United States
Environmental Protection
Agency
Office of Prevention, Pesticides,
and Toxic Substances
7101
EPA 747-R-93-007
February, 1995
vvEPA
COMPREHENSIVE
ABATEMENT PERFORMANCE
PILOT STUDY
VOLUME I: RESULTS OF
LEAD DATA ANALYSIS
-------
February 1995
EPA 747-R-93-007
FINAL REPORT
' for the
COMPREHENSIVE ABATEMENT PERFORMANCE PILOT STUDY
VOLUME I: RESULTS OF LEAD DATA ANALYSES
A'A
v ' Technical Programs Branch
Chemical Management Division
( Office of Pollution Prevention and Toxics
Office of Prevention, Pesticides, and Toxic Substances
U.S. Environmental Protection Agency
Washington, D.C. 20460
-------
DISCLAIMER
Mention of trade names, products, or services does not
convey, and should not be interpreted as conveying, official EPA
approval, endorsement, or recommendations.
-------
ACKNOWLEDGEMENTS
This study was funded and managed by the U.S. Environmental
Protection Agency. The study was conducted collaboratively by
two organizations under contract to the Environmental Protection
Agency, Battelle Memorial Institute and Midwest Research
Institute. Each organization's responsibilities are listed
below.
Battelle Memorial Institute (Battelle)
Battelle was responsible for the design of the study, for
producing the design documentation and the Quality Assurance
Project Plan, for developing training for the field teams, for
recruiting cooperators for the study, for providing the team
leader for the field team, for data management of combined study
data, for auditing the study data, for conducting the statistical
analysis of the data, and for writing the final report.
Midwest Research Institute (MRI)
Midwest Research Institute was responsible for participating
in the planning for the study, for writing certain chapters and
appendices in the Quality Assurance Project Plan, for developing
training for and training the field teams, for providing the
technicians who collected the field samples, for auditing the
field teams, for conducting the laboratory analysis of the field
samples, for managing the data associated with the field samples,
for auditing the laboratory results, and for contributing
sections of the final report.
U.S. Environmental Protection Agency (EPA)
The Environmental Protection Agency was responsible for
managing the study, for reviewing the design and the Quality
Assurance Project Plan, for assessing the performance of the
recruiters and the field teams, for reviewing the final report,
and for arranging the peer review of the design and the final
report. The EPA Work Assignment Managers were Benjamin Lim and
John Schwemberger. The EPA Project Officers were Gary
Grindstaff, Phil Robinson, Joseph Breen, and Janet Remmers.
111
-------
TABLE OF CONTENTS
Page
EXECUTIVE SUMMARY xi
1.0 INTRODUCTION AND SUMMARY 1
1.1 STUDY DESIGN 1
1.2 SUMMARY OF RESULTS 9
2.0 RECRUITMENT, RISK COMMUNICATION, AND FIELD EXPERIENCES . 14
2 .1 RECRUITMENT EXPERIENCES 14
2.2 RISK COMMUNICATION 15
2.3 SAMPLE COLLECTION, PREPARATION AND ANALYSIS
PROCEDURES 15
2.3.1 Sample Collection Procedures 16
2.3.2 Sample Preparation and Chemical Analysis . . 16
2.4 FIELD EXPERIENCES 17
3 . 0 DATA MANAGEMENT AND PRELIMINARY ANALYSES 18
3.1 DATA MANAGEMENT 18
3.1.1 Sample Collection 19
3.1.2 Analytical Data Transfer 22
3.1.3 Sampling and Analysis Deviations 22
3.1.4 Experiences From The Pilot Study 23
3.2 OUTLIER ANALYSIS 24
3.2.1 Outlier Analysis Approach 24
3.2.2 Data Groups 25
3.2.3 The Outlier Test 25
3.2.4 Resolution of Outlier Questions 27
3.3 DUST COLLECTED AND AREA SAMPLED 27
3.4 COMPARISON OF ICP AND GFAA RESULTS 32
4.0 DATA INTERPRETATION 35
4.1 DESCRIPTIVE STATISTICS 38
4.2 STATISTICAL MODELS 55
4.3 MODELING RESULTS BY MEASUREMENT TYPE 60
4.3.1 Estimates of Variance Components With
No Fixed Effects 61
4.3.2 Estimates of Renovation Effects, Abatement
Effects, and Variance Components 75
iv
-------
TABLE OF CONTENTS
(Continued)
4.4 MODELING RESULTS FOR DUST SAMPLES BY SAMPLE TYPE
92
4.4.1 Air Duct Samples 92
4.4.2 Bed/Rug/Upholstery Samples 93
4.4.3 Interior Entryway Samples 93
4.4.4 Floor Samples 94
4.4.5 Window Stool Samples 96
4.4.6 Window Channel Samples 98
4.5 MODELING RESULTS FOR SOIL SAMPLES BY SAMPLE TYPE . . 100
4.5.1 Boundary Soil Samples 100
4.5.2 Exterior Entryway Soil Samples 100
4.5.3 Foundation Soil Samples 101
4.5.4 Comparison of the Soil Sample Types 102
4.6 RELATIONSHIPS BETWEEN SAMPLE TYPES 102
4.7 COMPARISON OF VACUUM AND WIPE SAMPLING
PROCEDURES 115
4 .8 COMPARISON OF CAP PILOT DATA AND
HUD DEMONSTRATION DATA 121
5.0 STATISTICAL ANALYSIS OF QUALITY CONTROL DATA 133
5.1 BLANK SAMPLES 134
5.2 RECOVERY SAMPLES 139
5.3 DUPLICATE SAMPLES 144
5.4 INTERLABORATORY COMPARISON SAMPLES 147
6.0 REFERENCES 150
APPENDIX A. CAP PILOT STUDY DATA A-l
LIST OF TABLES
Table 1-1. Summary of Environmental Sampling
Planned for the CAP Study 5
Table 3-1. Unit Summary of Sample Collection 20
Table 3-2. Summary of Planned Samples, Collected
Samples and analytical results used in
analysis 21
Table 3-3. CAPS Pilot Study Outliers 26
Table 3-4. Descriptive Statistics for Amount of
Dust Collected (mg) and Area Sampled
(ft2) by Sample Type 28
Table 3-5. Symbols Used to Denote Sample Types in
Tables and Figures 28
-------
TABLE OF CONTENTS
(Continued)
Table 3-6. ICP and GFAA Measurements (Lead Loading
and Lead Concentration) for Samples
Analyzed by GFAA 33
Table 4-1. Descriptive Statistics for Lead Loading
(/zg/ft2) , Lead Concentration (/ig/g) , and
Dust Loading (mg/ft2) by Sample Type for
All Units Considered 40
Table 4-2. Geometric Mean for Lead Loading
(/zg/ft2) , Lead Concentration (/xg/g) , and
Dust Loading (mg/ft2) by Sample Type and
Unit 46
Table 4-3. Loading versus Concentration
Correlations for Dust Samples 54
Table 4-4a. Geometric Mean and Variance Component
Estimates from Model with No Fixed
Effects: Lead Loading (/xg/ft2) 62
Table 4-4b. Geometric Mean and Variance Component
Estimates from Model with No Fixed
Effects: Lead Concentration (/-tg/g) 63
Table 4-4c. Geometric Mean and Variance Component
Estimates from Model with No Fixed
Effects: Dust Loading (mg/ft2) 64
Table 4-5a. Estimated Renovation Effects, Estimated
Abatement Effects, and Variance
Component Estimates from Mixed Model
ANOVA: Lead Loading 76
Table 4-5b. Estimated Renovation Effects, Estimated
Abatement Effects, and Variance
Component Estimates from Mixed Model
ANOVA: Lead Concentration (/^g/g) 77
Table 4-5c- Estimated Renovation Effects, Estimated
Abatement Effects, and Variance
Component Estimates from Mixed Model
ANOVA: Dust Loading (mg/ft2) 78
Table 4-6a. Unit-to-Unit Correlations Among Sample
Types: Lead Loading 104
Table 4-6b. Unit-to-Unit Correlations Among Sample
Types After Correction for Renovation
and Abatement Effects: Lead Loading 108
Table 4-7a. Unit-to-Unit Correlations Among Sample
Types: Lead Concentration Ill
Table 4-7b. Unit-to-Unit Correlations Among Sample
Types After Correction for Renovation
and Abatement Effects: Lead
Concentration 113
Table 4-8a. Vacuum versus Wipe Comparison Data:
Floor Lead Loadings (/xg/ft2) 116
vi
-------
TABLE OF CONTENTS
(Continued)
Page
Table 4-8b. Vacuum versus Wipe Comparison Data:
Window Stool Lead Loadings (^g/ft2) 116
Table 4-8c. Vacuum versus Wipe Comparison Data:
Window Channel Lead Loadings (/ig/ft2) 117
Table 4-9. Geometric Means for CAP Pilot and HUD
Demonstration Data by Room: Interior
XRF/AAS Results (mg/cm2) and Dust Lead
Loadings (pig/ft2) 122
Table 4-10. Geometric Means for CAP Pilot and HUD
Demonstration Data by Side of Unit:
Exterior XRF/AAS Results (mg/cm2) and
Soil Lead Concentrations (fJig/g) 130
Table 5-1. Descriptive Statistics Tolerance Bound
for fjig Lead/Sample in Blank Samples 136
Table 5-2. Descriptive Statistics and Tolerance
Bounds for Percent Recovery in Recovery
Samples 142
Table 5-3. Descriptive Statistics and Tolerance
Bounds for the Ratio of Duplicate
Samples 145
Figure 1-1.
Figure 3-1.
Figure 3-2.
Figure 3-3.
Figure 4-la.
Figure 4-lb.
Figure 4-lc.
LIST OF FIGURES
Example of Sampling Locations Within
a Unit, with Sample Type Identified
for Each Location as Reflected in
Table 1-1
Amount of dust collected (mg) by
sample type
Area sampled (ft2) by sample type.
Plot of ICP lead amounts vs. GFAA
lead amounts for cassette samples
analyzed by GFAA
Lead loading measurements
(/xg/ft2) and geometric mean lead
loading for all units by sample
type
Lead concentration measurements
(/xg/g) and geometric mean lead
concentration for all units by
sample type
Dust loading measurements
(mg/ft2) and geometric mean dust
loading for all units by sample
type
30
31
34
42
43
44
VII
-------
TABLE OF CONTENTS
(Continued)
Page
Figure 4-2.
Figure 4-3a.
Figure 4-3b.
Figure 4-3c.
Figure 4-4a,
Figure 4-4b.
Figure 4-4c.
Figure 4-5.
Figure 4-6a.
Figure 4-6b.
Figure 4-6c.
Figure 4-7a.
Figure 4-7b,
Figure 4-7c.
Geometric means over all units for
lead loading (L) in /xg/ft2, lead
concentration (C) in /xg/g, and dust
loading (D) in mg/ft2 by sample
type 45
Geometric means by unit for floor
and upholstery samples: lead
loading (/xg/ft2) 47
Geometric means by unit for floor
and upholstery samples: lead
concentration (/ig/g) 48
Geometric means by unit for floor
and upholstery samples: dust
loading (mg/ft2) 49
Geometric means by unit for
window and air duct samples:
lead loading (/zg/ft2) 50
Geometric means by unit for
window and air duct samples:
lead concentration (/xg/g) 51
Geometric means by unit for
window and air duct samples:
dust loading (mg/ft2) 52
Geometric means by unit for soil
samples: lead concentration (jug/g) 53
Variance component estimates from
model with no fixed effects:
lead loading (/zg/ft2) 67
Variance component estimates from
model with no fixed effects:
lead concentration (/xg/g) 68
Variance component estimates from
model with no fixed effects:
dust loading (mg/ft2) 69
Estimated geometric mean in
unrenovated control units for
lead loading in /xg/ft2, lead
concentration in /xg/g, and dust
loading in mg/ft2 by sample type _. . 80
Estimated multiplicative effects
of renovation and abatement from
mixed model ANOVA: lead loading
(Mg/ft2) si
Variance component estimates from
mixed model ANOVA: lead loading
) 82
Vlll
-------
TABLE OF CONTENTS
(Continued)
Figure 4-7d. Estimated multiplicative effects
of renovation and abatement from
mixed model ANOVA: lead
concentration (/xg/g) 83
Figure 4-7e. Variance component estimates from
mixed model ANOVA: lead
concentration (/>ig/g) 84
Figure 4-7f. Estimated multiplicative effects
of renovation and abatement from
mixed model ANOVA: dust loading
(mg/ft2) 85
Figure 4-7g. Variance component estimates from
mixed model ANOVA: dust loading
(mg/ft2) 86
Figure 4-8a. Scatterplot matrix of geometric
unit means for different sample
types: lead loading (/xg/ft2) 105
Figure 4-8b. Scatterplot matrix of geometric
unit means for different sample
types after correction for
renovation and abatement effects:
lead loading (/xg/ft2) 109
Figure 4-9a. Scatterplot matrix of geometric unit
means for different sample types:
lead concentration (/xg/g) 112
Figure 4-9b. Scatterplot matrix of geometric
unit means for different sample
types after correction for
renovation and abatement effects:
lead concentration (/xg/g) 114
Figure 4-10a. Vacuum vs. wipe comparison:
geometric means by sample
location for floor lead loadings
(/ig/ft2) 118
Figure 4-10b. Vacuum vs. wipe comparison:
geometric means by sample
location for window stool
loadings (/xg/ft2) 120
Figure 4-11. CAP pilot vacuum versus HUD
Demonstration wipe data:
geometric mean floor lead loading
(/xg/ft2) by room 125
Figure 4-12a. CAP pilot and HUD Demonstration
floor lead loadings (tig/ft2)
versus HUD Demonstration XRF/AAS
results: geometric means by
room 126
ix
-------
TABLE OF CONTENTS
(Continued)
Figure 4-12b.
Figure 4-12c.
Figure 4-13.
Figure 4-14.
Figure 5-1.
Figure 5-2.
Figure 5-3.
Figure 5-4.
CAP pilot and HUD Demonstration
window stool lead loadings
(^g/ft2) versus HUD Demonstration
XRF/AAS results: geometric means
by room
CAP Pilot and HUD Demonstration
window channel lead loadings
(/zg/ft2) versus HUD demonstration
XRF/AAS results: geometric means
by room
CAP Pilot soil versus HUD
Demonstration soil data:
geometric mean foundation soil
lead concentration (/xg/g) by side
of unit
CAP Pilot and HUD Demonstration
soil lead concentrations (/J-g/g)
versus HUD Demonstration XRF/AAS
results: geometric means by
side of house
Individual measurements and
tolerance bounds for ng lead/sample
in blank samples
Individual measurements and
tolerance bounds for percent
recovery in recovery samples
Individual measurements and
tolerance bounds for the ratio of
duplicate samples
Secondary laboratory lead
concentrations (/ug/g) versus primary
laboratory lead concentrations for
interlaboratory comparison soil and
cassette samples
127
128
129
131
138
143
146
148
-------
EXECUTIVE SUMMARY
This report presents the results from the pilot study
that preceded the Comprehensive Abatement Performance Study. The
goal of the Comprehensive Performance Study was to assess the
long-term impact of lead-based paint abatement. The pilot study
was conducted to test the sampling and analysis protocols that
were intended for the full study. These protocols called for
determining the levels of lead in dust and soil samples collected
at residential units. The pilot study was conducted at six
houses, and all steps that were planned for the full study were
included in the pilot.
The major finding of the pilot was the difference
between wipe and vacuum methods for collecting dust. The choice
of method had a noticeable impact on the level of lead associated
with the collected sample.
All other sampling and analysis aspects of the pilot
study were completed successfully. In particular, an inter-
laboratory comparison of dust and soil samples indicated no
systematic difference in lead levels between the two
laboratories. In addition, intra-laboratory comparisons of
sample results by inductively coupled plasma-atomic absorption
spectrometry (ICP) and the more sensitive graphite furnace atomic
absorption spectrometry (GFAA) indicated good agreement within
the common domain of instrument detection limits. The pilot
study suggested that GFAA analysis would not be necessary for the
full study, if sufficient amount of sample was collected for ICP
analysis.
Other important findings from the pilot study were
results related to variance components. Estimates of random
house-to-house, room-to-room, and side-by-side sample variability
were obtained for most of the sample types in the study. These
estimates were used for determining the number of houses and
number of samples per house for the full study.
XI
-------
1.0 INTRODUCTION AND SUMMARY
This report presents final results from the Comprehensive
Abatement Performance Pilot Study, conducted in 1991 by Battelle
Memorial Institute and Midwest Research Institute (MRI) for the
U.S. Environmental Protection Agency's Office of Pollution
Prevention and Toxics (OPPT). The objectives, approach, and
design of this study, although briefly summarized here, are
completely described in the "Quality Assurance Project Plan for
the Abatement Performance Pilot Study" (Battelle and MRI, 1991).
1.1 STUDY DESIGN
Under an interagency Memorandum of Understanding, the
Environmental Protection Agency (EPA) is provided technical
support to the Department of Housing and Urban Development (HUD)
with respect to the abatement of lead-based paint hazards in
public and private housing. As part of its lead-based paint
research activities, HUD carried out a Demonstration Program in
ten cities to assess the costs and short-term efficacy of
alternative methods of lead-based paint abatement. A variety of
abatement methods were tested in approximately 120 multi-family
public housing units in three cities -- Omaha, Cambridge, and
Albany -- and in 172 single-family housing units in the FHA
inventory in seven metropolitan areas -- Baltimore, Birmingham,
Denver, Indianapolis, Seattle, Tacoma, and Washington. The FHA
portion of the Demonstration has now been completed, and OPPT is
planning to conduct a follow-up study (referred to as the
Comprehensive Abatement Performance (CAP) Study) of these housing
units with the following objectives:
1. Compare abatement methods or combination of methods
relative to performance over time. Assess whether
there are differences in performance.
2. Characterize levels of lead in household dust and
exterior soil over time for HUD Demonstration and
control homes.
3. Investigate the relationship between lead in household
dust and lead from other sources, in particular,
-------
exterior soil, rugs, upholstered furniture, and air
ducts.
The CAP Study is one of two major field studies currently
being conducted by OPPT. While the CAP Study will examine
relatively high-cost lead-based paint abatement alternatives
tested by HUD in their Demonstration Program, OPPT will also
examine lower-cost repair and maintenance methods for dealing
with lead-based paint and associated lead contaminated dust
(Battelle and Kennedy Krieger Institute, 1992). Like the CAP
Study, the first step in the Repair and Maintenance Study was to
conduct a pilot program to test the sampling and analysis
protocols planned (Battelle and Kennedy Krieger Institute, 1992).
This document describes the results from the CAP Pilot Study.
The Pilot Study was intended to investigate the field,
laboratory, and statistical analysis procedures planned for the
full CAP Study. In particular, the objectives of the Pilot Study
were as follows:
Test the sampling and analysis protocols;
Evaluate the questionnaires and other field data forms;
Provide variance estimates to help determine the final
design of the full CAP Study;
Assess the performance (i.e., sensitivity, accuracy,
and precision) of the sampling and analysis methods;
Compare analytical results for the MRI (primary) and
Kennedy Krieger Institute (secondary) laboratories; and
Compare the vacuum/total digestion protocol planned for
the full CAP Study with the wipe/ashing protocol
previously used in the HUD Demonstration Study.
The first five objectives are all necessary precursors that
will help to refine the study design and methods for the full CAP
Study. The final objective is intended to further enhance our
ability to assess the HUD abatement methods by providing a bridge
-------
between earlier dust measurements from the HUD Demonstration
obtained with a wipe sampling method, and our current dust
measurements obtained with vacuum sampling.
Our data analysis approach for the Pilot Study focused on
three statistical study objectives: variance component
estimation, comparison of vacuum and wipe protocols, and
assessment of the performance of the sampling and analysis
methods. Because this study was a pilot, we did not state our
Data Quality Objective (DQO) in terms of a specific statistical
hypothesis to be tested for the full CAP Study. Instead, our
objective for the Pilot was to collect sufficient information to
allow us to estimate variance components that are key to the
subsequent design of the full study. Specifically, our DQO was
to collect a minimally sufficient amount of data to allow
estimation of the following important sources of variation that
may be found in measurements of lead in interior dust:
Variations between houses abated with different
methods;
Variations between houses abated with the same method;
Variations between rooms abated with the same method
within a house;
Variations between sampling locations and abated
components within a room; and
Variations from non-paint sources.
