8.0
11.7
5.0
8.0
10.1
7.4
7.5
10.6
6.4
7.2
6.9
10.0
8.2
(
(
(
(
(
(
(
(
(
(
(
(
(
(
(
1.6)
2.8)
2.5)
2.8)
1.5)
2.0)
3.4)
3.3)
3.4)
4.3)
3.6)
2-7)
2.8)
3.6)
1.9)
t parts per million
^Standard error not estimable.
Source: Calculated by the Research Triangle Institute from the National Adipose
Tissue Survey's computer accessible data files.
-------
Table B.8 Weighted Percentage Distribution of PCB Residue Levels
CO
in Adipose Tissue
Domain
Sample
Size
for Fiscal Year 1979
Population
PCB Residue Levels
Estimate
(in thous. )
Age (years):
0 -14
15-44
45+
Sex:
Male
Female
Race:
White
Non-white
Census Region:
Northeast
North Central
South
West
Number of PCB
Industries per
State:
0
1
2+
Overall
203
346
361
477
433
740
170
189
189
438
94
281
259
370
910
57
84
61
104
99
178
24
49
56
62
34
57
60
85
203
,900
,167
,145
,131
,081
,897
,315
,041
,572
,795
,804
,160
,086
,966
,212
Not Less than Between 1
Detected 1 PPM t and 3 PPM 1
percent (standard error)
4.1
0.0
0.2
1.3
1.2
1.0
3.1
0.9
0.0
2.7
1.1
2.3
0.8
0.9
1.2
(2.2)
( * )
(0.2)
(0.6)
(0.8)
(0.4)
(2.3)
(0.7)
( * )
(1.5)
(1.3)
(1.6)
(0.8)
(0.5)
(0.6)
89.0
80.3
66.6
76.5
80.9
79.5
72.7
55.7
93.0
81.4
82.9
85.6
92.2
64.6
78.7
( 3.5)
( 4.2)
( 3.9)
( 3.5)
( 3.5)
( 3.2)
( 6.1)
(11.7)
( 2.8)
( 4.4)
( 5.7)
( 5.2)
( 1-9)
( 6.8)
( 3.3)
6.9
15.0
26.5
17.7
14.6
15.9
18.4
30.5
6.9
15.4
12.4
11.2
5.5
26.9
16.2
(3.1)
(2.8)
(3.9)
(3.1)
(2.3)
(2.6)
(3.8)
(6.7)
(2.8)
(3.3)
(6.5)
(3.6)
(1.8)
(4.0)
(2.4)
Greater than
3 PPMt
0.0
4.7
6.6
4.5
3.3
3.7
5.8
12.9
0.1
0.6
3.5
0.8
1.5
7.7
3.9
( * )
( 2.9)
( 4.3)
( 1.9)
( 3.2)
( 2.5)
( 2.8)
(10.5)
( 0.1)
( 0.4)
( 2.1)
( 0.6)
( 1.3)
( 6.0)
( 2.5)
t parts per million
""Standard error not estimable.
Source: Calculated by the Research Triangle Institute from the National Adipose
Tissue Survey's computer accessible data files.