In order to assess these sources of variation, our DQO was to
successfully collect and measure lead levels in a nearly complete
set (i.e., 95% data completeness) of 258 dust and soil samples.
The field sampling design for the Pilot Study included
samples to address the variance component estimation and
comparison of vacuum and wipe sampling. All of these samples are
shown in Table 1-1 and Figure 1-1. A summary of the most
-------
important design considerations for the Pilot Study is contained
in the following points:
To assess variability associated with different housing
units and different abatement methods, six housing
units in Denver were sampled. Two units were selected
from those predominantly abated by
encapsulation/enclosure methods, two units were
selected from those predominantly abated by removal
methods, and two units were selected from those control
houses already tested by HUD and found relatively free
of lead-based paint.
To assess variability from different sources within a
house, a total of 18 regular vacuum dust samples was
collected in each house (Table 1-1). Sampling was
performed in two different rooms of each house. When
selecting two abated rooms, rooms were chosen that were
both predominantly abated by the same method used for
the house in general.
Soil samples were collected in the Pilot Study to help
assess potential non-paint sources of lead
contamination in interior dust. For two sides of each
house, soil samples were collected both at the
foundation of the house and at the property boundary.
In addition, soil samples were collected immediately
outside the front and rear entryways.
For each of the six housing units included in the Pilot
Study, one room was selected for comparative vacuum and
wipe sampling. This room was a third room added to the
two sampled rooms discussed above. Within each room
selected for comparative sampling, a randomized side-
by-side arrangement of paired vacuum samples and paired
wipe samples was collected from the floor. In
addition, paired samples were collected on both the
stool and channel of the two windows in the room. The
window stool was defined as the horizontal board inside
the window -- often called the window sill. The window
channel was defined as the surface below the window
sash and inside the screen and/or storm window. One
window was typically designated for either paired
vacuum or paired wipe samples; while to other window
was designated for paired vacuum-wipe sampling (see KIT
in Figure 1-1).
-------
Table 1-1.
Summary of Environmental Sampling Planned for the
CAP Study
Sample Type
Recrular Samples
1 . Vacuum Dust
a. Floor (2 per room)
b. Window Stool (2 per room)
c. Window Channel (2 per room)
d. Rug/Upholstery (1 per room)
e. Air Ducts (1 per room)
f . Entryway (Front & Back)
2. Soil Cores
a. Near Foundation
b . Property Boundary
c. Entryway (Front & Back)
Vacuum vs . Wipe Samples
3 . Vacuum Dust
a. Floor
b. Window Stool
c. Window Channel
4 . Wipe Dust
a. Floor
b. Window Stool
c . Window Channel
Quality Control Samples
5 . Interlaboratory Comparison
a. Vacuum Floor Dust
b. Soil Cores
6 . Field Blanks
a. Vacuum Dust
b. Wipe Dust
c. Soil Core Liners
7. Side-by-Side Samples
a . Vacuum Floor Dust
b. Soil Cores
Total
Number of Samples
Planned
Samples
Per Unit
4
4
4
2
2
2
2
2
2
2
2
2
2
2
2
1
1
1
1
1
1
1
1
43
Total
(6 Units)
24
24
24
12
12
12
12
12
12
12
12
12
12
12
12
6
6
6
6
6
6
6
6
258
-------
p
4m\
11
s
m
o
8
a
\
a
m
Q
i
11
S 7
oc
o
o
Figure 1-1.
Example of Sampling Locations Within a Unit, with
Sample Type Identified for Each Location as
Reflected in Table 1-1.
-------
<
CD
CM
QC
OQ
CO
DC
m
cc
m
QC
O
O
TJ
CM
Figure 1-1. (Continued)
-------
c
m
GO
x*~X
O
t
6C - Field Blank
posite Sample
g
oc
uj
in
cc
1
I
0.
Figure 1-1. (Continued)
-------
Seven quality control samples (i.e., field side-by-
sides, field blanks, and interlaboratory comparison
samples) were collected to assess variability
introduced by the sampling method, sample handling, and
laboratory effects.
It should be emphasized here that control houses are houses
which were classified as not needing abatement because they were
found by HUD as being relatively free of lead-based paint. Thus,
this study does not assess lead levels before and after
abatement. Instead, this report provides a comparison of lead
levels in abated houses to those in houses not needing abatement.
In addition, during sampling it was discovered that one
control house (Unit 19) was undergoing partial renovation, and
one encapsulation/enclosure house (Unit 51) was undergoing full
renovation. In order to evaluate the impact of renovation and
also control for its effect when estimating the abatement effect,
a renovation measure was included in the statistical models as a
covariate (see Section 4.2).
1.2 SUMMARY OF RESULTS
This report, which is organized into two volumes, provides a
complete description of the CAP Pilot Study results. Volume I
summarizes the findings from a thorough statistical analysis of
the lead measurements collected for interior dust and exterior
soil samples. Volume II describes the results of a multivariate
statistical analysis of lead, cadmium, chromium, titanium, and
zinc measurements made on those same samples.
Section 2.0 of this Volume I presents findings concerning
recruitment, risk communication, and experiences in the field.
Next, in Section 3.0, results of the data management activities
are provided. This section completely describes all of the data
collected in the Pilot Study, and summarizes our suggestions for
enhancements to the data management system for the full CAP
Study. Section 4.0 presents the findings from the statistical
analysis of the Pilot Study data. In keeping with the Pilot
-------
Study design, the analysis considered a wide variety of topics,
including estimation of renovation and abatement effects,
estimation of variance components, comparison between lead levels
in different sampling media (e.g., soil and dust) and at
different sampling locations (e.g., floors and window stools),
comparison of lead levels measured by the vacuum and wipe
sampling protocols, and comparison of CAP Pilot sampling results
with earlier results from the HUD Demonstration study. Finally,
Section 5.0 presents results of the statistical evaluation of
various field and laboratory quality control data collected in
the study.
The results of the CAP Pilot Study can be organized into
three categories: findings pertaining to the three CAP Study
objectives listed in Section 1.1, those pertaining to other
important topics, including comparisons between vacuum and wipe
dust sampling protocols, and conclusions concerning operational
aspects of the study, such as recruitment, risk communication,
field data collection, and data management. The major findings
of the Pilot Study which pertain to the three primary objectives
of the CAP Study are as follows:
1. Levels of Lead in Dust and Soil -- Environmental
samples for six houses in Denver were analyzed for lead
levels in two media (dust and soil) and at several
different sampling locations (e.g., floors, windows,
foundation soil, boundary soil) . For dust samples,
geometric average lead concentrations ranged from a
high of 1440 jug/g for window channel samples, to a low
of 174 /xg/g for bed, rug, and upholstery samples. For
soil samples, geometric average lead concentrations
ranged from 217 /ig/g for foundation samples, to 121
f°r boundary samples.
2. Compare Abatement Methods -- Two of six houses sampled
were unabated, uncontaminated control houses; two
houses were abated by encapsulation/enclosure methods;
and two houses were abated by removal methods. In
addition, two of the six houses were undergoing full or
partial renovation at the time of sampling.
10
-------
a. Units under renovation had average dust lead
loadings (/ig/ft2) on floors and window stools many
times higher than those in unrenovated control
units.
b. Floor lead loadings (^ig/ft2) in abated rooms were
comparable to those in control units; however,
floor lead loadings in unabated rooms of abated
units were many times higher than those in abated
rooms of the same units.
3. Relationships Between Lead in Different Media and
Locations -- Lead levels were compared for six
different interior locations (i.e., floors, entryways,
window channels, window stools, air ducts, and
bed/rug/upholstery) and three different exterior
locations (i.e., entryways, foundation, and property
boundary).
a. Soil lead concentrations (/xg/g) were generally
well correlated among the three exterior sampling
locations. The soil lead concentrations were also
often correlated with interior dust lead
concentrations.
b. Average lead concentrations in boundary soil
samples (121 /xg/g) were significantly lower than
those in entryway soil samples (196 jug/g) and
foundation soil samples (217 /xg/g) suggesting that
the housing unit may contain additional sources of
lead (e.g., lead-based paint) which contaminate
nearby soil beyond the contamination introduced by
other area sources, such as fallout from
automotive or other combustion processes.
Other major findings from the Pilot Study which are not
necessarily directly related to the three primary objectives of
the CAP Study are:
4. Vacuum Versus Wipe Sampling -- A total of 64 vacuum and
wipe samples were collected in the Pilot Study for
comparative analysis. The wipe sampling procedure
produced lead loadings (//g/ft ) for floor samples that
were approximately 5 times higher, with a 95%
confidence interval of 2 to 15, and lead loadings for
window stool samples that were approximately 5 times
higher, with a 95% confidence interval of 3 to 8, than
those by the vacuum sampling procedure.
11
-------
5. CAP Pilot Data Versus HUD Demonstration Data -- CAP
Pilot soil concentration data were highly correlated
with HUD Demonstration soil concentration data,
although the HUD Demonstration data were moderately
higher (approximately 25%) than the CAP Pilot data.
Both the CAP Pilot and HUD Demonstration dust and soil
lead data appear to be only weakly correlated with the
HUD Demonstration XRF/AAS measurements of lead in
paint.
6. Interlaboratory Comparison -- A total of 68 vacuum dust
and soil samples were collected and randomly assigned
to the primary and secondary laboratories for
comparative analysis. No systematic differences were
found in the lead concentrations reported by the two
laboratories for matching pairs of samples.
Major findings from the Pilot Study concerning operational
aspects are as follows:
7. Recruitment -- Most occupants who were contacted about
the Pilot Study were enthusiastic about participating.
Telephone calls in combination with next-day delivery
mailings provided an effective means of contacting
these individuals. Also, due to the observed magnitude
of the renovation effects on lead levels, future
studies should make an effort to control this factor in
the selection of homes. At the very minimum,
renovation should be controlled for in any data
analysis.
8. Risk Communication -- Recruitment mailings and written
reports of the Pilot Study results provided effective
means of communicating to residents the potential
health risks of lead exposure.
9 Field Data Collection -- The sampling protocols for
dust and soil performed well in the field, although the
Pilot Study results indicated the need for a more
efficient vacuum sampling device. Sampling required 5
to 7 hours of work at each pilot house; but with the
reduced number of samples and modified dust sampling
device planned for the full study, this time is
expected to be reduced to about 1& to 3 hours.
10. Data Management -- Use of separate field and laboratory
sample IDs proved very helpful for effectively tracking
samples. Detailed instructions for completing field
data collection forms, formally capturing laboratory
12
-------
analysis comments, and frequent meetings between field,
laboratory, data management, and statistical analysis
personnel are recommended for the full CAP Study to
more effectively communicate important information to
the entire project team.
13
-------
2.0 RECRUITMENT, RISK COMMUNICATION, AND FIELD EXPERIENCES
This section presents a summary of recruitment, risk
communication, and field sampling experiences from the Pilot
Study.
2.1 RECRUITMENT EXPERIENCES
In order to meet the goals of recruiting a minimum of six
occupied units for participation in the Pilot Study, 20 houses
were targeted for recruitment. Owners or occupants were
contacted by telephone or next-day delivery letter to explain the
purpose of the study, why their home was selected for this study,
and to solicit their cooperation in allowing a team of
investigators visit their home to collect dust and soil samples.
A script was used for recruitment. Recruitment letters and a
brochure were also mailed to residents.
A high level of interest and a willingness to participate in
the study was displayed by occupants reached by telephone.
However, reaching people by telephone required late-night efforts
because of the time difference between the East Coast and Denver.
Next-day delivery letters were found to be appropriate for
recruiting residents of investor-owned units. Because the names
of these occupants were not known, use of next-day delivery
conveyed an importance that would not have been conveyed had
regular mail been used. Next-day delivery service also proved to
be an inexpensive method for determining if the unit was
unoccupied. Thirteen of the original 20 houses were unoccupied
or unreachable. In addition, one resident (removal house)
refused delivery of the recruitment package claiming they did not
know Battelle. However, because they did not accept delivery,
they did not know what they were refusing, and therefore this
refusal probably had no biasing effect on the study results.
Pilot testing of the telephone interview identified
questions that needed modification or elimination. Pilot testing
of the recruitment script indicated the appropriateness of the
script.
14
-------
2.2 RISK COMMUNICATION
Risk communication efforts employed in the Pilot Study
consisted of two components: (1) risk communication associated
with recruitment into the study, and (2) risk communication
resulting from conduct of the study. During the recruitment
phase of the project, all subjects were solicited by telephone
for participation in the study. This telephone solicitation was
the first information received by owner-occupants, while
residents of investor-owned property were solicited by telephone
after they responded to the next-day delivery letter addressed
to "resident". The telephone solicitation was done according to
a pre-designed script, one of whose purposes was to describe
potential hazards associated with lead exposure.
All participating residents also received mailings which
described the potential hazards associated with lead exposure.
The mailings comprised the second risk communication effort for
owner-occupants and the first risk communication effort for
residents of investor-owner units. Two separate letters were
sent in these mailings. Each letter described the health hazards
associated with lead exposure. Along with these letters a
brochure was enclosed describing the study and the hazards
associated with lead exposure.
Letters and reports of the visual inspection and laboratory
analysis were sent to the Pilot Study participants informing them
of the results of the data collection effort. By highlighting
results that indicate potential "hot spots" of lead, areas in
need of better housekeeping were brought to their attention.
Study participants were referred to their local health department
for more information.
2.3 SAMPLE COLLECTION, PREPARATION AND ANALYSIS PROCEDURES
This section summarizes the collection and analysis methods
used for the vacuum dust, wipe dust, and soil samples.
15
-------
2.3.1 Sample Collection Procedures
Vacuum samples of surface dust were collected from floors,
window stools and channels, upholstered furniture, rugs and air
ducts. The vacuum sampling device consisted of a Teflon pick-up
nozzle mounted on a pre-weighed 37-mm, mixed-cellulose ester
filter cassette (0.8-^tm pore size) . This device was coupled to a
rotary-vane vacuum with Tygon tubing. The area vacuumed was
nominally 4-ft2 for floor samples, 1-ft2 for upholstery and rug
samples, and the entire accessible surface for window stools,
channels, and air ducts. Vacuuming time for each square foot was
nominally two minutes.
Wipe samples of surface dust were collected from uncarpeted
floors, window stools and window channels. The surfaces were
wiped with standard, name-brand wipes, using a sampling method
used in the HUD Demonstration. The area wiped was 1-ft2 for
floor samples and the entire accessible surface for window stool
and channel samples.
Soil samples were collected with a 1-inch internal diameter
soil recovery probe and a 12-inch stainless steel core sampler
with cross-bar handle and hammer attachments. Each sample was a
composite consisting of three to five soil cores, each 0.5 inches
in depth as measured from the top of the soil surface.
2.3.2 Sample Preparation and Chemical Analysis
Dust vacuum samples were analyzed using a modified version
of EPA SW-846 Method 3050, followed by EPA SW-846 Method 6010,
Inductively Coupled Plasma (ICP). Lead levels in sample digests
which fell below ten times the ICP instrumental detection limit
were reanalyzed by EPA SW-846 Method 7421, Graphic Furnace Atomic
Absorption Spectrometry (GFAA). Gravimetric analysis of the
sampling cassettes was performed in a humidity-temperature
stabilized environment prior to field collection and prior to
digestion in order to measure the amount of dust collected and
calculate results on a concentration basis.
16
-------
Dust wipe samples were first prepared using an ashing
procedure followed by digestion using a modified version of NIOSH
7082, and then analyzed by Flame AA (SW-846 Method 7000 Series).
Soil samples were first prepared using a drying and
homogenization step followed by digestion using a modified
version of EPA SW-846 Method 3050, and then analyzed using a
modified version of EPA SW-846 Method 6010, ICP.
2.4 FIELD EXPERIENCES
In general, the field sampling protocols for dust and soil
performed well in the field. Two issues that warrant special
mention are the time required to sample at each house, and the
efficiency of the vacuum nozzle for collecting interior dust
samples.
Initially, it was estimated that sampling at each house
would take from two to three hours. However, the time actually
required in the Pilot Study was from five to seven hours for a
single house. This was with a field crew of three people
collecting between 33 and 38 samples per house. Factors
contributing to the time required included cleaning of the
sampling equipment between each sample, the time required to
collect vacuum dust samples, and the initial learning curve for
field sample collection.
The time required to sample interior dust is inversely
proportional to the efficiency of the sampling protocol. During
the training period for the Pilot Study, it appeared that the
vacuum protocol selected, and in particular the sampling device
used, was inefficient at collecting all of the dust from several
common surfaces (e.g., floors, window channels). Specifically,
the sampler appeared to be incapable of collecting all the dust
that was visibly present in a number of cases. Subsequent to the
pilot field work, a new vacuum sampler was developed for the full
study.
17
-------
3.0 DATA MANAGEMENT AND PRELIMINARY ANALYSES
This section presents a summary of the data management and
data verification activities, as well as preliminary data
analysis leading up to the full statistical analysis.
3.1 DATA MANAGEMENT
There were several sources of data in the Pilot Study,
including the recruitment, field data collection, and laboratory
analysis activities. The individual data sources are described
below:
Cover Sheet - Contains unit information such as
city, address, categorized abatement method, name
of unit occupant, owner, and members of the
sampling team. Each record corresponds to a
different housing unit.
Interview - Contains interview questionnaire
information regarding demographics, habits, pets,
hobbies, etc. of the occupants. Each record
corresponds to a different housing unit.
Visual Observation Form - Contains information on
the physical surface condition of abated
components. Three interior rooms as well as the
exterior of each house were observed in the Pilot
Study. Each record corresponds to a different
abated component, and the current condition of
these observed components is designated.
Field Sample Log - Contains information used to
identify the planned sampling location, sample
medium, sample type, etc. and the link between
field and laboratory sample IDs. Each record
corresponds to a planned and/or collected field
sample.
Field Analytical Results - Contains laboratory
analysis results for dust and soil samples. Each
record corresponds to a collected and analyzed
field sample.
Quality Control Analytical Results - Contains
quality control results for each laboratory batch.
Each record corresponds to a reported calibration
or quality control sample.
18
-------
Data from the Cover Sheet, Interview, Visual Observation Form,
Field Sample Log, and Analytical Results were processed using the
procedures stated in Section 5.0 of the Quality Assurance Project
Plan (QAPjP) for the Pilot Study (Battelle and MRI, 1991) . These
data are organized into SAS datasets.
3.1.1 Sample Collection
A summary of the field samples planned, field samples
collected, analytical data received, and analytical data used in
the statistical analysis for each unit is provided in Table 3-1.
For completeness, Table 3-1 also summarizes all of the laboratory
QC, trip blank and laboratory comparison data received from the
primary and secondary laboratories. A further breakdown of this
information by sample type and medium is provided in Table 3-2.
There were a total of seven housing units recruited for the
Pilot Study, six participating and one alternate. There were 258
samples planned; 228 were actually collected, and 225 analytical
results were reported by the primary and secondary laboratories.
A total of 19 extra (i.e., unplanned) field samples was
collected:
three small nozzle field blanks,
one small nozzle air duct,
two replacement samples for samples mistakenly
collected with the wrong name-brand baby wipes,
one sample taken to replace a sample with an excessive
amount of saw dust, and
twelve soil samples split to create 12 extra samples
for interlaboratory comparison.
Among the planned samples, 72 (36 pairs) were to be
collected for the vacuum versus wipe comparison. All twelve of
19
-------
Table 3-1. Unit Summary of Sample Collection
Housing
Unit ID
33
43
17
19
80
51
Sub-Total
ICPS(c) for
Reported GFAA
Lab QC Samples
Baltimore
Lab comparison
Trip Blanks
Total
Abatement
Method
Control
Removal
Removal
Control
Encaps/Enclose
Encaps/Enclose
Renovation
Performed
None
None
None
Partial
None
Full
Planned
Samples
43
43
43
43
43
43
258
258
Planned
Samples
Collected
34
38
34
33
36
34
209
209
Extra
Samples
Collected
3
2
2
5
4
3
19
19(b)
Analytical
Results
Received
37
40
36
35(a)
40
37
225
33
383
38
53
732
Analytical
Results Used
in Analysis
37
40
36
33(b)
40
36(b)
222(b)
33
383
38
53
729(b)
(a) Three collected samples do not have data reported: two collected with wrong name-brand baby wipes, one sample spilled
in laboratory. Two samples were mistakenly collected into the same cassette (03 and 09). Therefore, only one analytical
result was received but is counted as two results.
(b) Three samples were deleted from the analysis. Unit 19, sample #03 and #09 as described above and unit 51, sample #12
because cassette was filled with sawdust after only one square foot had been sampled.
(c) Inductively coupled plasma atomic emission spectroscopy.
20
-------
Table 3-2.
Summary of Planned Samples, Collected Samples
and analytical results used in analysis
Sample Type
and Medium
Regular
1 . Vacuum Dust
a. Floor
b. Window Stool
c. Window Channel
d. Upholstery/Carpet
e. Air Duct
f. Entry Way
2. Soil Core
a. Foundation
b. Boundary
c. Entry Way
Vacuum vs. Wipe
3. Vacuum Dust
a. Floor
b. Window Stool
c. Window Channel
4. Wipe Dust
a. Floor
b. Window Stool
c. Window Channel
Quality Control
5. Interlab Comparison
a. Vacuum Dust (Fir)
b. Soil Core
6. Field Blanks
a. Vacuum Dust
b. Wipe Dust
c. Soil Core Liners
7. Side-by-side
a. Vacuum Floor Dust
b. Soil Core
Total
Planned
Samples to be
Collected
24
24
24
12
12
12
12
12
12
12
12
12
12
12
12
6
6
6
6
6
6
6
258
Planned
Samples
Collected
24
15
8
8
10
12
12
12
12
12
10
3
12
12
6
6
6
6
6
6
5
6
209
Extra Samples
Collected
1
4(3)
2
3
1
1
19
Analytical
Results Reported
24
15
8
8
11
12
16
14
12
12
10
3
12
12
6
6
12
9
6
6
5
6
225
Analytical
Results Used in
Data Analysis
22
15
8
8
10(c>
12
16
14
12
12
10
3
12
12
6
6
12
9
6
6
5
6
222(c)
(a> A total of 12 soil samples were split, and half of each sample was sent to the primary laboratory and the secondary
laboratory for chemical analysis.
(b) Two samples were collected into the same cassette (19-03 floor and 19-09 air duct) and only one analytical result was
received but counted as two results.
(c) Samples 51-12, 19-03, and 19-09 were excluded from the analysis.
21
-------
the planned vacuum-wipe floor sample pairs were collected. Among
the window stool samples, all 3 of the planned wipe-wipe pairs
were collected, 2 of the 3 planned vacuum-vacuum pairs were
collected, and all 6 of the vacuum-wipe pairs were collected.
Among the window channel samples, 2 of the 3 planned wipe-wipe
pairs were collected, 1 of the 3 planned vacuum-vacuum pairs was
collected, and 1 of the 6 planned vacuum-wipe pairs was
collected. In addition, 1 window channel wipe sample was
collected, but the corresponding vacuum sample was not.
3.1.2 Analytical Data Transfer
Ten batches of data were received from the primary
laboratory: four batches of vacuum cassette dust data, four
batches of wipe dust data, and two batches of soil data. The
secondary laboratory provided one batch of laboratory comparison
data.
The primary laboratory also reported data for a total of 53
trip blanks: one regular batch of 52 trip blank data, and one
trip blank that was reported with a batch of vacuum cassette
data. There were two batches of ICP results reported, including
29 data for regular samples and 4 quality control results. These
ICP data were used to compare with GFAA results generated for the
same samples.
The secondary laboratory provided 18 laboratory comparison
data for samples collected in Denver that are part of the
subtotal in Table 3-1.
3.1.3 Sampling and Analysis Deviations
A sampling and analysis deviation was considered to have
occurred if any of the following criteria was met:
a sample was not collected in the field
more than 43 planned samples were collected at any one
house
22
-------
a collected sample's analytical results were not
reported, or
a sample was collected with the wrong protocol.
The last two criteria are pertinent to the analysis and are
further discussed below.
There were three samples collected in unit 19 for which
analytical results were not reported: two samples collected with
the wrong brand of baby wipes1, and one sample spilled in the
laboratory. There were also three sampling protocol violations.
In unit 19, two planned samples were collected into the same
cassette; thus, only one analytical result could be reported for
two different planned samples. In unit 51, the analytical
results were invalidated for a regular cassette that was filled
with sawdust after sampling only one square foot. These three
samples were excluded from all of the statistical analyses
discussed in Section 4.0 of this report.
3.1.4 Experiences Prom The Pilot Study
The field preparation, forms processing, data transfer, and
data tracking went well for the Pilot Study. However, the
following observations will be addressed to make improvements to
the data management system:
Use of a separate field ID and laboratory ID proved to
very helpful, for example it helped determine that a
trip blank was reported as a regular sample.
Development of a more detailed set of instructions for
completing field data forms is needed since more than
one sampling team is usually collecting samples.
Scheduling a meeting of the field and data management
personnel after returning from the field was useful to
alert everyone of unusual occurrences in the field.
1 It was discovered that baseline measures of lead vary
across brands. For comparability, it was decided to use the same
brand of wipes as was used in the HUD Demonstration.
23
-------
Development of a system to formally capture laboratory
comments about individual sample results was useful in
the subsequent statistical analysis.
3.2 OUTLIER ANALYSIS
This section begins the presentation of results from
statistical analysis of the CAP Pilot study data. A complete
listing of these data is provided in the Appendix. The data are
sorted by unit ID, the room or yard in which the sample was
collected, and the component sampled. The sample location
variable is a general location measure (e.g., within a room)
which facilitates the pairing of side-by-side samples for later
analysis. Only two field samples, other than field blanks, had
levels of lead below the detection limit. These samples, floor
wipe measures in unit 33, were set at the detection limit of
13.77 ,ug/ft2.
In this section are presented the outlier analysis
statistical approach, the outliers identified, and the findings
of the laboratory review of the outlier data.
3.2.1 Outlier Analysis Approach
Formal statistical outlier tests were performed on the
natural logarithms of the lead concentration data and lead
loading data. Data were placed into groups of comparable values,
and a maximum absolute studentized residual procedure was used to
identify potential outliers. When a potential outlier was
identified, that value was excluded from the group, and the
outlier test was performed again. This procedure was repeated
until no additional outliers were detected. After all potential
outliers were identified, a list of these samples was sent to the
laboratory for rechecking. The following sections further
explain this procedure.
24
-------
3.2.2 Data Groups
The following homogeneous groups of data were identified for
each indicated sample type:
Vacuum cassette dust samples (7 groups): air duct,
upholstery (including bed coverings and throw rugs),
interior entryway, floor (excluding entryway), window
stool, window channel, and floor (including entryway);
Wipe dust samples (3 groups): floor, window stool, and
window channel;
Soil Samples (4 groups): boundary, foundation, exterior
entryway, and all exterior samples combined.
Initially, data for all six units in the Pilot Study were
combined for the outlier tests in these groups. Subsequent
outlier tests were also performed by segregating the data in each
group by abatement method and by housing unit, but only if there
were at least three samples in the resulting subgroups.
3.2.3. The Outlier Test
The SAS procedure GLM (SAS PC, ver. 6.04) was used to
compute the studentized residual for each data value in a group
by fitting a "constant" model (i.e., mean value plus error term)
to the log-transformed data in each group. The absolute values
of the studentized residuals were then compared to the upper
.05/n quantile of a t distribution with n-2 degrees of freedom,
where n is the number of data values in the group. If the
maximum absolute studentized residual was greater than or equal
to the .05/n quantile, the corresponding data value was flagged
as a potential outlier. The outlier test was then repeated,
excluding additional potential outliers, until no more outliers
were detected. Table 3-3 lists the outliers found as a result
this test.
Of the 135 lead loading values reported, four (or 3%) were
identified as potential outliers. This includes 3 out of
25
-------
Table 3-3. CAPS Pilot Study Outliers
LOADING OUTLIERS
Sample
Processing
Batch #
CRS
CIS
CSS
WSS
Laboratory
ID
900383
900337
900041
900849
Medium
Cassette
Cassette
Cassette
Wipe
Study ID/Sample ID
80/06
19/08
51/08
51/34
Loading (ug/ft2)
13087.15
187.30
59.42
1628.77
CONCENTRATION OUTLIERS
Sample
Processing
Batch #
CLS
CLS
CRS
SSS
SSS
SSS
SSS
CKC
CSS
Laboratory
ID
900357
900009
900383
901067
901057
901095
901074
901119
900105
Medium
Cassette
Cassette
Cassette
Soil
Soil
Soil
Soil
Cassette
Cassette
Study ID/Sample ID
19/09
51/21
80/06
43/26
17/23
80/24
33/27
17/01
17/32
Concentration (ug/g)
69.53
4026.20
61573.85
289.61
363.88
941.59
167.51
50.00
63.69
26
-------
105 cassette samples and 1 out of 30 wipe samples. Of the 153
lead concentrations reported, 9 (or 6%) were identified as
potential outliers. This includes 5 out of 105 cassette samples
and 4 out of 48 soil samples.
3.2.4 Resolution of Outlier Questions
Potential outliers were screened by a statistician to
eliminate those which were merely numerical anomalies due to
sample sizes of only 3 or 4. A list of the remaining outliers
was sent to the laboratory for review. After rechecking, the
laboratory verified that no transcription errors had occurred in
reporting the results for these samples.
3.3 DUST COLLECTED AND AREA SAMPLED
When planning a field study to collect dust samples in a
residential setting, information about the amount of dust
collected and the square footage sampled is invaluable for
interpreting the resulting lead loadings and concentrations.
Detection limits for dust lead concentrations are a direct
function of the amount of dust collected. The area sampled
information for window stools and channels is quite useful for
design purposes since it provides information on the size of
these components. In Table 3-4, descriptive statistics are
reported by sample type for the amount of dust collected (mg) by
the vacuum sampling method, and the area sampled (ft2) by both
the vacuum and wipe sampling methods. The descriptive statistics
presented are the geometric mean, logarithmic standard deviation,
minimum, and maximum for the amount of dust collected and the
arithmetic mean, standard deviation, minimum, and maximum for the
area sampled. The symbols (abbreviations) used in Table 3-4 to
represent the different sample types are described in Table 3-5.
These symbols will be used repeatedly in the text, tables, and
figures in this report.
It is important to understand what is meant by "abatement
effect" in this study. The control houses were houses
27
-------
Table 3-4.
Descriptive Statistics for Amount of Dust Collected
(mg) and Area Sampled (ft2) by Sample Type
ARD
BRU
EWY-I
FLR-V
FLR-W
WST
WST
(1/2)
WCH
WCH
(1/2)
Amount of Dust (mg)
N
Geometric Mean
LN Standard Deviation
Minimum
Maximum
10
154.26
1.13
25.4
561.3
8
48.62
1.24
8.7
388.6
12
287.35
1.37
13.3
1819.0
39
204.70
1.36
21.3
1902.5
0
15
49.38
1.20
6.2
283.6
10
26.78
1.39
4.8
385.5
8
221.96
0.80
78.0
1001.4
3
279.64
1.48
103.6
1522.9
Area Sampled (ft2)
N
Arithmetic Mean
Standard Deviation
Minimum
Maximum
10
0.54
0.59
0.22
1.67
8
1.00
0.00
1.00
1.00
12
4.00
0.00
4.00
4.00
46
3.98
0.15
3.00
4.00
13
1.00
0.00
1.00
1.00
15
1.23
0.70
0.35
2.60
22
0.65
0.44
0.23
1.56
8
0.42
0.26
0.09
0.88
9
0.20
0.10
0.06
0.34
Table 3-5.
Symbols Used to Denote Sample Types in Tables and
Figures
Sample Type
Air Duct Dust
Bed Cover-Rug-Upholstery Dust
Entryway Dust (Interior)
Floor Dust
Window Stool Dust
Window Channel Dust
Soil
Symbol
ARD
BRU
EWY (-I)
FLR
FLR-V
FLR-W
WST
WST(1/2)
WST-V
WST-W
WCH
WCH(1/2)
WCH-V
WCH-W
BDY
EWY-0
FDN
Description
Dust samples from an air duct
Dust samples from a bed cover, rug, or upholstered furniture
Dust samples from inside an entryway
Dust samples from the floor
Vacuum dust samples from the floor
Wipe dust samples from the floor
Dust samples from a window stopj
Dust samples from a split window stool
Vacuum dust samples from a window stool
Wipe dust samples from a window stool
Dust samples from a window channel
Dust samples from a split window channel
Vacuum dust samples from a window channel
Wipe dust samples from a window channel
Soil samples from the boundary of the property
Soil samples from outside an entryway
Soil samples near the foundation of the unit
28
-------
tested by HUD and found to be relatively free of lead-based
paint. Therefore these houses were not abated. The abated
houses were houses tested by HUD and found to contain sufficient
lead-based paint to warrant abatement. Therefore these houses
were abated. The data analyzed in this report were obtained by
dust and soil sampling conducted subsequently at both types of
homes. Thus the "abatement effect" is really a measure of the
difference in lead levels between abated houses (which were
abated due to presence of lead) and unabated houses which were
previously identified by XRF as being relatively free of lead-
based paint. In some sense, it is a measure of how well
abatement brings dust and soil lead levels in line with
corresponding levels in houses determined to be relatively free
of lead-based paint.
The amount of dust collected is illustrated graphically in
Figure 3-1. The area sampled is similarly illustrated in Figure
3-2. In these figures, box and whisker plots are displayed for
each sample type. The boxplot is a useful scheme for portraying
the center, scatter, and skewness of a dataset. The lower and
upper quartiles of the data are represented by the bottom and top
of the box, respectively. At least 50% of the data lies within
the box. The bar within the box represents the median of the
data. The lower and upper tails of the distribution of the
sample data are represented by the whiskers extending from the
bottom and top of the box. Extreme data points are classified as
either minor (pluses) or extreme outliers (stars) based on the
distance of the data value from the quartiles relative to the
distance between the upper and lower quartiles (interquartile
range). The arithmetic mean of the data is portrayed with a
diamond. Split window stools and channels in the bridge rooms
are separated from full window stools and channels in the regular
rooms since the split stools and channels provide only about half
the sampling area of a full window stool or channel, as shown in
Table 3-4 and Figure 3-2.
29
-------
\
s
(8m)
4J
0)
to
O
0)
JJ
U
0)
O
U
JJ
to
O
-p
1
CO
0)
D)
-H
Eh
30
-------
\
CO
04
pofdraBg
0)
0)
to
J?
JJ
M-l
-0
0)
m
id
0)
cs
0)
I
31
-------
As illustrated in Figure 3-1, the amount of dust collected
by the vacuum sampler was seldom less than 10 mg (the amount
targeted by the laboratory chemists in the study plans),and never
exceeded 2 grams (2000 mg). Bedcover, rug, and upholstery
samples, and window stool samples, provided the smallest amounts
of dust primarily due to the area sampled. The large amount of
dust collected from window channel samples is due to a very high
dust loading which compensates for the very small area available
for sampling (less than for window stool samples).
As illustrated in Figure 3-2, the area sampled for bedcover,
rug, and upholstery (1 ft2), interior entryway (4 ft2), and floor
wipe (1 ft2) samples was always the same. The area sampled for
floor vacuum samples was 4 ft2, with a single exception. The
average area sampled for air duct samples was slightly over 1/2
ft2, for full window stool samples was slightly over 1 ft2, and
for full window channel samples was slightly under 1/2 ft2.
3.4 COMPARISON OF ICP AND GFAA RESULTS
The protocol for analysis of the vacuum cassette dust
samples called for an initial analysis by ICP. This analysis
method was denoted by ICP-V in the previous section. If the ICP
result was less than 10 times the ICP detection limit, the sample
was reanalyzed by GFAA. The ICP and GFAA results for the samples
reanalyzed by GFAA are reported in Table 3-6. The table presents
the location, type and amount of lead collected (/ig lead per
sample) for each sample. The samples are listed in increasing
amounts of lead as measured by ICP. These results are
illustrated graphically in Figure 3-3. Separate plotting symbols
are utilized in the figure to distinguish between the various
sample types. The three samples reported by ICP as having a
negative concentration (i.e., well below the detection limit) are
plotted against 0.1 /ig/sample on the ICP axis. As shown in the
figure, agreement between the two methods was very good,
indicating that the supplementary analysis by GFAA was probably
unnecessary.
32
-------
Table 3-6.
TCP and GFAA Measurements (Lead Loading and Lead
Concentration) for Samples Analyzed by GFAA
Unit
544
564
N/A
571
506
506
564
507
571
544
564
571
571
588
564
506
506
507
506
506
507
506
544
544
507
506
507
N/A
N/A
N/A
N/A
564
564
564
Room
KIT
LVG
N/A
KIT
BD2
BD2
LVG
DIN
BAT
BD1
LVG
BD3
EWY
BAT
BD1
LVG
BD2
DIN
BD2
LVG
LVG
BD2
LVG
KIT
LVG
LDY
KIT
N/A
N/A
N/A
N/A
KIT
EWY
LVG
Component
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
BRU
WSL
BRU
FLR
N/A
BRU
BRU
FLR
FLR
FLR
WSL
FLR
WSL
WSL
FLR
WSL
WSL
WSL
N/A
N/A
N/A
N/A
WSL
FLR
FLR
MRIID
900098
900284
900458
900436
900273
900271
900353
900239
900435
900114
900373
900456
900446
900033
900360
900261
900249
900241
900255
900250
900197
900247
900103
900119
900205
900274
900229
900484
900485
900481
900468
900351
900347
900365
Sample Type
Field Blank
Field Blank
Trip Blank
Field Blank
Field Blank
Field Blank
Field Blank
Field Blank
Field Blank
Regular
Regular
Regular
Regular
Field Blank
Regular
Regular
Regular
Regular
Regular
Regular
Regular
Regular
Regular
Regular
Regular
Regular
Regular
Reference Material
Reference Material
Reference Material
Reference Material
Regular
Regular
Regular
ICP Amount
(/yg/sample)
0*
0"
0*
0.47
0.51
0.53
0.62
0.72
0.82
0.87
1.544
1.74
2.176
2.638
3.644
3.942
3.952
4.120
4.23
4.501
4.78
5.215
5.863
6.059
6.077
6.372
6.769
12.3038
13.11
13.43
39.84
62.240
78.878
285.538
GFAA
Amount
(//g/sample)
0.061
0.22
0.16
0.44
0.16
0.21
0.20
0.17
0.24
1.02
2.090
1.56
2.947
3.131
4.216
3.777
3.896
3.920
4.19
4.660
4.74
4.254
5.824
6.400
5.979
6.504
6.626
12.134
13.0814
10.74
40.1520
68.15
91.606
326.298
* The calculated final concentration was negative.
Note: Some of the ICP amounts are estimates. They were calculated using the weight and concentration of the sample.
33
-------
A
X
A
X
** x *
X
x
X
X
EJ
o
9 1
O i
O [
O i
«
+ *
+ *
+ *
a
Q)
m
a)
4J
4J
0)
a>
to
o
o
M-I
ra
4J
I
a
n>
d)
H
En
O
to
(0
4J
T) ft
id 3
0) Pn
H U
Pi >i
H
d
-------
4.0 DATA INTERPRETATION
Interpretation of the study data began with the production
of descriptive statistics in both tabular and graphical form.
These descriptive statistics are presented in Section 4.1. Next,
statistical models were fitted to the measurement data to
estimate various variance components (unit-to-unit, room-to-room,
exterior side-to-side, sampling location-to-sampling location,
and duplicate-to-duplicate) and to estimate the effects of
renovation and abatement. The statistical models employed are
defined in Section 4.2.
It is important to understand what is meant by "abatement
effect" in this study. The control houses were houses tested by
HUD and found to be relatively free of lead-based paint.