-------
Table B.9 Weighted Percentage Distribution of PCB Residue Levels
Domain
Age (years):
0 -14
15-44
45+
Sex:
Male
Female
Race:
White
Nonwhite
Census Region:
Northeast
North Central
South
West
Number of PCB
Industries per
State:
0
1
2+
Overall
Sample
Size
26
47
24
53
44
76
21
19
24
• 34
20
34
30
33
97
in
Adipose
Tissues for Fiscal Year 1981
Population
PCB Residue Levels
Estimate
(in
51,
115,
59,
126,
99,
182,
44,
49,
58,
75,
43,
68,
66,
91,
226
thous . )
390
750
365
700
805
205
299
137
854
349
165
718
161
626
,505
Not Less than Between 1
Detected 1 PPMt and 3 PPMt
percent (standard error)
0.0 (
0.0 (
0.0 (
0.0 (
0.0 (
0.0 (
0.0 (
0.0 (
0.0 (
0.0 (
0.0 (
0.0 (
0.0 (
0.0 (
0.0 (
* )
* )
* )
•/t ^
* )
* )
* )
* )
* )
**» ^
* )
* )
* )
J. \
* )
92.8
69.5
58.4
65.9
79.4
80.7
35.4
73.5
74.0
73.8
63.6
69.2
82.5
66.1
71.8
( 5.3)
(10.6)
(12.8)
(10.8)
( 9.3)
( 6.8)
(13.3)
(15.9)
(13-4)
(16.4)
(22.6)
( 1-7)
(10.6)
(13.0)
( 8.5)
7.2
29.0
41.6
32.7
20.6
19.3
60.7
26.5
23.0
26.2
36.4
28.3
17.5
33.9
27.4
( 5.3)
(10.5)
(12.8)
(10.4)
( 9.3)
( 6.8)
(14.0)
(15.9)
(12.2)
(16.4)
(22.6)
(16.0)
(10.6)
(13.0)
( 8.4)
Greater than
3 PPMt
0.0 (
1.5 (
0.0 (
1.4 (
0.0 (
0.0 (
3.9 (
0.0 (
3.0 (
0.0 (
0.0 (
2.5 (
0.0 (
0.0 (
0.8 (
* )
1.5)
* )
1.4)
* )
"A" ^
4.0)
* )
3.0)
* )
j- \
2.5)
* )
* )
0.8)
t parts per million
^Standard error not estimable.
Source: Calculated by the Research Triangle Institute from the National Adipose
Tissue Survey's computer accessible data files.
-------
B.2 Statistical Models for Trends
To estimate the trends in PD and percent GTS for the U.S. Popu-
lation, the regression model:
Y(i) = B(0) + B(l) (FY-FY) + B(2) (FY-FY)2 (B.2.1)
was used where Y(i), i=l,2, is the dependent variable, FY denotes fiscal
year, FY denotes the mean of the fiscal years. Y is an indicator random
variable taking on the values:
( 0 if PCB = 0
Y(l) = <
( 1 if PCB > 0
when PD is being investigated and
( 0 if PCB < 3 ppm
Y(2) = I
( 1 if PCB > 3 ppm
when percent GTS is being investigated.
For the dependent variable Y(l), both the linear and quadratic
"V *£
terms were significant . Only the quadratic term was significant for
Y(2). Hence, the estimated trend equations are
PD = 96.0 + 1.80 (FY - 76.11) - .189 (FY-76.11)2
and
GTS = 8.15 - .31 (FY-76.11)2.
These equations are plotted in Figures 4 and 7 in Chapter 3.
Besides the overall model, the complex design model
Y(i) = AGEG+RACE+SEX+CR+IND+(FY-FY)+(FY-FY)2+
(FY-FY) * (AGEG+RACE+SEX+CR+IND)
+(FY-FY)2 * (AGEG+RACE+SEX+CR+IND) (B.2.2)
*
Assuming no sampling or measurement bias.
B-43
-------
was tested using SURREGR (Holt 1977) where was Y(i), i=l,2, is defined
as in equation (B.2.1),
AGEG denotes the age effect for the three levels 0-14,
15-44 and 45+,
RACE denotes the race effect for the two levels white
and nonwhite,
SEX denotes the sex effect for the two levels male
and female,
CR denotes the Census Region effect for the four levels
Northeast, North Central, South and West,
IND denotes the PCB industry effect for the three levels
none (0), one (1) and more than one (2+),
FY-FY denotes the difference between the fiscal year and
the mean of the fiscal years,
(FY-FY)"(AGEG+RACE+SEX+CR+IND) denotes the linear interactions of
fiscal year with the main effects, and
(FY-FY) *(AGEG+RACE+SEX+CR+IND) denotes the quadratic interactions
of fiscal year with the main effects.
The categories CR and IND are defined in Tables B.10 and B.ll respec-
tively. Each observation is assigned to the appropriate category by
using the State in which the tissue specimen was collected.
*
The statistical test indicated only one significant second order
2
interaction (significance level (SL) < .05), (FY-FY) *AGEG for GT3 and
none for PD. The model in equation (B.2.2) was simplified by dropping
2
all second order terms with the exception of the (FY-FY) *AGEG for GT3.
The statistical test on the first order interactions indicated no
.A.
significant" interactions among (FY-FY) and RACE, SEX, CR or IND. The
model was again simplified yielding the two models:
*
Assuming no sampling or measurement bias.