Therefore these houses did not warrant abatement. The abated
houses were houses tested by HUD and found to contain lead-based
paint. These houses were abated. The data analyzed in this
report were obtained by dust and soil sampling conducted
subsequently at both types of homes. Thus the "abatement effect"
is really a measure of the difference in lead levels between
abated houses and unabated houses. In some sense, it is a
measure of how well abatement brings dust and soil lead levels in
line with corresponding levels in houses determined to be
relatively free of lead-based paint.
Modeling results are presented in Section 4.3. Two
different models were fitted to the data. The first model
contains only an overall geometric mean and random effects; no
fixed effects are included. The purpose of this model is to
assess general variability without attributing the variability to
any particular cause. Results from the first model are reported
in Section 4.3.1. The second model fitted includes fixed-effect
terms to represent renovation and abatement effects which attempt
to explain portions of the unit-to-unit and room-to-room
variability. Results from the second model may be found in
Section 4.3.2.
35
-------
In Section 4.3, results for dust samples are reported
separately for two different statistical models, three different
measured values (lead loading, lead concentration, and dust
loading), and in some cases two different sampling methods
(vacuum and wipe). In Section 4.4, the modeling results are
summarized for each of the six dust sample types (air dust,
bed/rug/upholstery, entryway, floor, window stool, and window
channel). These summaries span the results from the two
different statistical models and the different measurement types
for each sample type. Section 4.5 provides similar summaries for
the three soil sample types (boundary, entryway, and foundation).
Having summarized the data by sample type, relationships
between the sample types were then examined. These relationships
are characterized in terms of correlation matrices and
scatterplot matrices in Section 4.6.
As stated earlier, one of the objectives of the Pilot Study
was to compare the vacuum sampling protocol with the wipe
sampling protocol. Paired measurements for these two sampling
protocols are compared statistically in Section 4.7. Finally, in
Section 4.8, the data collected in this study were compared to
data previously recorded for the housing units as part of the HUD
Demonstration.
All sampling was done in six houses, and the results should
be interpreted with this in mind. As a result of the analyses
and comparisons performed, the following broad conclusions may be
drawn:
1. Units under renovation had relatively high interior lead
loadings on readily available surfaces such as
entryways, floors and window stools; floor lead loadings
in the units undergoing full renovation were estimated
to be 70 times higher than those in control units; both
higher lead concentrations in the dust (5 times higher)
and higher dust loadings (14 times higher) appeared to
contribute to the higher lead loadings.
2. There is some evidence that abated units had higher
interior lead loadings on readily available surfaces; it
36
-------
appears that this is due primarily to higher lead
concentrations.
3. For floor lead loadings, abated rooms had lead levels
which were comparable to those in control units;
however, lead loadings in unabated rooms in abated units
were 10 times higher than abated rooms in the same unit;
higher dust loading appeared to be the primary cause.
4. With window stools as the exception, differences in dust
lead loadings among different sample types can be
attributed to differences in both dust lead
concentration and dust loading on the surface being
sampled; dust lead concentration and dust lead loading
were positively correlated from sample type to sample
type.
5. The higher lead loadings for window stools relative to
floors can be attributed primarily to higher lead
concentrations in the dust, and not to higher dust
loadings.
6. Soil lead concentrations for the three types of samples
collected (boundary, entryway, and foundation) were
highly correlated from unit to unit, both before and
after correcting for renovation and abatement effects.
Also, the lead concentration in boundary soil samples
was significantly lower than that in entryway and
foundation soil samples.
7. Interior dust lead concentrations for the six types of
samples collected (air duct, bed/rug/upholstery,
entryway, floor, window stool, and window channel)
generally were not highly correlated even after
correction for renovation and abatement effects;
exceptions were:
entryway samples with floor samples before
correction for renovation and abatement effects, and
air duct samples with bed/rug/upholstery samples and
floor samples, window stool samples, and window
channel samples as a group after correction for
renovation and abatement effects.
8. Interior dust lead concentrations were generally
correlated with soil lead concentrations:
37
-------
before correction for renovation and abatement
effects, entryway, floor and window stool dust lead
concentrations were all positively correlated with
soil lead concentrations for all three soil sample
types;
after correction for renovation and abatement
effects, dust lead concentrations for all interior
dust sample types, except entryway samples, were
positively correlated with soil lead concentrations
for all three soil sample types.
9. Based on paired data for the two sampling procedures,
the wipe sampling procedure appeared to produce lead
loadings on the order of 5 times higher than the vacuum
method; this would be consistent with a sampling
efficiency of approximately 10-20% for the vacuum
sampler.
10. The CAP Pilot Study soil concentration data were highly
correlated with HUD Demonstration soil concentration
data with the HUD Demonstration data being 25% higher on
average; floor lead loadings for the two studies did not
appear correlated.
11. For floor and window stool lead loadings and soil lead
concentrations, results from both the CAP Pilot Study
and the HUD Demonstration appeared to be somewhat
positively correlated with XRF/AAS measurements of paint
lead loading from the HUD Demonstration. However, if
anything, window channel lead loadings appeared to be
negatively correlated with the XRF/AAS measurements.
This negative correlation cannot be explained simply by
window replacement as none of the windows were replaced
during abatement of the units examined in the Pilot
Study.
4.1 DESCRIPTIVE STATISTICS
Three basic types of measurements were examined for the dust
and soil samples. They are:
Lead Loading: Amount of lead (p.g) in household dust per
square foot (ft2) of surface area sampled
Lead Concentration: Amount of lead (jug) per gram (g) of
household dust sampled or amount of lead (/xg) per gram
(g) of soil sampled
38
-------
Dust Loading: Amount of household dust (mg) per square
foot (ft2) of surface area sampled.
Vacuum dust samples produce all three measurements. Wipe dust
samples produce only lead loading measurements since the amount
of dust collected cannot be determined. For soil samples, lead
concentration was determined because a volume, not a surface, was
sampled.
Descriptive statistics for all units combined are presented
by sample type in Table 4-1 for all three measurement types. The
abbreviations used to denote the different sample types have been
defined previously in Table 3-5. The descriptive statistics
reported include the number of samples, geometric mean, median,
arithmetic mean, logarithmic standard deviation, minimum, and
maximum.
Log-transformed responses (lead loadings, lead
concentrations, and dust loadings) were used in all of the
statistical analyses. Using log-transformed environmental lead
measures is common and supported in the literature. Reeves et
al. (1982) found that the normal distribution was statistically
rejected for each of the environmental measures they studied
(lead in paint, soil, and house dust), and that the data tend to
be closer in form to the log-normal distribution. Based on the
data obtained in this study, one obvious reason for using log-
transformed data is the fact that in many cases, the responses
range over two to three orders of magnitude (see Figures 4-la,
b), especially for lead loadings. Another justification for
using this transformation is that the geometric means are often
much closer to the median than the arithmetic mean (see Table 4-
1). This is evidence that the distributions are more symmetric
on a log scale than on a linear scale. Also, examining residuals
from a partial model fit to the log-transformed data (the full
model leaves only 2 to 4 degrees of freedom for error) including
the fixed effects and a random unit effect, only one of the
eighteen lead sample types (floor vacuum lead loading) was
39
-------
14-1
a>
rH
3
EH
^
I
s
§
§
§
£
LL
Z
s
o
>
GJ
Q
CD
>
1
r?
r
52
§
1
S
si
m
Q
9
CM
CM
1^
(O
in
-
CM
en
CM
oo
o
z
%
1
O)
_C
1
o
CO
.3
Sco oo co oo h-
co en r- co co
o w w T m en
O CO CM
oo co in
in CM t *t o> m
o ^ co CM co oo
>t ^ "* in co co
*- CM
co i^ co CM r^ co
o> CM r^ ^ t*~ in
o ^ y, co i^ cb
* ^ en *
CM T- CM
i- en
Si~- CM co o in
r- CM CM 00 »-;
Q
co
CD o> in ^- h- Tt
r- CD CM O) CO
co co o <-
T-
to ^ <° CM cri r-:
»- * *
CM O>
§tr T- o m o
m en i~- co CM
<£) CM CM fyj Q Q-j
en * en
T- 00
t^ 00 O CM O> T-
co en r^ ^- CM oo
^ ^~ CM co in
"
o *- o oo oo m
CO ^ 1^ 00 00 CM
» ^ ^ CD co
^ CO CM
T h-
in CM co in co in
CO CO CM CO 00 CO
co "° C: * cJ co
CM r^ h-
t-- 10
CO
t- CO CO CM O CO
CM CM T- m o T-
iri ^ * CM d cb
in 10 -^i-
CM *
CM
t^ T- CM CD CM 0
CM (O O CD ^t CM
TT ^ <° CM 00 CD
t- 00 CM
CO O
T- CO CM O CM 1^
co in co co co in
CO *~ ^ CM ID TT
h- (D CO
T- ^J-
oo ^ en co ^ co
CD >- m i^ ^t co
co <° °° ^ co en
^r en * co in co
oo CM o co in en
CM °° S2 c\i * cri
CD in *
CO CM M-
CO
in t co co o oo
o * co oq co in
t^l CM ^ in
in
r*- co 01 co o r^-
m co ^- en co T-
^- ^ co in rt
(O
i^ CM ^t co ^t in
oo t^ co en co r-»
T- "*" CO CO ^
in in t m T- o
co in en ^- i^ co
00 CO 00 00
ao a> *- cr> vt
-------
rejected as non-normal. However, when the untransformed data was
fit to this model, eight sample types were rejected, includingall
three floor dust lead responses, all window stool dust lead
responses, and both entryway dust lead responses.
The geometric mean and logarithmic standard deviation are
natural summary parameters for lognormally distributed data. The
geometric mean is calculated by first taking the natural
logarithm of the data values, calculating the arithmetic mean of
the logarithms, and then exponentiating (taking the antilog of)
the resulting arithmetic mean. The logarithmic standard
deviation is calculated by first taking the natural logarithm of
the data values, then calculating the usual standard deviation.
Figures 4-la through 4-lc contain box-and-whisker plots of
lead loadings, concentration and sample loadings for various
sample types. The symbols used in these plots have been defined
in Section 3.3.
Lead loading measurements along with the geometric mean lead
loading for all units are plotted versus sample type in Figure 4-
la. Similar plots for lead concentration and dust loading
measurements are presented as Figures 4-lb and 4-lc,
respectively. Figure 4-2 is a bar graph illustrating the
geometric means for all three measurement types by sample type;
and Table 4-2 presents geometric means for each individual
housing unit.
The geometric means from Table 4-2 are plotted versus unit
number in Figures 4-3 through 4-5. Figure 4-3 illustrates
geometric means for the floor and upholstery sample types: BRU,
FLR-V, FLR-W, and EWY-I. Lead loadings, lead concentrations, and
dust loadings are presented in Figures 4-3a, 4-3b, and 4-3c,
respectively. Figure 4-4 illustrates geometric means for the
window and air duct sample types: WCH-V, WCH-W, WST-V, WST-W, and
ARD. Lead loadings, lead concentrations, and dust loadings are
presented in Figures 4-4a, 4-4b, and 4-4c, respectively.
Finally, geometric mean lead concentrations in soil samples are
presented in Figure 4-5.
41
-------
1 1 rl i 1
* * 0 1 EE3H
I
_, ,
\
4
\
\
\
1
%^ 1
*
7^
^JL
^f
\
^
%.
a O O O O O -:
Is 8 8 8 0 "^ °
3 % H '*""
3s-
tn
A
H
m
0
rH
O
nt
-------
«-,
1
1 ^
, n
1 A I^~^l^^~~~*^~l 1
, ,
\
^r
j*
^v
SML
Ty
\
\
^jjL
i
V, f
T' 5
jf
\
<**
xs
\
^v
^
i^
9 p p P g P 5
I*"
3
T)
n)
0)
H
mean
u
rl
M
0)
3
0
0)
0>
O 0)
^.rH
«^ !jj
CO
0)
C*-»
rM
0)
0) 4J
tl -_!
M n
m 0
(d
%1
m
^
0 h
-H 0
0)
1-4 rt
4-> 0
0) ^J
u nJ
O 4J
o a
0)
o o
n> a
0) O
J O
xi
H
0)
1
fe
43
-------
*s
\
O
8
c
H
0
o
4J
a
§
o
H
0)
0)
0)
d
9
M
JJ
01 -U
4J
a
0) H
3 m
m
(0 ^t
0) ,Q
a
tI»4J
(d
O H
4J
a
8
o
H
0)
at
44
-------
dp
-E ,
T '
S
I
19
a
Oil
&
I
\
Q
O fl)
at H
0) (2,
a
4-1
o
at tn
o a
rH -H
d
C at
at o
a) H
H
4J
b to
a
^J
H
tn
H ^
H 0)
id =t
^ a
0) -H
u
a
a> o
O at
H M
M 4J
u C!
0) 0)
§ d
U (H
0) 0
O O
01
I
a>
M
g,
H
45
-------
Table 4-2. Geometric Mean for Lead Loading (jig/ft2) , Lead
Concentration (/xg/g) ,
Sample Type and Unit
and Dust Loading (mg/ft ) by
Unit
33
43
17
19
80
51
| ARD
N
Lead Loading
Lead Concent.
Dust Loading
N
Lead Loading
Lead Concent.
Dust Loading
N
Lead Loading
Lead Concent.
Dust Loading
N
Lead Loading
Lead Concent.
Dust Loading
N
Lead Loading
Lead Concent.
Dust Loading
N
Lead Loading
Lead Concent.
Dust Loading
2
649
875
742
2
1313
834
1574
2
38
511
74
1
57
624
91
3
505
861
587
0
BRU
1
4
117
32
2
14
141
102
1
1
67
15
2
28
483
58
2
6
151
43
0
EWY-I
2
6
106
60
2
12
394
29
2
23
270
86
2
44
193
228
2
4
275
16
2
415
1605
259
FLR-V
7
2
131
19
7
3
221
14
7
12
166
74
5
57
173
330
7
9
305
29
6
262
1227
213
WST-V
4
10
425
23
3
12
525
23
6
17
368
47
3
10
139
72
5
152
3828
40
4
273
1854
147
WCH-V
1
3697
7238
511
2
1658
1175
1411
1
977
1141
856
1
1201
368
3263
3
1954
2914
671
3
507
828
613
BDY
2
86
2
133
3
59
3
57
2
325
3
325
EWY-O
3
79
3
338
2
160
3
73
3
380
2
674
FDN
3
147
3
246
3
68
2
108
3
515
3
599
FLR-W
2
14
2
21
2
24
2
35
2
29
2
2498
WST-W
3
141
3
25
1
24
1
191
1
163
3
1345
WCH-W
0
3
518
0
1
1530
0
2
1112
46
-------
* * o .
\
^
\
^
^^
f\
9
<* A
'x 4
^b
' OT
\
^
\
/^
^
%,
9 o o o o o T-.
eo o o T- o
r^
<0
0)
rH
..
a
a>
rH
CO
a>
4J
a
rH
a
S1
TJ
(0
M
O
o
rH
rl
O
H
4J
H
^
^
S .
O 01
§ >M
I "&
0
-H
M tn
u C
0)-H
8-8
0) 0
O H
10
n
0)
1
-H
Pu
_
O)
C
'a
E
n
o
4)
1
.g
ra
renov
ra
r
(0
a
O)
c
1
a>
a>
a
a>
T
'c
D
(0
c
_o
a
0)
3
O)
'5
D)
0)
|
(0
*
^
tn
'c
3
47
-------
o
n>
o>
0.
CO
O
0)
o
g
?
.2
a
D)
'6
o>
|
3
10
(0
0)
8
(I)
M
g,
H
c
3
(D
|
0)
I
3
CO
10
in
48
-------
-G
8
4J
a
a
a)
a
0)
4J
a>
H
o
xi
ft
M
o
o
M
o
Mrs
ti ^
rt «n
O '
H
H rj)
U fl
0) -H
§?
0) 0
O rH
o
ro
4)
M
o>
I
4)
|
«
a
in
'£
49
-------
I I
i i i
O)
a o o o o o f-.
£ o o o c> T- o
Tl
0)
0)
ft
(Q
JJ
O
(0
8
8
H
O
H
jf
OJ M-l
0) -^
6 tn
o ~"
H
4-> C
0) -H
6 -0
O nS
0) 0
OJ rH
T
0)
a
is
a)
+5
C
O
0)
ra
'5
D)
re
0)
(0
>
D)
C
c
D
V>
i
in
'c
50
-------
8
Ol
flS
Q)
CO
0)
H
ft
CO
JJ
o
-H
id
W
9
o
t)
a
H
H
a
'oi
A Di
to ^
a
a c
0) O
*1
o (d
_j (j
li | i
-u C
0) 0)
0) O
o o
_c
"5.
i
^5
0)
_g
(D
o
a>
a
D)
0>
I
3
(A
10
01
C
D
(0
o
o
Q)
O
D)
o
Cfl
CD
C
D
Q)
g.
H
EE.
51
-------
8
4J
OH
O
H
-H
fl -U
Of 14
U
-H
M tn
4J (H
0) -H
§3
0) O
O H
u
0)
H
fa
a
(0
(0
"o
a
&
o
r
(0
a
o>
'o
D)
a>
a
E
'o
O)
(0
(0
LO
.*;
52
-------
Cn
^
d)
3.
o
H
0)
O
tf
0
o
I
a
(0
in
i^
o
0)
CO
c
n
1
0)
O)
8
1
C
(0
I
0)
k
"5
5
S>
I
CO
-------
The correlation between loadings and concentrations was
assessed for the six types of vacuum samples. Table 4-3
displaysthe estimated correlations for each of these along with
the significance level of each estimate. These estimates are
based on the log-transformed data.
For five of the six sample types the estimated correlation
was significantly different from zero. Pooling across sample
types, the average correlation was 0.77, and this was highly
significant.
Table 4-3. Loading versus Concentration Correlations for Dust
Samples
Sample
Type
Air Ducts
Bed/Rug/Upholstery
En try way
Floor
Window Stool
Window Channel
Across Sample
Types
Number of
Samples
10
8
12
39
25
11
105
Estimated
Correlation
.59
.72
.76
.69
.86
.62
.77
Significance
Level
.0738
.0461
.0041
.0001
.0001
.0433
.0001
The effect of renovation on dust lead loadings for floor and
bed/rug/upholstery samples is evident in Figure 4-3a. Control
Unit 19 was undergoing partial renovation and
encapsulation/enclosure Unit 51 was undergoing full renovation.
The remaining four units show similar lead loadings. Examination
of Figures 4-3b and 4-3c lead to the conclusion that renovation
produces both higher lead concentrations and higher dust loadings
for these sample types, which both contribute to higher lead
loadings. In contrast, no effect for abatement or abatement
method is evident in Figure 4-3.
54
-------
There is no clear renovation effect or abatement effect for
the sample types plotted in Figure 4-4, window and air duct
samples. One possible exception is that the renovation of Unit
19 has produced higher dust loadings where the dust has a lower
lead concentration. In Figure 4-5, three of the four abated
units show consistently higher soil lead concentrations across
all three sampling locations.
4.2 STATISTICAL MODELS
In this section, the statistical models that were fitted to
the lead loading, lead concentration, and dust loading data are
described. These models are the basis for the statistical
analyses described in Sections 4.3, 4.4 and 4.5. All statistical
models for dust samples contained an overall geometric mean. The
models contained random effects for unit-to-unit, room-to-room,
sampling location-to-sampling location, and duplicate-to-
duplicate variability. At the unit level, there were fixed
effects for renovation and abatement. At the room level, there
was a fixed effect for abatement. The mathematical form of the
fullest model for dust samples was
ln(Xijkm) = In(CGA) + In (BRENO) RENOi + ln(BHP)HPi + H± +
ln(BRP)RPi:j + Ri:) + Sijk + Dijkm (1)
for
i = 1, ... , 6 (# units)
j = 1, ... , # rooms/unit
k = 1, ... , # sampling locations/room
m = 1, ... , # duplicates/sampling location
where
55
-------
measured lead loading (L^^) , lead concentration
(Cijkm) , or dust loading (Dijkm) for the mth
(replicate) sample at the kth sampling location in
the jth room in the ith unit,
CGA = overall geometric average of the dependent variable
for unrenovated control units,
BRENO = fixed multiplicative increase in the dependent
variable due to an ongoing full renovation of the
unit,
RENOi = 1 if ith unit is being fully renovated (Unit 51);
1/2 if the unit is being partially renovated (Unit
19); zero if the unit is not being renovated (other
4 units),
BHP = fixed multiplicative increase in the dependent
variable due to abatement having been performed
somewhere in the unit,
HP-j^ = 1 if the ith unit was abated; zero otherwise,
H.j_ = random effect for the ith unit; assumed to follow a
normal distribution with mean zero and standard
deviation aH,
BRP = fixed multiplicative increase in the dependent
variable due to abatement having been performed
somewhere in the room,
RPij = 1 if the jth room in the ith unit was abated; zero
otherwise,
RJ^J = random effect for the jth room in the ith unit;
assumed to follow a normal distribution with mean
zero and standard deviation crRI
Sijk = random effect for the kth sampling location in the
jth room in the ith unit; assumed to follow a normal
distribution with mean zero and standard deviation
as,
D.j_.jkm = random effect for the mth sample at the kth sampling
location in the jth room in the ith unit; assumed to
follow a normal distribution with mean zero and
standard deviation aD (includes variability due to
the sample collection process and variability due to
the laboratory analysis process).
56
-------
Two versions of the model were fitted to the data. The first
model contained no fixed effects. That is, the terms in the
model involving RENO.^, HP^ and RP.^ were excluded. The second
version of the model included the fixed effects. The model was
tailored to each of the sample types as follows:
For air duct, bed/rug/upholstery, and entryway samples,
it was not possible to estimate sampling location-to-
sampling location and duplicate-to-duplicate
variability.
For floor wipe samples, it was not possible to estimate
room-to-room and sampling location-to-sampling location
variability.
For window channel wipe samples, it was not possible to
estimate room-to-room variability.
The room level abatement term, RPi7 was estimated only
for lead loadings from vacuum floor samples; it was not
statistically significant for any other measurement
type.
Because of an insufficient number of samples, it was not
possible to estimate abatement or renovation effects on
wipe loadings for window channels or bed/rug/upholstery
measurements.
The statistical model for soil samples was similar to the
model for dust samples. However, side-to-side replaced room-to-
room as the within-unit variability source. Since samples were
taken at only a single sampling location on each side of the
unit, the sampling location-to-sampling location random effect
was confounded with the side-to-side random effect. Also, since
exterior abatement information was not available by side of unit,
a fixed effect for abatement was included only at the unit level.
The mathematical form of the fullest model for soil
concentrations was
ln(Cijm) = In(CGA) + ln(BRENO) _ ... . _
Sij + Dijm (2)
57
-------
for
where
Cijm -
CGA =
B
RENO
RENO, =
B
HP
HI =
i = 1, ... , 6 (# units)
j = 1, 2 (# sides/unit)
m = 1, ... , # duplicates/side
measured lead concentration for the mth (replicate)
sample on the jth side of the unit in the ith unit;
overall geometric average of the lead concentration
for unrenovated control units,
fixed multiplicative increase in the lead
concentration due to an ongoing full renovation of
the unit;
1 if ith unit is being fully renovated (Unit 51);
1/2 if the unit is being partially renovated (Unit
19); zero if the unit is not being renovated (other
4 units),
fixed multiplicative increase in the lead
concentration due to abatement having been performed
somewhere in the unit;
1 if the ith unit was abated; zero otherwise,
random effect for the ith unit; assumed to follow a
normal distribution with mean zero and standard
deviation aH,
random effect for the jth side of the unit at the
ith unit; assumed to follow a normal distribution
with mean zero and standard deviation as,
random effect for the mth (replicate) sample on the
jth side of the unit at the ith unit; assumed to
follow a normal distribution with mean zero and
standard deviation aD (includes variability due to
the sample collection process and variability due to
the laboratory analysis process).
Just as for dust measurements, two versions of the model were
fitted to the soil concentrations, the first containing no fixed
effects. No tailoring of the soil concentration model was
necessary for the individual sample types.
58
-------
The following random effects were allowed to be correlated,
so that different samples and sample types within a unit, within
a room, on the same side of a unit, or from the same window could
be correlated:
unit-to-unit random effects for all dust measurements
and all soil concentrations within a unit
room-to-room random effects for all dust measurements
within a room
side-to-side random effects for soil concentrations on
the same side of a unit
sampling location-to-sampling location (window-to-
window) random effects for dust measurements within a
window.
As is standard with mixed models of this type, all other random
effect terms in the models were assumed to be independently
distributed.
All statistical analyses were performed with the Statistical
Analysis System (SAS) software. For each component sampled,
sample medium and response (lead loading, lead concentration, or
dust loading), the modeling results could be obtained from
several runs of the SAS PROC GLM procedure. The random effects
are specified in a RANDOM statement employing the test option, in
the order appearing in the tables. For both fixed effects and
random effects, all tests are based on Type I sums of squares
using the proper denominator based on expected mean squares
approximations. The fixed effect tests and estimates are those
obtained by including each fixed effect last among the fixed
effects, but before all random effects. Thus, each fixed effect
is tested for significance controlling for all other fixed
effects in the model, and comparing it to the proper linear
combination of random effects for an error term. A separate GLM
run would be required for each fixed effect in the model.
59
-------
For the estimates and confidence bounds of the random
effects, linear combinations of the observed mean squares were
used. Therefore, a generalization of Satterthwaite's
approximation (for the 2-sample t-test) is used to estimate
degrees of freedom.
In fact, the above procedure was implemented in SAS/IML to
avoid the need for multiple GLM runs and the resulting voluminous
output, including many pages listing all Type I estimable
functions.
4.3 MODELING RESULTS BY MEASUREMENT TYPE
Twenty-four (24) different types of measured values were
fitted to the statistical models described in Section 4.2. These
measured values fall into three main categories:
Lead Loading: Air duct, bed/rug/upholstery, interior
entryway, floor vacuum, window stool vacuum, window
channel vacuum, floor wipe, window stool wipe, and
window channel wipe samples
Lead Concentration: Air duct, bed/rug/upholstery,
interior entryway, floor vacuum, window stool vacuum,
window channel vacuum, boundary soil, exterior entryway
soil, and foundation soil samples
Dust Loading: Air duct, bed/rug/upholstery, interior
entryway, floor vacuum, window stool vacuum, and window
channel vacuum samples.
Results from fitting each type of measured value to two different
models are provided. Results from the first model, which
included no fixed effects, are reported in Section 4.3.1.
Results are reported in Section 4.3.2 for the second version of
the model, which included fixed-effect terms. The p-values
reported are the observed significance levels for the given test.
A reported p-value of 0.00 indicates that the actual p-value was
less than .005.
60
-------
4.3.1 Estimates of Variance Components With No Fixed Effects
The first model fitted to the 24 measured values contained
no fixed effects. The purpose of this model was to assess
general variability without attributing the variability to any
particular cause. Note that for all models, the dependent
variable was the logarithm of the measurement of interest.
The statistical models for dust samples always contain an
overall geometric mean and can contain random effects for unit-
to-unit, room-to-room, sampling location-to-sampling location,
and duplicate-to-duplicate variability. The statistical models
for soil samples always contain an overall geometric mean and can
contain random effects for unit-to-unit, side-to-side, and
duplicate-to-duplicate variability. As indicated in Section 4.2,
the model was tailored to the individual sample types. Thus,
some models contain only a subset of the four random effect
terms. Generally, if the variance component associated with a
random effect term can be estimated it is included in the model.
The results of fitting the random effect models to the 24
measured values are reported in Table 4-4. Results for lead
loading measurements, lead concentration measurements, and dust
loading measurements are reported in Tables 4-4a, 4-4b, and 4-4c,
respectively. The rows of the table are defined by the sample
type which can be vacuum, wipe, or soil, and the component type
which can take on the following values:
Vacuum: Air duct, bed/rug/upholstery, entryway, floor,
window stool, and window channel
Wipe: Floor, window stool, and window channel
Soil: Boundary, entryway, and foundation.
Each row of the table represents a separate fit of the model to a
particular set of measurements.
An estimate of the overall geometric mean is provided as the
top value in each box in the fourth column. The bottom value is
61
-------
rd
^
i
^
0)
rH
3
H
*
c
o
1
Q
T3
OJ
T)
C
*" E?.
CM «0
o
CM
re
>.
c
LLJ
II Vacuum
i^
* P
o c
en co oo
oo in *
p+ a> o
d in d
*-^
r^ 57 o
in o o
^^
in csT
en *
*"
d
S
o
o
LL
II Vacuum
CM
CO ?J
d ~
m ro
CO .tf CO
d ~ d
to r-
co *o Q
r^ O
CO ^ CO
O CO -
"" *
~+ 5)
0 »
r+ CN
CO «>
O
in
CM
"5
o
CO
Window
Vacuum
in
CO P
d ~
CO P O5
q oo ^-;
r-' O O
r> CM CM
00 00
^ d d
in CM
O CO O3
""""
in m
CM 00
"""""
Is
^
Window
Channel
II Vacuum
(O
CO (O
d ~
CM CO O
on oo o
~~*
in ?
on ^
*- in
r- 0)
in i">
d
CM
O
O
U-
.1
Q
^ co"
d ~"
O) 00
r^ CM o
d ~ d
in
~ d
en ^t oo
to r^ oo
"~~
CO CO
t- LO
*~ d
CM
~o
o
CO
0
c
Q.
§
re
a
c
re
I
-n OJ
_o c OT
QJ re 2_
° to Q-
a o c
ill
a3 -a
s
.§ « -2
fe > .i
01 a >
o o ai
-i I- -o
62
-------
0)
iH
9
)eviation*
(D
C
(0
S ID
3 °-
Q- ro
« CO
OC "3
O) r~
.S o
0
ID O
CO -1
o
o ^
DC
'E
15
1-
Geometric
Mean*
CO
a. CD
1 i/5
c
CD
C
o
a.
J
CN
O ~
^ CO CO
o o o
t 0
CO CO
o co
» CO
1 CO
*~ O
00
.c
a.
Bed/Rug/U
E
3
ID
>
0 _
in
CM «-
0
CO CO O
o ^- o
en co co
* CO O
o * o
'~'
f- CO r-
l> - O
o co o
r^ to
m *
d o
in co
m co
"6
en
eo
o
o
LL
E
D
O
ID
>
en
CM CM
o ~
* m. CN
ci d
CM en eo
o oo o
r- r-. O
o ? *
O CO O
«- CM O
w »-
in *>
*~"_
CM in
"* CD
in
CM
"o
0
Window S
E
3
O
IS
>
f
in
0 CO
o ~"
«- CO O
co en o
o m o
en CN oo
co eo o
O CM O
CM P
en in
"~~ *
CM CO
*~ 6
in
Boundary
_
o
CO
f
in ^
d ~
in CM p~
^ O CM
O r- O
co en oo
r^ m o
O CM 0
m P
O CM
"""""
co r^
en co
" CD
co
Entryway
0
CO
*- co
o ~
co in CM
co eo o
O * o
r
i- en en
00 * 0
O CM O
S5
*~ r^
CO O
N 6
2
Entryway
(2 outliers
deleted)
o
CO
px
«- in
d ~
Sen .-
o o
o in o
~
r- co in
r^ en o
O CM o
CO ?
on en
o eo
r. co
- CO
"6
-
Foundatior
o
CO
0)
a
ID
o
c
I
CD
E
o
T3
C
ID
O)
c
-4-<
CD
E
o
-o
CD
ID
CD
T3
CD >
=0 »
5 g,
E
1 1 1
t £ M
(D "O "O
.c -p., CD
*-£;>
3 -g fc
5 c «
w £ c
= £ 0)
a *
D) CL T3
O O C
I I CO
63
-------
,£3
4J
o
'it1
a)
H
8
H
*
*
c
g
0)
Q
ID
a
c
ID
CO
0)
yz to
1!
o> in
CC OT
J: o
ra
E o
ID O
W -"
o
o
cc
.*;
c
"S
1-
0
£j *
E §
0 .g
o> -S
O
0)
Q. 0)
p N
i CO
CO
c
0>
c
o
a.
E
o
o
0)
II
ID |-
co
§-
«- ~
to ? -
""""
(O O
^ CO
*~ SB
t- 0)
i- in
* d
o
o
Q
E
3
u
ID
co
r* en o
o £ d
co a? o
«- 1^ O
t- CO O
in co
* *
*" 2
CM CM
in ^>
o
O)
CO
0
o
u.
E
u
ID
CM (M
O
IN p *
d oi d
in CM oo
^ in *-
0 co o
0) 00
q en
^ ^
PN oo
d
in
CM
Window
Stool
|
3
0
ID
00 _
d ~
t- S a>
e*
o ^J- oo
*"*"
CM R
o *
1- IN
co oo
«0 r-
00 d
^
>
o c
o c
C ID
§6
E
D
D
u
ID
"S
ID
U
a>
S:
0}
o
a
c
ID
D)
C
O
a
o>
o>
a>
o>
o
ID
E
s;
>
. S
5 S
c §
d .2 S
0) T3 .2
T3 O c
III
11
w
o .« ro
Hi
S > .5
O) D- >
O O CD
-
64
-------
the logarithmic standard error of this estimate. The logarithmic
standard error is the standard error of the logarithm of the
estimate.
In the last four columns, the top value in each box is an
estimate of the corresponding variance component in standard
deviation form. As indicated previously, it was not possible to
include all random effects in every model. When it was not
possible to include a random effect in the model, the box
associated with the random affect is left blank and the line
separating it from one or more of the other boxes is eliminated.
In these cases, the random effects associated with these long
boxes are confounded and the standard deviation estimate reported
corresponds to the combined variability from all corresponding
random sources. That is, if a standard deviation estimate is not
reported for a particular source of variability, then the
estimate reported to the left of the blank area includes the
variability contributed by the source for which no estimate was
reported.
For example, for wipe floor samples, the estimate reported
in the unit standard deviation column is actually an estimate of
the combined unit-to-unit, room-to-room within unit, and sampling
location-to-sampling location within room variation. However,
the replicate sample standard deviation is indeed an estimate of
the side-by-side standard deviation of wipe floor samples.
The top value in each box in the fifth column of Table 4-4
is an estimate of the total standard deviation. Note that this
value and all other standard deviation estimates in Table 4-4 are
logarithmic standard deviations. For dust samples, the total
standard deviation is the standard deviation of a measured value
from a randomly selected duplicate sample from a randomly
selected sampling location in a randomly selected room in a
randomly selected unit. For soil samples, the total standard
deviation is the standard deviation of a soil lead concentration
from a randomly selected duplicate sample from a randomly
65
-------
selected sampling location on a randomly selected side of a
randomly selected unit.
In most cases, the total variance (the total standard
deviation squared) is simply the sum of the individual variances
(the individual standard deviations squared). This will not be
the case, however, if any of the individual standard deviation
estimates is reported as zero. Due to the small number of
degrees of freedom available for estimating certain variance
components, some of the individual variance estimates were
initially negative. Since all variances are by definition
nonnegative, when this occurred the estimate presented in Table
4-4 is zero. When calculating the total variance, however, it is
appropriate to use the negative estimate of an individual
variance component in the sum.
The value in parentheses below each standard deviation
estimate is the approximate number of degrees of freedom
associated with the estimate. The larger the number of degrees
of freedom, the more precise the estimate. In the unit, room or
side, and sampling location standard deviation columns, a value
is sometimes reported below the approximate degrees of freedom.
This value is the observed significance level (OSL) of the test
of the hypothesis that the corresponding standard deviation is
equal to zero. A small value of the OSL is an indication that
the standard deviation is significantly larger than zero. This
test can be performed for all but the lowest order variance
component (farthest to the right).
The variance component estimates are illustrated graphically
in Figure 4-6a for lead loading, in Figure 4-6b for lead
concentration, and in Figure 4-6c for dust loading. These
figures provide a pictorial view of the estimates in Tables 4-4a,
4-4b, and 4-4c, respectively. (In Figure 4-6b, EWY2-0 refers to
the analysis reported in Table 4-4b in which 2 outliers were
deleted.) For each sample type, the estimated standard
deviations have been squared to convert them to estimated
variances. In order to see how each variance component
66
-------
'
I
1
I
9 t
Sample Type
iponBnt:
Unit
Room
T.nr.fltinn
Replicate
Figure 4-6a. Variance component estimates from model with no
fixed effects: lead loading (^ig/ft2) .
67
-------
g
a
T
I
Variance Component
Unit
Room or Side
Sampling Location
Replicate
Sample Type
Figure 4-6b. Variance component estimates from model with no
fixed effects: lead concentration
68
-------
Ov
Variance Component:
Unit
Room
Sampling Location
Replicate
1
1
Sample Type
Figure 4-6c. Variance component estimates from model with no
fixed effects: dust loading (mg/ft2) .
69
-------
contributes to total variability, the estimates are stacked. In
most cases, the total height of the bar is the total variance
(square of the total standard deviation). This will not be the
case if any of the individual standard deviation estimates is
reported as zero, as discussed above.
The following subsection contains a discussion of which
variance components are estimable for which sample types. This
discussion is followed by individual summaries of the modeling
results for lead loading, lead concentration, and dust loading.
Later, in Section 4.4, global summaries of the modeling results
are presented by sample type (e.g., floor samples, window channel
samples).
Estimable Variance Components
For air duct, bed/rug/upholstery, and entryway vacuum
samples, the unit-to-unit variance component was estimable as
well as the combined room-to-room, sampling location-to-sampling
location, and replicate-to-replicate variance component. The
last three variance components are confounded here because only a
single sample was taken in each room.
For floor, window stool, and window channel vacuum samples,
all variance components could be estimated. There were more
vacuum samples for floors (39) and window stools (25) than for
any other sample type in the study.
For floor wipe samples, the replicate-to-replicate variance
component can be estimated as well as a combined unit-to-unit,
room-to-room, and sampling location-to-sampling location variance
component. The first three variance components are confounded
here because floor wipe sampling was conducted at a single
sampling location in a single room in each house.
For window stool wipe samples, all variance components were
estimated, but there was very little data available for these
estimates. By design, the unit-to-unit and room-to-room variance
components should be confounded, since window stool wipe samples
were to be taken in only one room per house. However, sometimes
70
-------
a secondary bridge room was selected, allowing the possibility of
assessing room-to-room variability when window stool wipe samples
were taken in both rooms.
For window channel wipe samples, it was possible to estimate
the combined unit-to-unit and room-to-room variance component,
the sampling location-to-sampling location variance component and
the replicate-to-replicate variance component. However, since
there were only six samples, few degrees of freedom are available
to estimate the variance components. The first two variance
components are confounded since window channel wipe samples were
taken in only one room per house.
The soil data followed a simple structure for all three
sample types (boundary, entryway, foundation). This structure
permitted estimates of the unit-to-unit variance component, the
combined side-to-side and sampling location-to-sampling location
variance component, and the replicate-to-replicate variance
component. For soil samples, the sampling location-to-sampling
location random effect is confounded with the side-to-side random
effect since samples were taken at only a single sampling
location on each side of the unit.
Lead Loadings
The following is a summary of the modeling results for lead
loading for the model with no fixed effects. These results are
reported in Table 4-4a and the variance component estimates
illustrated in Figure 4-6a.
The geometric average lead loadings for the different sample
types in decreasing order are:
Window channel (vacuum 1250 ^g/ft2, wipe 801
Air duct (308 /ig/ft2)
Window stool (vacuum 34 /zg/ft2, wipe 144 /ig/ft2)
Entryway (23 /ig/ft2)
71
-------
Floor (vacuum 13 /ig/ft2, wipe 51
Bed/rug/upholstery (8 /zg/ft2) .
Interestingly, the window channel samples had the lowest two
total standard deviations (1.25 for vacuum, 0.61 for wipe). For
the remaining sample types, the total standard deviation was
fairly consistent ranging from a low of 1.66 (bed/rug/upholstery)
to a high of 2.04 (vacuum window stool).