B-44
-------
Table B.10 List of States by Census Region and Census Division Within Region
Region
North East
Division
New England
Middle Atlantic
South
South Atlantic
East South
Central
West South
Central
States
Connecticut
Maine
Massachusetts
New Hampshire
Rhode Island
Vermont
New Jersey
New York
Pennsylvania
Region
North Central
Division
East North
Central
West North
Central
Delaware
District of Columbia
Florida
Georgia
Maryland
North Carolina
South Carolina
Virginia
West Virginia
Alabama
Kentucky
Mississippi
Tennessee
Arkansas
Louisiana
Oklahoma
Texas
West
Mountain
Pacific
States
Illinois
Indiana
Michigan
Ohio
Wisconsin
Iowa
Kansas
Minnesota
Missouri
Nebraska
North Dakota
South Dakota
Arizona
Colorado
Idaho
Montana
Nevada
New Mexico
Utah
Wyoming
Alaska
California
Hawaii
Oregon
Washington
-------
Table B.ll List of States By PCB Industry Category
No PCB Industries One PCB Industry
Alaska
Arkansas
Connecticut
3 Delaware
y Florida
Hawaii
Idaho
Iowa
Kansas
Louisiana
Maine
5 Michigan
<; Minnesota
Missouri
Montana
Nebraska
Nevada
New Hampshire
New Mexico
North Dakota
Oklahoma
Rhode Island
South Dakota
Utah
Washington
J West Virginia
^ Wisconsin
Wyoming
Arizona
Colorado
"^Georgia
^Indiana
^ Mississippi
u North Carolina
5 Ohio
Oregon
y Tennessee
Texas
Vermont
Two or more PCB Industries
S- Alabama
California
5" Illinois ,
Massachusetts £~/
New Jersey /^
New York
3 Pennsylvania
Ij South Carolina
Virginia
Source: Compiled by the Research Triangle Institute from lists of
U.S. transformer and capacitor manufacturing industries
using PCBs provided by the U.S. Environmental Protection
Agency.
B-46
-------
Y(l) = AGEG+RACE+SEX+CR+IND+(FY-FY)+(FY-FY)2+(FY-FY)*AGEG
and
Y(2) = AGEG+RACE+SEX+CR+IND+(FY-FY)+(FY-FY)2+(FY-FY)*AGEG+(FY-FY)2*AGEG
(B.2.3)
for PD and GTS respectively. The above models was used to test for the
main effects with the exception of AGEG. The results of this test are
summarized in Chapter 3.
To illustrate the trends in the three age groups, the coefficients
for two simple models were estimated. The equations for PD are
PD = 91.0 + 2.58 (FY-76.11) - .189(FY-76.II)2
for the 0-14 age group,
PD = 98.2 + 1.35(FY-76.11) - .189(FY-76.II)2
for the 15-44 age group, and
PD = 97.8 + 1.60(FY-76.11) - .l89(FY-76.II)2
for the 45+ age group. These equations were used to calculate the
trends illustrated in Figure 3. The equations for percent GT3 are
GT3 = 1.83 - .055 (FY-76.11)2
for the 0-14 age group,
GT3 = 8.55 - .293 (FY-76.11)2
for the 15-44 age group, and
GT3 = 13.6 - .585 (FY-76.11)2
for the 45+ age group. The linear terra in FY was excluded because it
*
did not contribute significantly to explaining the changes in GT3 over
time. These equations were used to calculate the trends in Figure 5.
To illustrate the trends in the race groups, the equations for
percent GT3 are
GT3 = 7.8 - .34(FY-FY)2
*
Assuming no sampling or measurement bias.
B-47
-------
for whites and
GT3 = 11.5 - .34(FY-FY)2
for nonwhites were plotted in Figure 6. Trends for the two sex groups
and the Census Regions would have essentially the same coefficients for
the fiscal year terms but would have different intercepts. Table B.12
summarizes the intercepts for the factors race, sex, Census Region and
industry.
B.3 Assessing Limitations of the Data
In an attempt to ascertain the impact of the change in chemical
analysis on the estimated trends, overall trends were calculated using
data collected during fiscal 1976 to 1979 and 1981. The overall trend
equation for PD is:
PD = 98.98 + .24 (FY-78.2),
and the overall trend equation for percent GTS is:
GT3 = 6.00 - 1.66 (FY-78.2).