The unit-to-unit variance component is statistically
significant (at the 0.05 level) for only two sample types: floor
vacuum samples and floor wipe samples. This variance component
is marginally significant for air duct, entryway, and vacuum
window stool samples. The estimated variance component is
negative for bed/rug/upholstery and vacuum window channel
samples. With the exception of these last two sample types, the
unit-to-unit variance component is a substantial contributor to
total variability. Of those sample types for which the room-to-
room variance component could be tested for significance, it is
significant for only vacuum floor and vacuum window stool
samples.
Lead Concentrations
The following is a summary of the modeling results for lead
concentration for the model with no fixed effects. These results
are reported in Table 4-4b and variance component estimates
illustrated in Figure 4-6b. As reported in Section 5.3, one pair
of side-by-side soil samples differed significantly (unit 19).
Because of the effect of this pair of data values on the
replicate standard deviation, modeling results for entryway soil
samples are also reported with this pair of values (outliers)
eliminated.
The geometric average lead concentrations for the different
sample types in decreasing order are:
72
-------
Window channel dust (1448 /xg/g)
Air duct dust (749 ^g/g)
Window stool dust (724 /xg/g)
Entryway dust (314 /xg/g)
Floor dust (255 /xg/g)
Foundation soil (217 /xg/g)
Entryway soil (196 /xg/g)
Bed/rug/upholstery dust (174 /xg/g)
Boundary soil (121 /xg/g) .
The six dust sample types are in the same exact order as for lead
loadings. Note also that the soil lead concentrations are lower
than all the dust lead concentrations except for the
bed/rug/upholstery sample type.
The smallest total standard deviation was observed for air
ducts (0.53) and the largest was for window stools (1.53) . The
remainder of the total standard deviations were fairly consistent
from a low value of 0.84 (bed/rug/upholstery) to a high of 1.08
(vacuum window channel). Note that the variability in lead
concentrations is substantially lower than the variability in
lead loadings. This is logical since the variability in lead
loadings includes variability due to both lead concentrations and
dust loadings.
For the vacuum dust sample types, the unit-to-unit variance
component is statistically significant for entryways, floors, and
window stools. The room-to-room variance component is also
marginally significant for floors and window stools. For all
three soil sample types, both the unit-to-unit variance component
and the side-to-side variance component were at least marginally
significant. With the exception of air duct and vacuum window
channel samples, the unit-to-unit variance component was a
73
-------
substantial contributor to total variability for both dust and
soil samples.
Dust Loading
The following is a summary of the modeling results for dust
loading for the model with no fixed effects. These results are
reported in Table 4-4c and variance component estimates
illustrated in Figure 4-6c.
The geometric average dust loadings for the different vacuum
sample types in decreasing order are:
Window channel (863 mg/ft2)
Air duct (411 mg/ft2)
Entryway (72 mg/ft2)
Floor (52 mg/ft2)
Bed/rug/upholstery (49 mg/ft2)
Window stool (47 mg/ft2) .
The average dust loading values fall in exactly the same order as
for lead loadings and concentrations, with one major exception.
Window stools have dropped from third to last place in the list.
These results lead to two conclusions concerning lead loadings:
The higher lead loadings for window stools relative to
floors can be attributed primarily to higher lead
concentrations in the stool dust and not to higher dust
loadings.
With window stools as the exception, differences in dust
lead loadings among different sample types can be
attributed to differences in both dust lead
concentration and dust loading on the surface being
sampled; dust lead concentration and dust lead loading
are positively correlated from sample type to sample
type (see Table 4-3) .
74
-------
The smallest total standard deviation was observed for
window channels (1.02) and the largest was for air ducts (1.46) .
The four other total standard deviations varied throughout this
range. Note again that the variability in dust loadings is
substantially lower than the variability in lead loadings.
Again, this is logical since the variability in lead loadings
includes variability due to both lead concentrations and dust
loadings.
Floor samples had the only statistically significant
variance components. Both the unit-to-unit and room-to-room
variance components were observed to be significant. For air
duct, entryway, and floor samples, the unit-to-unit variance
component is a substantial contributor to total variability. For
bed/rug/upholstery, window stool, and window channel samples, the
unit-to-unit variance component is only a minor contributor to
total variability.
4.3.2 Estimates of Renovation Effects, Abatement Effects, and
Variance Components
A second statistical model was fitted to the data for each
of the sample types. The second model is exactly like the model
fitted in Section 4.3.1, except that fixed-effect terms
representing renovation and abatement effects have been added to
the model. These terms attempt to explain portions of the unit-
to-unit and room-to-room variability. Due to the limited number
of units and the importance of the renovation effect, it is not
possible to include fixed-effect terms for type of abatement or
amount of abatement without reducing the degrees of freedom for
unit-to-unit variability to an unreasonably low value. Estimates
of the geometric mean, estimated fixed effects for abatement and
renovation, and variance component estimates are reported in
Table 4-5a for lead loading, in Table 4-5b for lead
concentration, and in Table 4-5c for dust loading.
Rather than representing an overall mean for all units, the
geometric mean now represents the expected value of the dependent
75
-------
(0
ID
0)
(0
H
«
c
o
.2
Q
-o
to
c
re
w
to
UJ
o
Tl
C
to
OC
to
O
tu
TJ
tu
X
u.
%
H
= £
Q. re
° W
OC
O) f
= .0
E o
to o
E
o
o
oc
'c
o
1
r- tU
E £
§ 1
DC re
JD
C
0) |£
3
I «
-O
*"
c
*J
o
c
tu
oc
tJ fc_ CU
mil
S * 1 o i
O c
tu
a. tu
CO (/)
tu
c
o
~
E
0
O
"a. tu
CO f
0)
*"" m
^ ~
in CM o
i- <- O
"
in CM
a> -
i- CO
O) CO tO
* o rx
o CM o
i~ CO CO
q CM ^
d in d
0) <0
U3 ,_'
O
£
3
O
,±
<
E
3
3
U
re
>
(O
in CD
i-^ ~
O (O CD
O t- O
"
O O
co en
t- 00
|x o ^
in CD in
i- O O
on * CM
§ ci d
CM CM
(O m
to d
CM
m
5
t
C
UJ
E
3
U
re
>
^ *~
CO
0) CO 00
CO *t «-
0^0
^ 00
to ^- o
d ~ d
CO in ^
»t in CM
O O O
-1
8?
d fj
co o in
«- en o
o o o
co 57 tfl
on en o
O) O O
O 0 T-
O
ix O O
co d
O5
CO
o
o
"-
E
3
O
to
>
CO CM
6 ~
in en co
CO CO CO
o o o
(O **" r-
iO . O
v ^ O
^^
O) oo r**
(O CM CM
o o o
~*-*
ȣ
T- ^
^- O CM
in r- o
t- T* ID
«-; CO CM
uJ ^ d
O in
p^ en
id o
in
CM
g
o
c I
i? *~*
> w
E
3
0
to
>
in
CO ,-
d ~
(O «- O>
O 00 r-
«- o o
rx CM" CM
CO 00 ^J"
i- O O
§5? oo
oo en
O CM O
"**
CM R
O 0
« ^-
8) ? in
in to TJ-
O o o
o> *
CM «
d *
CO 5?
O
E
3
3
O
to
>
CO
CO (O
d ~
00 tB -^
* en o
o «- o
^^
a) Co
in r-
0 -
co r^ r^
m ^- o
CO O O
«t IX ^
m ">. 9
03 O O
CO ^
to *t
** 2
CM
0
O
"-
tu
Q.
*
^ CO
d ~
O) O) 00
IX CO O
o-o
O fx in
O oo to
o o o
^- P en
en «- m
O CM O
fx 00
O 00
*" 52
o ?
^- en *
o o o
CM o en
od o o
CM ^ 0
O C/3
CU
D.
^
value is logarithmic standard error of estimate, and bottom value (when present]
tu
TJ
TJ
'E
tn
to
E
CO
tu
tu
'^
o
3
E
£r
C
to
tu
eometric m
O)
to
tu
>
Q.
0
1
tu
J5
tu
o
c
to
o
'c
O)
'to
TJ
0)
*-
to
.0.
0
alue is estimated degrees of freedom for estimating the random effect standard
^
_aj
o
E
c
o
.2
tu
TJ
o
to
TJ
c.
tn
a
c
^
,1±
re
OJ
stimated lo
tu
to
tu
"to
>
a
0
t-
ance level; the last standard deviation estimate (on the right) cannot be tested fo
o
'c
g>
U)
TJ
I
tu
to
-Q
O
.£2
c
tu
CO
tu
Q.
C
tu
JZ
z
tu
3
bottom val
TJ
C
re
c
g
re
5
tu
TJ
tu
J£
ficance
'c
.5"
'to
tu
>
tu
0
0
c
to
JZ
tu
"id, therefoi
m
icance ;
**-
'c
O)
'to
s effect was negative and therefore a significance level based on this test cannot
^
~
"5
ficance
'c
DJ
'to
f
*J
to
0
o
'JJ3
.55
re
to
uL
1
c
CO
enomin
TJ
tp
r-
t-
to
3
_O
re
o
76
-------
w u
in
0)
rH
8
t
B
Q
"S
ra
TJ
Fl
CO
UJ
TJ
C
ra
QL
*
«
o
UJ
TJ
CO
£ CO
ra ^
c
& <°
5? c
= .0
ro o
co -i
r- CO
E TJ
gco
'E
D
2
P
Room
Abatement
t:
CO CO
« E
2 co
1 5
c
o
ra
CO
o:
TJ
itfli
o>
Q. CO
if CO
CO
'c
0
o.
E
o
o
l&
t >>
ra (-
CO
coG?
0 >"
o CM" oo
O CM CO
O t- O
*~-^
10 at
0 C£_
SPh-
CO CO
0 0 O
T- ^- CO
10 co co
O T- o
10 CO
l*~ CO
oo ci
0
t3
Q
?
E
8
>
0 ^
co co
d ^
o co° o>
O T}- CO
o co o
*"'
Ij 0
O CO.
IO P T-
CN t- O
ri o_ci
10 f- 0
00 CN O
^ 0 O
co T
o> o^
CN
^
ra
%
c
m
E
3
>
in ^~
CM »-
o ^
co ^ co
c*> co o
0 * 0
10 ?co
^ m o
0 JO O
« in CM
0 O O
*"-"'
of
d jj
n tn *
h- CD CO
° 2.O
CO P" 00
co r^ r-
CN 0 O
co o in
oo in o
* 0 O
§S
*" S
cn
CO
o
o
LJ_
E
o
>
cn
*! CM
o ^^
3O CO
OO CN
0 T- 0
CN ST CD
O CO O
*" t, °
CO CO *
o in o
T- T- O
*~ '
CO S"
10 cn
*~ 5S,
co o" o>
O T- CM
W T- O
CN in *-
^ CO OO
T- »- O
10 S^
N 2.
m
CM
3
Q
"D o
§ W
E
o
>
^*,
^ ^~
es ^
o co co
O CM CO
o o o
co in" oi
O CO T-
T- CM O
u) co" m
CO O <»
o o o
* '
5?
^
10 CM" co
cn o> o>
d ci ci
CO O) CO
O 0 O
0 CO*
U) to
CN 0
-
$ CO
o c
TJ c:
is
E
3
re
>
IO
0 CO
0 ^
T- CO 0
CO CD O
o in o
IO ?T-
CD CO »-
O T- 0
x '
oo r-
0 S
T- o~cn
^ t^ CM
CM 0 O
^- in *
T- 00 *
CN 0 O
CO r-
fO .
IO O,
m
>%
ra
TJ
c
m
o
T- _^
ci "
10 CM" i~-
* O CM
O ^0
0 ^~CO
o T- m
e o o
*^^
co *":
d j;
«- ?CN
t^ CO O
* O 0
cs m co
o> -^
ra
3
C*
LU
.
Q
CO
|^
T m
o ' *
5a> i-
o o
O Q O
T- P ^r
00 00 O
0 *- 0
o> P"
o> ^
0 s
1- S" CD
cn oo TC
r- 0 0
ro 05 ^
CM O 0
l|
t
c
.0
TO
TJ
C
£
o
CO
, , -
r- « 1
77
-------
O
in
i
0)
<-\
3
H
*
c
e
<0
&
p
ra
T3
c
*-*
fi
UJ
E
o
T3
C
«
«
t3
ff
UJ
-£,
0)
.X
LL
f
" F
V *
0) c
f=8
ra o
W -1
E
*
«
C
D
-*-J
|2
E I
si
o; oj
ti
<" E
o
<
c
.2
2
o
c
0)
£
o
O ,_ 0)
iflis
o>
Q. 0>
C N
CO W
W
C
c
0
Q.
E
o
O
II
ra h-
W
o ^^.
O to
T~
JO CM O
T- *- O
IO P"
1^ 1^
O *~ O
?
O CO
*~ S2.
00 CM CM
^ CD CO
000
co r** ^*
CO N. O
00 O O
s|
CM
ra
!>
LU
|
1
>
to *-
o C-
o S *
O CO CO
O 0 O
O T- CO
^ co o
o r^o
o »- co
h- ^- o
0^0
o ?
o o>
*~ t
OO CO CO
T- O T-
° c. °
N- O5 ^~
^ *~ co
O T O
«§§
5 0,0
"I
o>
CO
o
o
LL
E
^
3
O
5
N CM
O ^^
IO CD T-
O CM^O
U> CM CO
h- in T-
o co.o
O 00 00
O CM 1^
O CO O
y
00
o °2
IO O CO
M CO CO
i- 0 O
N 1^ CM
CO CO O
^ 0 O
CM °^
m
CM
"o
0
OT
S
o
c
1
1
$
oo *
o *~ '
T-
CM en o
^ 00
n co co
co o in
o o_o
o t^ in
o en r--
o o o
CO O
o m
t--
-------
variable for unrenovated control units. As in Table 4-4, the
estimated geometric mean is reported as the top value in the
fourth column with the logarithmic standard error reported in
parentheses below.
For renovation, house abatement, and room abatement effects,
estimated effects are reported as the top value in the fifth
through seventh columns of Table 4-5. The estimate is an
estimate of the multiplicative effect of the presence of that
condition. For example, to determine an estimate of the
geometric average lead concentration (Table 4-5a) for vacuum
window stool samples in abated, unrenovated houses, multiply the
geometric mean for unrenovated control houses by the estimate for
house abatement :
6.70 * 5.47 = 36.56
Below each of these estimates, the logarithmic standard error of
the estimate is reported in parentheses. The bottom value
reported in these columns is the observed significance level of
the test that the true multiplicative effect is equal to one
(i.e., no multiplicative effect) versus the alternative that the
multiplicative factor is not equal to one. The estimated
geometric means from the mixed model analysis are presented in
Figure 4-7a for lead loading, lead concentration, and dust
loading. The estimated multiplicative effects are illustrated
graphically in Figure 4-7b for lead loading, in Figure 4-7d for
lead concentration, and in Figure 4-7f for dust loading. Effects
with an observed significance level of 0.05 or less are marked
with an asterisk (*) .
The last four columns of Table 4-5 provide estimates of the
various variance components after controlling for the fixed
effects listed previously for that sample type. The structure is
the same for these as it was in Table 4-4. Notice that the
degrees of freedom for the unit standard deviation are smaller
than in Table 4-4. The variance component estimates are
79
-------
Y///7///77/7777///////.
V////////////////777.
r/////7///77/777/77.
V/////////////////////////////////77,
\
Y////////////7///77///7///.
\
Y/////////////////////.
o
\/////////////////////< A.
Y///////////////////////////////.
a
o>
M
g.
-H
80
-------
i
V//////////////////A
W//////////////ZV//.
V///////7///77///////'.
\
\
t
0)
n>
8
fi
O
H
4J
0) MH
M\
cn
«H 3.
o «-
a cn
-u 0
o -H
a) -a
(M (0
(M O
01 H
a) -a
> at
H 0)
i> H
at
o
H
0)
H 0)
4J X
a -H
"f
ID
I
H
81
-------
I
a
I
3
0
Variance Component:
Unit
Room
Sampling Location
Replicate
Os
£
*
Sample Type
Figure 4-7c. Variance component estimates from mixed model
ANOVA: lead loading (/*g/ft2) .
82
-------
J
1
1
J
J
K
r
u
i
i
11
J
II
12 I
V//s
\
* v//////,
d
Y///J
I
V/////,
]~_
Y///A
* |
* V////,
* 1
8 S
9
y////////////////////.
y////////////////////.
y////////////////////.
y////////////////////.
y////////////////////.
^^^^^M
y////////////////////.
y////////////////////.
v////////////////////.
i_
§ 3 5
4 0 C
1
0
T:
£
^
^
^i«.
"<<
A>.
^
^
^
^
+
\
^
3
3
In
*
4
*
tt.
\
>
U (U
0) U
IM fi
4-I O
fl) o
fl) Tl
> id
H ID
i> H
n<
O
-H
12
^
0)
TJ TJ
0) O
J_J ri
(2 T-l
Figure 4-7d. Estii
mixec
83
-------
8
11
Variance Component:
Unit
Room or Side
Sampling Location
Replicate
i
i
I
Sample Type
Figure 4-7e. Variance component estimates from mixed model
ANOVA: lead concentration (jig/g) .
84
-------
O
O
d
9
o
a
o
o
4J
G
0)
4J
«
A
(0
a
§
H
c u
ra
-H 3
4J -d
n)
u
H
H <
£§
0)
T) -O
0) O
H d)
U X
[Q -H
H 6
14-1
0)
M
3
Oi
-H
85
-------
a
Variance Component:
[=1 Unit
Room
Sampling Location
Replicate
t 9 t t
Sample
Figure 4-7g. Variance component estimates from mixed model
ANOVA: dust loading (mg/ft2) .
86
-------
illustrated graphically in Figure 4-7c for lead loading, in
Figure 4-7e for lead concentration, and in Figure 4-7g for dust
loading. Comparing the estimates of variance in Figure 4-7 to
the estimates in Figure 4-6 gives an indication of the amount of
variability explained by the fixed effects. Due to the small
amount of data, the degrees of freedom for estimating the random
effects are small. Therefore, conclusions about the significance
of these effects should be made with caution. Also, in general,
variance components are expected to decrease when fixed-effect
terms are added to the model. However, because of the small
number of degrees of freedom, some variance components (e.g.,
lead loading variance for air duct samples) may increase when
fixed effects are added.
The room abatement effect was significant only for floor
lead loading. In fitting lead loading and lead concentration to
various models for components other than floors, the room
abatement effect was never even marginally significant (the
significance level was never below 0.20). Therefore, room
abatement is only included in models for the floor samples.
Each fixed effect was tested for significance when added
last among the fixed effects in the model, but before all the
random effects in the model. The denominator mean square used in
each test was the proper linear combination of the estimated
variance components as determined by expected mean square
equations.
Lead Loadings
The following is a summary of the modeling results for lead
loading for the model with fixed effects included. These results
are reported in Table 4-5a. The geometric means are illustrated
in Figure 4-7a, estimates of the fixed effects of renovation and
abatement are illustrated in Figure 4-7b, and the variance
component estimates are illustrated in Figure 4-7c.
87
-------
The geometric average lead loadings expected in unrenovated
control houses for the different sample types in decreasing order
are :
Window channel (vacuum 2873 /xg/ft2)
Air duct (649 pig/ft2)
Window stool (vacuum 6.70 jug/ft2, wipe 100 pig/ft2)
Floor (vacuum 3.76 //g/ft2, wipe 7.63 /^g/ft2)
Entryway (6.62 /ig/ft2) .
The floor samples had the lowest two total standard
deviations (0.99 for vacuum, 0.59 for wipe). Air ducts (1.95)
and vacuum window stool (1.86) samples were the highest.
The unit-to-unit variance component is statistically
significant (at the 0.05 level) only for floor wipe samples.
This variance component is marginally significant for air duct
samples. The estimated variance component is negative for
entryway and window stool samples. With the exception of these
two sample types, the unit-to-unit variance component is a
substantial contributor to total variability even after
controlling for the fixed effects. Of those sample types for
which the room-to-room variance component could be tested for
significance, it is significant only for vacuum window stool
samples.
The renovation effect was only statistically significant in
explaining the responses for entryway samples, and both vacuum
and wipe floor samples. For all sample types except air ducts
and vacuum window channels, the estimated effect of renovation
was to increase lead loadings. The effect was strongest for both
vacuum and wipe floor samples.
In general, abatement history of a house was found to be
less significant than renovation for lead loading. For no
component was this effect strongly significant. In the cases of
88
-------
vacuum and wipe floor samples, a marginal significance was
observed. For all sample types except air ducts, vacuum window
channels, and wipe window stools, houses which have been abated
in the past have higher lead loadings.
The effect of room abatement was found to be significant
only for floor vacuum samples. In abated houses, abated rooms
were observed to have lower floor lead loadings than unabated
rooms.
Lead Concentrations
The following is a summary of the modeling results for lead
concentration for the model with fixed effects included. These
results are reported in Table 4-5b. The estimates of the fixed
effects and variance components are illustrated in Figures 4-7d,e
respectively.
The geometric average lead concentrations estimated for
unrenovated control houses for the different sample types in
decreasing order are:
Window channel dust (2150 pcg/g)
Air duct dust (875 /Kj/g)
Window stool dust (245 /xg/g)
Foundation soil (109 /-ig/g)
Floor dust (106 /xg/g)
Entryway dust (96.3 jug/g)
Entryway soil (65 /xg/g)
Boundary soil (54 /-ig/g) .
The three dust sample types with the highest concentrations are
in the same exact order as for lead loadings. Note that none of
the soil lead concentrations are very high, but foundation soil
levels are close to the floor and entryway dust levels.
89
-------
The smallest total standard deviation was observed for
entryways (0.47) and the largest was for window stools (1.56).
The unit-to-unit variance component is statistically
significant for window stool dust samples and foundation soil
samples. The room-to-room (side-to-side) variance component is
significant for boundary and foundation soil samples. The unit-
to-unit variance component is a substantial contributor to total
variability for floor, window stool, boundary, and foundation
sample types.
For concentrations, the renovation effect was only
statistically significant in explaining the data for entryway
samples and floor samples. As was the case for lead loadings,
for all sample types except air ducts and vacuum window channels,
the estimated effect of renovation was to increase lead
concentrations. Also consistent with the results for lead
loadings, the effect was seen to be strongest in floor samples.
House abatement was only found to be significant in both
types of entryway samples (vacuum and soil). As for renovation,
for all sample types except air ducts and vacuum window channels,
houses which have been abated in the past have higher lead
concentrations. The component with the strongest estimated
abatement effect was soil entryway samples.
Again, a room abatement effect was only included in the
model for lead concentrations on floors. However, it was not
observed as significant. The estimated room abatement effect of
0.73 indicates that in abated houses, abated rooms were observed
to have slightly lower floor lead concentrations than unabated
rooms.
Dust Loading
The following is a summary of the modeling results for dust
loading for the model with fixed effects included. These results
are reported in Table 4-5c. The estimates of fixed effects and
variance components are illustrated in Figures 4-7f,g
respectively.
90
-------
The geometric average dust loadings expected in unrenovated
control houses for the different vacuum sample types in
decreasing order are:
Window channel (1336 mg/ft2)
Air duct (742 mg/ft2)
Entryway (69 mg/ft2)
Floor (36 mg/ft2)
Window stool (27 mg/ft2) .
The average dust loading values fall in exactly the same order as
for the uncorrected geometric means.
The smallest total standard deviation for dust loadings was
found for window stools (0.88) and the highest for air ducts
(1.65) . The remaining four sample types had very consistent
total variation (1.09, 1.09, 1.06).
Floor samples had the only statistically significant
variance components. The unit-to-unit variance component was
marginally significant (p=0.06), and the room-to-room variance
was significant (p=0.03). For air duct and floor samples, the
unit-to-unit variance component is a substantial contributor to
total variability. For entryway, window stool, and window
channel samples, the unit-to-unit variance component is only a
minor contributor to total variability.
The renovation effect was only statistically significant in
explaining the variability in dust loadings for vacuum window
stool samples. For all sample types except air ducts and vacuum
window channels, the estimated effect of renovation was to
increase dust loadings. The effect was strongest for vacuum
floor samples.
Abatement was not found to be significant for any of the
components. The strongest estimated effect was for floors. For
91
-------
all sample types except air ducts and vacuum window channels,
houses with an abatement history have higher dust loadings.
The effect of room abatement was not found to be
statistically significant for floor samples. However, in abated
houses, unabated rooms were observed to have approximately five
times higher floor dust loadings than abated rooms.
4.4 MODELING RESULTS FOR DUST SAMPLES BY SAMPLE TYPE
In Section 4.3, results for dust samples were reported
separately for two different statistical models, three different
measured values (lead loading, lead concentration, and dust
loading), and in some cases two different sampling methods
(vacuum and wipe). In this section, results are reported by the
following sample types:
Air duct samples
Bed/Rug/Upholstery samples
Interior entryway samples
Floor samples
« Window stool samples
Window channel samples.
An attempt is made to draw global conclusions that span the two
different statistical models and the different measurement types
for each sample type.
4.4.1 Air Duct Samples
There were only 10 air duct samples collected in the Pilot
Study and used in the statistical analyses; their geometric mean
lead loading was 308 /xg/ft2. An estimate of the corresponding
mean in unrenovated control houses was 649 ^g/ft2. The geometric
mean lead concentration was 749 /-tg/g. For unrenovated control
houses, the estimate was 875 /xg/g.
The variation in lead loadings (standard deviation 1.72) was
equally due to unit-to-unit and room-to-room differences.
92
-------
However, for lead concentrations, the differences were virtually
all due to room-to-room differences.
Air duct and window channel (vacuum) samples were the only
sample types in the Pilot Study for which renovation and
abatement were estimated to reduce both lead loadings and
concentrations. For air duct lead loading, the estimated
multiplicative effect of renovation was 0.01; the multiplicative
effect of abatement was 0.49. For concentrations, the effect was
0.51 for renovation, and 0.84 for abatement. Neither of these
effects was observed as statistically significant, and neither
reduced the substantial unit-to-unit variation in lead loading.
4.4.2 Bed/Rug/Upholstery Samples
Only eight bed/rug/upholstery samples were collected in the
Pilot Study, allowing only a limited statistical analysis. Unit
51, which was under full renovation, had none of these items
present to sample. The geometric mean lead loading was the
lowest of all sample types (8 /xg/ft2} , and the geometric mean
concentration was second lowest (174 /xg/g) only larger than
boundary soil samples).
The variation seen in the lead loadings and concentrations
of these samples was not due to differences between units; it was
primarily due to within-unit differences. Due to the small
amount of data, tests for renovation and abatement effects were
not attempted.
4.4.3 Interior Entryway Samples
There were 12 entryway vacuum samples collected in the Pilot
Study, front and back entryway samples from each of the six
units. The geometric mean lead loading for these samples (23
jug/ft2) was almost twice as high as the mean for other floor lead
loadings (13 /xg/ft2) . This is primarily due to the presence of
more dust, but also partially due to a higher concentration of
lead in the dust (314 /-ig/g) as compared to the corresponding
results for other floor samples (255 /xg/g) .
93
-------
Both abatement (p=0.01) and renovation (p=0.00) were
statistically significant in explaining the variation for
concentrations. As compared with unrenovated control houses,
lead concentrations were about 3 times higher in abated houses,
and about 5 times higher in renovated houses. For lead loadings,
only renovation was observed as significant here, but the
loadings for renovated control houses were more than 40 times
greater than those for unrenovated control houses.
4.4.4 Floor Samples
For floors, both vacuum and wipe samples were collected in
the Pilot Study. Vacuum sample modeling results are presented
first for lead loading, lead concentration, and dust loading.
Wipe sample modeling results are then presented for lead loading,
and compared to the vacuum sample modeling results.
Floor Vacuum Samples
The 39 floor vacuum samples collected in the Pilot Study
were by far the largest number collected for any sample type.
The overall geometric mean lead loading of these samples was 13
/xg/ft2. After controlling for renovation and abatement effects,
the estimate of this mean for unrenovated control houses was 3.8
^tg/ft2. The overall geometric mean for lead concentrations was
255 /ig/g. After controlling for renovation and abatement
effects, this figure was 106 ptg/g.
Most of the variability in lead loadings and lead
concentrations on vacuum samples from floors was due to unit-to-
unit differences (Table 4-4). Replicate-to-replicate variation
was observed near the same magnitude as sampling location-to-
sampling location variation. For both lead loadings and lead
concentrations, most of the unit level differences were explained
by the renovation and abatement factors. For lead loadings,
houses under renovation had lead loadings 70 times higher than
unrenovated control houses. This large lead loading is due both
to an increased concentration of lead in the dust (4.89 times
94
-------
higher), and a larger amount of dust on the floors in houses
under renovation (14.31 times larger).
Lead loadings were also found 10 times higher in abated
homes than in unrenovated control houses. The lead
concentrations were about 2.9 times higher, and the dust loadings
about 3.5 times higher.
Floor lead loadings taken by the vacuum method were the only
measurements in the Pilot Study for which room abatement history
was found to be statistically significant. Unabated rooms in
abated houses have about 8 times higher lead loadings than abated
rooms in abated houses, (i.e., the lead loadings in abated rooms
are about 13% of those in unabated rooms in abated houses). This
would suggest either these unabated rooms were contaminated by
dust prior to or during abatement and never completely cleaned,
or that there may be residual lead-based paint in these unabated
rooms.
The same phenomenon was evident in the lead concentrations
and dust loadings for vacuum samples, but it was not as
pronounced, and therefore the room abatement effect was not
observed as statistically significant.
Floor Wipe Samples
There were a total of 12 floor wipe samples collected in the
Pilot Study, two side-by-side samples from each unit. The
geometric mean of these samples was 51 /xg/ft2. After controlling
for renovation and abatement effects, the mean for unrenovated
control houses was 7.6 /zg/ft2.
The estimate of unit-to-unit standard deviation is actually
an estimate of the combined variation of unit-to-unit, room-to-
room, and sampling location-to-sampling location standard
deviation. As expected, this combined variation far exceeded the
replicate-to-replicate (side-by-side) variation. This variation
was mostly explained by renovation and abatement effects, but
even after controlling for these factors, the unit-to-unit
differences were still statistically significant. The results
95
-------
for floor wipe samples are similar to the results for floor
vacuum samples. The effect of renovation on lead loadings was
estimated as virtually the same under both methods (i.e., a 70-
fold difference). The multiplicative effect of house abatement
was higher for vacuum samples, but both were positive (9.9
compared to 3.5).
The variance component estimates for lead loadings from the
vacuum and wipe methods are also similar. The combined unit-to-
unit, room-to-room, and sampling location-to-sampling location
standard deviation was estimated (by pooling the three individual
standard deviations in Table 4-4) at 1.89 /zg/ft2 for vacuum
samples, compared to 1.92 for wipe samples. The replicate-to-
replicate standard deviation was 0.47 /xg/ft2 for vacuum samples,
and 0.33 /xg/ft2 for wipe samples.
Although this section provides some comparison of the vacuum
and wipe sampling results, a further detailed comparison of the
wipe and vacuum methods based on paired data collected in the
"bridge" rooms is given in Section 4.7.
4.4.5 Window Stool Samples
For window stools, there were also both vacuum and wipe
samples collected in the Pilot Study. As for floor samples in
the previous section, vacuum sample modeling results are
presented first for lead loading, lead concentration, and dust
loading. Wipe sample modeling results are then presented for
lead loading, and compared to the vacuum sample modeling results.
Window Stool Vacuum Samples
There were 25 window stool vacuum samples collected in the
Pilot Study. The geometric mean lead loading for these samples
(34 /xg/ft2) was relatively low, but the geometric mean lead
concentration (724 /zg/g) was among the highest observed, exceeded
only by window channels and air ducts. In addition the largest
loading observed in the entire study was found by vacuuming a
window stool (13087 /^g/ft2 found in unit 80) .
96
-------
Window stool vacuum samples were observed to have the
largest total variation of any of the sample types observed, for
both lead loadings and lead concentrations. This variation was
mainly due to room-to-room differences, and unit-to-unit
variation. Window-to-window within room differences and
replicate-to-replicate differences were small by comparison.
These variations could not be explained by renovation or
abatement effects (i.e., neither of these effects was found to be
statistically significant). However, on average, in houses under
renovation, lead loadings were 6.1 times higher, and dust
loadings were 4.3 times higher, than in unrenovated control
houses. In abated houses, lead loadings were 5.5 times higher
and lead concentrations were 4.1 times higher than in unrenovated
control houses. Thus, one might conjecture that higher lead
loadings in renovated houses were mainly due to larger amounts of
dust, while higher lead loadings in abated houses are possibly
due to higher concentrations of lead in the dust.
Window Stool Wipe Samples
There were a total of 12 window stool wipe samples collected
in the Pilot Study. The overall geometric mean of these samples
was 144 pig/ft2; while the mean in unrenovated control houses was
estimated at 100 /xg/f t2. The unit-to-unit differences were seen
to be the primary source of variation in these samples. However,
there was a marginally significant window-to-window within room
variation (p=0.08) observed. For the wipe method, the estimated
effect of renovation on window stool samples was to increase lead
loadings by a factor of 28. On average, abated houses had window
stool wipe lead loadings of less than half (0.40) of those in
unabated houses. However, neither of these effects was found to
be statistically significant. The lead loading window stool
results for wipe samples differ in several major ways from the
window stool results for vacuum samples. The main difference
between the two methods was in the observed effect of abatement.
By the wipe method, abated houses had a lower than average lead
97
-------
loading, while by the vacuum method, abated houses had 5.5 times
higher lead loadings than unabated homes.
A second qualitative difference seen in the results by the
two sampling methods was in the room-to-room variation estimates
(Table 4-4). By the wipe method, the room-to-room differences
were negligible, but by the vacuum method, room-to-room
differences were determined to be the primary source of variation
in the lead loadings. This difference may be due in part to the
low number of degrees of freedom available for estimating the
room-to-room variation for wipe samples.
Another difference observed between the two methods was in
the average lead loading observed. Comparing the results
discussed above and before controlling for the fixed effects, the
wipe method had a geometric mean 4.23 times larger than the
vacuum method. This difference is mostly due to a general
multiplicative bias factor of approximately 5 to 10 between the
two methods (see Section 4.7). After controlling for the fixed
effects, the results are even less comparable; the estimated
loading in unrenovated control houses is 15 times higher by the
wipe method (100 /xg/ft2) than by the vacuum method (6.7 p.g/ft2) .
4.4.6 Window Channel Samples
Similar to the case of floors and window stools, both vacuum
and wipe samples were collected from window channels in the Pilot
Study. As in these previous cases, vacuum sample modeling
results are presented first for lead loading, lead concentration,
and dust loading. Wipe sample modeling results are then
presented for lead loading, and compared to the vacuum sample
modeling results.
Window Channel Vacuum Samples
There were only 11 window channel vacuum samples collected
in the Pilot Study. The geometric mean lead loading (1250
/ig/ft2) and lead concentration (1448 p.<3/g) were highest for these
samples among all sample types taken. Oddly, the largest average
98
-------
loadings and concentrations were found in the unrenovated control
house. This sample type and air ducts were the only ones for
which renovation and abatement were estimated to reduce lead
loading and concentration. The average dust loadings on window
channels were also lower in abated homes and in renovated homes.
There was little unit-to-unit variation observed for window
channel vacuum samples (Table 4-4); this was consistent across
all three measurements (lead loading, lead concentration, and
dust loading). Variation was primarily attributed to room-to-
room differences, however, there was also a substantial
difference in lead loadings seen between windows within rooms.
Neither abatement nor renovation were observed to be significant
factors for these samples. This is not surprising, since these
factors are unit-level explanatory variables, and there were only
small differences observed between units.
Window Channel Wipe Samples
There were only six window channel wipe samples collected
from a total of three units. The geometric mean lead loading of
these samples was 801 jug/ft2, exceeded only by the mean lead
loading on window channels taken by the vacuum method. This
sample type had the smallest estimated total variability of all
lead loading measurements. There were not enough data available
for wipe window channel samples to fit a mixed model analysis of
variance. Thus, no results for renovation and abatement effects
are presented.
The estimate of sampling location-to-sampling location
(window-to-window) variation was observed as statistically
significant compared with the replicate-to-replicate variability,
which was estimated as the smallest among lead loadings for all
sample types. Both the vacuum and wipe sampling methods on
window channels produced the highest estimates of geometric mean
lead loading, and the lowest estimates of total variation among
lead loadings. Aside from the fact that the lead loadings were
99
-------
larger for the vacuum method than for the wipe method,
qualitatively, the results by the two methods were similar.
4.5 MODELING RESULTS FOR SOIL SAMPLES BY SAMPLE TYPE
In Section 4.3, results for soil samples were reported
separately for two different statistical models. In this
section, results are reported by the following sample types
Boundary samples
Exterior entryway samples
Foundation samples.
An attempt is made to draw global conclusions that span the two
different statistical models for each sample type.
4.5.1 Boundary Soil Samples
There were a total of 15 boundary soil samples collected in
the Pilot Study. The geometric mean lead concentration was 121
/xg/g. For unrenovated control houses, the mean was about half as
large (54 jug/g) .
The results of fitting the statistical model equation (2) to
these 15 samples is shown in Table 4-4b. The unit-to-unit
standard deviation (0.69) was about as large as the side-to-side
standard deviation (0.61). Both were statistically significant
(unit-to-unit was marginal). On average, houses under renovation
were estimated to have lead concentrations about twice as high as
others, and houses where abatement was performed had average
concentrations about 2.4 times higher. Neither of these factors
was seen as statistically significant.
4.5.2 Exterior Entryway Soil Samples
There were a total of 16 entryway soil samples collected in
the Pilot Study. The replicate-to-replicate variance in lead
concentrations was large in comparison with the other two soil
sample types (100 times larger than for boundary samples and 9
100
-------
times larger than for foundation samples). Therefore, the data
were examined to look for any gross inconsistencies.
For Unit 19, in the front yard, there were two side-by-side
soil samples taken near the entryway. The measured
concentrations here were 196.53 //g/g and 49.69 /ig/g. This was by
far the largest observed difference (on a log scale) between
side-by-side samples found for any of the soil sample types.
Computing the variance components for entryways without these two
samples gave an estimated replicate-to-replicate standard
deviation consistent with that for foundation soil samples (0.18
/xg/g compared with 0.17 /xg/g) . These samples are referred to as
outliers for lack of a better term, although because of the small
sample size, there is no proof that they will not be found
typical of soil lead concentrations in the full CAP Study.
The analysis of variance for entryway soil samples was
performed with and without these samples removed. The main
difference observed was that with the outliers removed, the side-
to-side variation was statistically significant, while with all
the data included, it was not. Since there was little difference
in the estimates from the mixed model, the results using all the
data are presented.
Using all the data, the geometric mean lead concentration
for entryway soil samples was 196 pig/g. The corresponding
estimate for unrenovated control homes was 65 //g/g. After
controlling for renovation, abatement was observed to have a
statistically significant (p=.02) effect on these lead levels;
the multiplicative effect of renovation was estimated at 1.9,
while abated houses had 4.7 times higher lead concentrations than
unrenovated control houses.
4.5.3 Foundation Soil Samples
There were 17 foundation soil samples collected in the Pilot
Study. The geometric mean lead concentration in these samples
was 217 /xg/g. For unrenovated control houses, this mean was
estimated to be 109 /xg/g.
101
-------
The total variation in these samples was similar to that
observed in the other soil samples. Most of this variation was
due to unit-to-unit and side-of-house differences. The variation
was not explained by renovation or abatement effects (i.e.,
neither of these factors was statistically significant.
Nonetheless, foundation soil lead concentrations were 2.4 times
higher in houses under renovation, and 2 times higher in abated
houses, as compared to unrenovated control houses.
4.5.4 Comparison of the Soil Sample Types
An analysis of variance was performed on the soil sample
types to determine whether there was a statistical difference in
the lead concentrations between boundary, entryway, and
foundation samples. Using all the soil data, there was
significant (p=.02) statistical evidence of a difference in the
results. Applying a multiple comparison test, there was a
significant difference between the boundary samples and each of
the other two soil sample types; however, the difference between
entryway and foundation soil sample results was not statistically
significant.