The trends for PD and GT3 are illustrated in Figures 9 and 10 respec-
JL
tively. The slopes of the lines are significant , with the slope for PD
JL
having a significance level of .0199 and the slope for GT3 having a
*
significance level of .0000.
To compare the slopes in the above equations with the estimated
trend using all fiscal years, one can calculate the average change per
year from fiscal 1976 to 1981. Using the equations:
PD(FY) = 96.0 + 1.80 (FY-76.11) - .189 (FY-76.11)2
and
GT3(FY) = 8.15 - .31 (FY-76.11)2,
Assuming no measurement of sampling bias.
B-48
-------
Table B.12. Intercepts for Trends of Percent Greater
Than 3 PPM* by Factors —
Factor Subpopulation
Race White
Nonwhite
Sex Male
Female
Census Region Northeast
North Central
South
West
Industry 0
1
2+
Intercept
(Percent)
7.7
11.4
9.3
7.1
11.6
6.8
7.7
6.2
7.4
6.4
9.9
PPM denotes parts per million.
Source: Calculated by the Research Triangle Institute from the
National Adipose Tissue Survey computer accessible data file.
B-49
-------
the average change for PD is (PD(81) - PD(76))/(81-76) = .90 and the
average change for GT3 is (GT3(8l) - GT3(76)) / (81-76) = -1.48. The
average changes per year are consistent in direction. The two estimated
rates of change for GTS are comparable for the time period. The magni-
tudes for PD are not as comparable but this can be explained by the
relatively low value in PD for fiscal 1973 which pulls down the trend
equation for all years. Hence, the change in chemical analysis method
does not appear to consequentially affect the estimated trends.
To develop an understanding of the impact of bias on the significance
level of trends and subpopulation differences, Table B.13 was calculated.
The table illustrates the true significance of a two sided test of
hypothesis when no bias exists and for levels of bias ranging from one
fourth the standard error (SE) to four times the SE. For simplicity,
the significance levels are calculated using the normal density function.
Even though the tests discussed in the report were based on the F distri-
bution, the table is not misleading because the degrees of freedom for
the denominator were large (69 for fiscal 1976-1981 and 103 for all
fiscal years) and the degrees of freedom for the numerator were often
only 1 (at most 3).
Using the table, one can estimate the magnitude of the relative
bias that may mislead the investigator into drawing erroneous conclusions.
For example, if a test was calculated to have a significance level of
.0001, then a bias of more than twice the size of the SE would be required
to raise the true significance level above the traditional .05 level.
Hence, because the significance level of trends were generally .0001 or
lower, it is unlikely that magnitude of the bias is large enough to
B-50
-------
Table B.13. Impact of Bias on the Significnace Level
of Statistical Test of Hypothesis'-
BIAS/SE
0.00
0.25
0.50
1.00
1.50
2.00
3.00
4.00
Significance Level
.0500
.0572
.0791
.1701
.3230
.5160
.8508
.9793
.0250
.0297
.0440
.1081
.2297
.4052
.7764
.9608
.0100
.0124
.0190
.0578
.1412
.2826
.6646
.9229
.0010
.0012
.0028
.0113
.0375
.1003
.3897
.7642
.0001
.0001
.0003
.0019
.0084
.0294
.1867
.5438
.00001
.00002
.00004
.00032
.00177
.00782
.07825
.33834
Calculated by the Research Triangle Institute using the equation:
z(a) + BIAS/SE
SL
exp(-x /2)/V2n dx
-z(a) + BIAS/SE
where z(.0500) = 1.960, z(.0250) = 2.24, z(.0100) = 2.576, z(.0010) =
3.291, z(.OOOl) = 3.891, z(.00001) = 4.417 and SE denotes the standard
error.
B-51
-------
affect the conclusions about trends. Similar reasoning can be used to
allow the reader to evaluate the conclusions about race, sex, or geogra-
phic differences.
B-52
-------
B.4 References
Holt MM 1977 SURREGR: Standard errors of regression coefficients from
sample survey data. Research Triangle Institute, Research Triangle
Park, NC 27709.
Shah DV 1979 SESUDAAN: Standard errors program for computing of stan-
dardized rates from sample survey data. Research Triangle Institute,
Research Triangle Park, NC 27709, RTI/1789/00-01F.
B-53
------- |