4.6 RELATIONSHIPS BETWEEN SAMPLE TYPES
In Sections 4.4 and 4.5, the pilot data have been summarized
by dust sample type and soil sample type, respectively.
Attention is now turned to relationships between the various
sample types. The primary methods employed to examine these
relationships are correlation matrices and scatterplot matrices.
The primary data employed to examine the relationships
between sample types are the geometric means by unit presented in
Table 4-2. Both the lead loading and lead concentration means
are examined.
102
-------
Lead Loading
The correlation matrix for lead loading unit means is
presented as Table 4-6a. To locate a correlation of interest,
locate the row corresponding to the first sample type and the
column corresponding to the second sample type. Correlation
information for the two sample types is presented in the
corresponding box. Within each box, the three values presented
are:
Top value: Correlation coefficient between the
logarithms of the geometric unit means
Middle value: Observed significance level of the test
of the hypothesis of no correlation (correlation
coefficient equal to zero)
Bottom value: Degrees of freedom associated with the
variance estimates used in calculating the correlation
coefficient.
Only the upper right-hand half of the matrix, above the shaded
diagonal, is filled in since the lower left-hand half of the
matrix would contain redundant information.
The lead loading unit means are presented graphically in
Figure 4-8a. This figure is a scatterplot matrix, or a
collection of bivariate plots organized into matrix form. As
with the correlation matrix, to locate a plot of interest,
identify the row associated with one sample type and the column
associated with the other sample type. The plot is presented in
the corresponding box. Within each box, the horizontal axis
represents increasing values of the column variable on a
logarithmic scale. Similarly, the vertical axis represents
increasing values of the row variable on a logarithmic scale.
The abbreviations employed on the diagonal to identify the
different sample types are defined in Table 3-5.
103
-------
Cn
a
-H
o
nf
O
d
ns
a)
itf
10
0)
H
A
a
CD
Q.
E
O
5 -5
0 C
T3 C
C CO
§6
0 0
| S
o
o
LL
5 13
0 C
-a c
.£ w
5 _
O o
-D 0
.E *^
0
o
LL.
CO
C
LU
D>
It
0)
CD
o
Q
r- C7) ±T
O 05 3.
o
«- co 5:
in co 2.
9
d
d
£25
o
* * *
i> r~- 3-
CM CD ^
0
O
Q
.^
<
E
CJ
CO
CM 00 ^
o
5 £ ?
o
co CD ±r
o o S
d
d
CO CD i
d
CM r~- S
0
Q.
QC
CD
CQ
^ *~ csT
in CD 21
o
22s
o
O) CN r?r
00 O S
O
O) CN
00 O S
d
d
m o S
d
CO
1
c
LU
CO § ^
o
* o £
o
to m ^
co o £
o
r*- co rPT
oo o £
d
d
6
o
LL.
Xgg
0
P> 10
(0 - £
o
in en rtr
r«. o £
O
d
"o
o
g
o
a
c
5
CO 03
^- CD 2i
9
O) in ~
O
O in
oo o £
9
5 B
0 C
T3 C
C CO
> O
CM 0
O
So£
d
0
o
LL
£
SS3
o
"o
0
w
0
T3
c
5
.".\ .::x/-
5 "CD
0 C
T3 C
c ro
> o
O
n
CD
CD
O)
CD
T3
E
o
C
o
.0
a
c
ro
C
O)
T3
0)
JD
O
a
T3
CD
O
O
C
o
o
o
T3
£
CO
a
o
104
-------
o>
I
m
tQ
-U
a
a>
M
a>
H
T3
TO
a
a)
6
4J
H
a
o
H .
M
-Ucs
0) 4J
0) Cn
Cn a.
o Cn
X-S
-U 0
nj H
4J m
o a>
H i-i
0)
-U CQ
4J 0)
W 4J
(C
00
I
0)
M
tn
H
105
-------
The ellipse plotted in each box of Figure 4-8a is
the ellipse that contains 95% of the probability associated
with the estimated bivariate normal distribution for the
plotted data. The narrower the ellipse, the stronger the
correlation between the two sample types. If the ellipse is
oriented from the lower left-hand corner of the box to the
upper right-hand corner of the box, the sample types are
positively correlated. If, on the other hand, the ellipse is
oriented from the upper left-hand corner of the box to the
lower right-hand corner of the box, the sample types are
negatively correlated.
Lead loadings for entryway dust were found to be
statistically significantly positively correlated with those
for both floor vacuum and floor wipe samples. Lead loadings
for window channel vacuum samples were found to be
significantly negatively correlated with each of these three
sample types. There was also a strong positive relationship
observed between the lead loadings of floor vacuum and wipe
samples.
It may be possible that correlation present in the
lead loading data, or conversely the lack of correlation, is
due to nonrandom factors such as renovation or abatement. For
example, if all units which were abated have high lead
loadings on both floors and window stools, and unabated
units have low levels for both of these sample types, then
floor loadings and window stool loadings will be highly
correlated, when there may be no correlation at all beyond
the effect of abatement history. To examine this
relationship, a correlation matrix and scatterplot matrix
were created for lead loadings after controlling for fixed
effects.
106
-------
Specifically for each sample type, the residuals from
the mixed model analysis of variance performed in Section 4.3.2
were averaged to produce average residuals for each unit.
These average unit residuals were used (in place of the
logarithms of the geometric unit means) in calculating the
correlation coefficients that are presented in Table 4-6b. The
average unit residuals were also plotted in scatterplot matrix
form in Figure 4-8b.
When controlling for the fixed effects, one must
realize that some degrees of freedom for estimation of
correlation are sacrificed to estimate the fixed effects.
This was accounted for in the significance levels and degrees
of freedom provided in Table 4-6b. Since only six houses were
sampled in the Pilot Study, and two house-level fixed effects
were found to be important, the reduction to 2 or 3 degrees of
freedom has a serious negative impact on the statistical power to
detect non-zero correlations in Table 4-6b. In particular, there
were insufficient data to test unit-to-unit correlations between
dust lead loadings collected on window channels and any other
sample type, after controlling for abatement and renovation
effects. This factor should not be a problem in the full CAP
Study.
After correcting for renovation and abatement affects,
none of the correlation estimates was observed to be significant.
However, there are several relationships worth noting. Whereas
lead loadings for entryway, floor vacuum, and floor wipe samples
were all found to be significantly positively correlated before
controlling for the fixed effects, they were all found to be
negatively correlated after correcting for the fixed effects.
This may suggest that the effects of renovation and abatement
override any house-to-house relationship between these sample
types.
107
-------
(0
E-i
ID
a.
5
E
o
3 a>
o c
§5
||
0
0
LL
3 O) £^?
O
g
Q
E
3
D
O
>
? 0) S
o
T* O> ^7"
o en CL
9
OJ 0 --*
Is ^ ^i-
o
gss
o
?§ S
o
§ s s
9
s:
CL
Bed/Rug
jess
o
tM 00 2^
CM N. G
9
"> "5 Z?
r^ CM G
9
52S
9
SS5
9
ro
c
LU
£S G
0
9 ' ~
" <° Si
CM r- G
9
n S G
d
0
o
LL
5§G
o
8$G
d
n r- -^
o c
D =
5 "
§ 0
o
o
in
o>
a?
§"
I
J3
TJ
C
ra
s
c
ro
o
o
«
0)
o>
8
c
g
S
a)
o
-SJ
ro
tn
-------
Q
5
m
ra a)
'35
00
0)
109
-------
Lead loadings for air ducts and bed/rug/upholstery samples
had the highest correlation coefficient after correction. Lead
loadings for window stool vacuum and wipe samples were also found
to be positively correlated. In addition, lead loadings for
entryway samples were found to be negatively correlated with
those for every other sample type after correction for the fixed
effects.
Lead Concentration
Table 4-7a contains unit-to-unit correlation coefficients
for the geometric mean lead concentration data. This table is
analogous to Table 4-6a, but is for lead concentrations rather
than lead loadings. The geometric mean lead concentration data
are plotted in scatterplot matrix form in Figure 4-9a. This
figure is analogous to Figure 4-8a.
There were several positive correlations found for lead
concentrations. Entryway and floor vacuum results were highly
correlated (0.94). Lead concentrations for floor and window
stool samples were also significantly correlated with those for
each of the soil sample types. In addition, lead concentrations
for all soil sample types had a statistically significant
positive correlation. The strongest of these correlations was
seen between boundary and foundation soil samples (0.98). It is
also interesting to note that there were no strong negative
correlations observed.
In Table 4-7b and Figure 4-9b, the relationship between
lead concentrations is examined after correcting for
renovation and abatement effects. This table and figure are
directly analogous to Table 4-6b and Figure 4-8b for lead
loadings.
110
-------
S
O
4J
-H
cd
r-
i
3
H
o
w
e
3
c
CO
T3
3
O
Li-
ra
5
^
c
LU
CO
a
c
3
0 C
T! C
C CO
§6
1 _
o
0
Li-
ra
c
LU
0)
«
Q
52s
o
«M 0 5-
«*) CD S-
o
°2 s
0
Sin Zr
CM S
o
Sin 5-
co S
o
SSs
o
9 '~
oo o ±r
o o> S
o
ts
Q
§
CO
SE5S
o
8!nS
9
? S s
9
CO S
9
n fe S
9
o co 5-
T- 00 S
d
9 ~
JZ
a.
3
1
g f° E"
o
wo £
0
8£ £
o
5?o£
9
§£ £
o
o
1
t:
LU
K§£
o
SS£
o
SS £
o
SS£
o
S£ £
o
o
o
u.
o S2,
o
E5S£
d
£>
CO
T3
C
3
O
V)
5S£
0
1
&
'-.'>"'<
-
c
o
01
T3
C
£
o
-o
0)
o
U)
_3
co
i
o
.a
c
CO
o
T3
I
o
-------
1
cc
CQ
CO
0)
CO
0)
h
0)
u-i
O
(4-1
ra
a
(d
0)
-rl
H D)
M 3.
Jj »* ^
0)
§§
0) -H
O)4J
(0
-H O
fc CS
-U O
(d u
o a)
rH rH
ft
^
a)
-P CQ
4-> 0)
m
o
cr>
tf
0)
H
3
tn
112
-------
O
(4-1
o
H
jj
o
a>
o a
u o
H
^ U
0) (0
JJ M
M-l 4->
*c d
0)
(Q U
0) ti
a o
Su
H
TJ
(U (0
H o
I -H
o -u
OJ (d
-u o
-H d
C
E
3
3
CO
c
g
U
T3
c
o
u.
CO
3
c?
"c
LLI
CO
T3
C
1
3 o>
o c
"O c
C CO
§6
3
O o
"O o
!*
^_
o
o
u.
CO
c
Ul
0)
Q; .c
"3 D
o>
m
ts
3
Q
O> O £?
CO CO Ci
o
0> CM S
0
SsS
ci
CN CO Ci
ci
O> O JCT*
IO CO Ci
O
ESS
ci
3°S
O
* * *
O> O) C7*
0> O Ct
d
T5
Q
'5
E
3
3
TO CM CT
TO CO Ci
O
ass
ci
TO 5 S
O
to co cp
N CD Ci
O
sss
ci
» f~- m'
ro CD C-
O
u> !? 52.
ci
SSS
ci
cSSS
o
o en 52.
9
"> "> Z?
n co 53-
0
^
CO
UJ
sss
ci
co i--. c;-
o> o J2.
o
0> 0 2-
o
KSS
ci
CM CO S^
o> o 52,
ci
o
o
u.
g
3
o
13
c
5
58S
o
58S
ci
£8S
ci
3 "g
Is
«O CM Z?
o> o £2.
o
S 2 S
ci
£
CO
T3
c
i
1
o> o 52,
0
^
CO
1
.0
To
o
c
3
£
113
-------
CO
5
Li-
ra
0)
0)
fa
ra a)
b ra
0) 4J
<4-l O
«H C
0)
n> 3
ti 0)
m 4J
a) a)
JJ
-H
0
pi
rt
o a
H O
fc -H
J-> 4J
a) at
M 0)
JJ Q) O
(tJ -U fl
O
-------
After correction for renovation and abatement effects, the
relationships among the lead concentrations appear stronger than
before controlling for them. The reduction in degrees of freedom
increases the threshold at which correlation estimates are
considered statistically significant, but there are many positive
relationships exhibited in the data, and some were statistically
significant.
Lead concentrations for floor samples are significantly
correlated with soil samples taken at the boundary (0.97),
entryway (0.93, marginal), and foundation (0.96). The lead
concentrations for soil samples are still strongly correlated
after controlling for the fixed effects. This may indicate that
it is not the fixed effect of renovation or abatement which
causes the data for these soil sample types to be correlated.
4.7 COMPARISON OF VACUUM AND WIPE SAMPLING PROCEDURES
One of the objectives of the Pilot Study was to compare the
vacuum and wipe sampling protocols. In each of the units a
"bridge room" was selected and side-by-side vacuum and wipe
samples were taken. The purpose of collecting these data was to
build a "bridge" between the sampling method for the full CAP
Study, the vacuum method, and the wipe sampling method employed
in the HUD Demonstration.
The vacuum versus wipe comparison data for floor lead
loadings, window stool lead loadings, and window channel lead
loadings are listed in Tables 4-8a, 4-8b, and 4-8c, respectively.
In Table 4-8a, all side-by-side duplicate floor lead loadings are
included even when they are not from the "bridge" room. These
measurements contain information on the expected variation
between side-by-side samples when they are taken using the same
sampling protocol.
115
-------
Table 4-8a. Vacuum versus Wipe Comparison Data:
Loadings (jig/ft2)
Floor Lead
Unit
33
43
17
19
80
51
Room
Kitchen
Living Room
Dining Room
Kitchen
Front Bedroom (BD1)
Living Room
Kitchen
Back Bedroom (BD3)
Kitchen
Front Bedroom (BD1)
Back Bedroom (BD3)
Sampling
Location1
3
1
1
3
1
3
3
1
3
3
1
Vacuum
#1
3.13
5.57
2.56
876
45.18
9.84
39.42
1069
2.50
59.42
31243
Vacuum
#2
2.08
4.21
4.64
4.57
36.03
8.68
31.82
8.31
1.45
374.03
409.98
Vacuum
Geo.
Mean
2.552
4.842
3.447
6.327
40.346
9.242
35.417
9.428
1.904
149.080
357.897
Wipe
#1
13.772
18.42
18.42
33.45
36.95
3832.53
Wipe
#2
13.772
24.27
30.12
36.95
22.96
1628.77
Wipe
Geo.
Mean
13.77
21.14
23.55
35.16
29.13
2498.46
Sampling location identifies a general location sampled in each room.
The lead levels in these two samples were below the level of detection for the wipe analytical method; value reported is
the detection limit.
Table 4-8b. Vacuum versus Wipe Comparison Data:
Lead Loadings (fig/ft2)
Window Stool
Unit
33
43
17
19
80
51
Room
Kitchen
Utility Room
Kitchen
Kitchen
Living Room
Living Room
Kitchen
Kitchen
Pantry
Front Bedroom (BD1)
Front Bedroom (BD1)
Sampling
Location1
4
1
1
4
1
4
1
4
1
1
4
Vacuum
Vacuum Vacuum Geo.
#1 #2 Mean
25.84
6.72
6.33
16.48 12.20 14.179
96.47
147.85 83.62 111.190
33.91
600.26
Wipe
#1
105.80
217.91
27.43
18.39
Wipe
Wipe Geo.
#2 Mean
121.73 113.486
30.53 23.695
24.42
190.75
163.11
421685
1142.59
504.54 759.264
Sampling location identifies a general location sampled in each room.
116
-------
Table 4-8c. Vacuum versus Wipe Comparison Data: Window Channel
Lead Loadings (/*g/ft2)
Unit
43
80
51
Room
Kitchen
Kitchen
Kitchen
Front Bedroom (BD1)
Sampling
Location1
1
4
4
4
Vacuum
#1
Vacuum
Vacuum Geo.
#2 Mean
9246.81
3771.04
6167.62 4822.69
Wipe
#1
335.38
658.39
Wipe
#2
631.05
Wipe
Geo.
Mean
644.58
1008.29
1225.76
1111.72
1 Sampling location identifies a general location sampled in each room.
With regard to window channel samples, the Pilot Study
design called for sampling from two split windows in the "bridge"
room in each unit. One window was to have both vacuum and wipe
samples taken and the other was to have either two vacuum samples
or two wipe samples taken. As is evident in Table 4-8c, sampling
window channels turned out to be a difficult task (e.g., windows
painted shut). Only four split window channels were actually
sampled, and only one window was sampled with both the vacuum and
wipe sampling methods.
The paired floor lead loadings from Table 4-8a are plotted
in Figure 4-10a. In the figure, lead loadings from wipe samples
are plotted versus lead loadings from vacuum samples. A
reference line which represents complete agreement between the
two sampling methods is also included. With one exception, the
lead loadings from wipe samples exceed the lead loadings from
vacuum samples. A statistical analysis was performed to quantify
this relationship.
Both the vacuum lead loadings and wipe lead loadings are
assumed to follow a lognormal distribution. For this reason a
log-linear model was employed to characterize the relationship
between wipe and vacuum lead loadings. The model fitted to the
data was
log(W) = log (a) + /3 log (V) + log(E)
(3)
117
-------
s
8
. o
E
!
s
a
o
H
4J
n)
u
o
0)
rH
n
j?
(0
u
-H
M
4J
0)
0)0o
U h
(d o
(d
o
H
i
0)
118
-------
where W and V are the geometric means for vacuum and wipe
samples, respectively, from Table 4-8a; E represents a random
error term which is assumed to follow a lognormal distribution.
Restating the model in terms of the wipe lead loading results
W = a V'3 E. (4)
If /? is not equal to one, the multiplicative bias between the two
sampling methods changes with the magnitude of the measurements.
However, if 13=1, there is a fixed multiplicative bias (a) between
the sampling methods which does not change with the magnitude of
the measurements. Also, for /?=!, the model of Equations (3) and
(4) simplifies to the assumption that the ratio W/V follows a
lognormal distribution with geometric mean 01.
This model of Equations (3) and (4) was fitted to the six
pairs of floor lead loading measurements plotted in Figure 4-10a,
and the hypothesis H0: 13=1, the hypothesis of a fixed
multiplicative bias, was tested. The estimate of /3 is 1.05 and
the observed significance level of the test is 0.90. Since the
hypothesis could not be rejected, the model was then refitted
with the j8 parameter set to one (1) . The estimate of the
multiplicative bias (a) of wipe over vacuum measurements is 4.76
with a 95% confidence interval of (1.52, 14.95). This result
implies that, on average, the wipe lead loadings are 4.76 times
larger than matching vacuum lead loadings on floors. The reader
should note that the slope of the estimated regression line
(dashed) in Figure 4-10a is strongly influenced by the
observation from the house with the highest loadings by both
methods.
The paired window stool lead loadings from Table 4-8b are
plotted in Figure 4-10b. The statistical analysis performed for
floor lead loadings was repeated for the window stool lead
loading data. For window stool lead loadings, the estimate of 0
is 1.07 and the observed significance level of the test of a
119
-------
8
E
?
5
£
c
o
H
4J
id
u
o
H
1
XI
to
e
id
§
o
-u .
0)
s tQ
co 6
> T)
U M
id o
X)
o
<«
-------
fixed multiplicative bias (H0: (8=1) is 0.66. Since, again, the
hypothesis could not be rejected, the model was refitted with the
18 parameter set to one (1) . The estimate of the multiplicative
bias (a) of wipe over vacuum measurements is 4.55 with a much
tighter 95% confidence interval of (2.66, 7.78). This
resultimplies that, on the average, the window stool wipe lead
loadings are 4.55 times larger than matching vacuum lead
loadings.
As evidenced in Table 4-8c, only one pair of window channel
lead loadings is available. Therefore, no statistical analysis
of the window channel data was performed.
The precision of the vacuum and wipe measurement techniques
can also be compared by examining the replicate sample log
standard deviation results in Tables 4-4a and 4-5a. The
replicate sample standard deviation (reported in the last column)
provides an estimate of the standard deviation of duplicate
samples taken side-by-side for each sample type. Examining these
values for floors, window stools, and window channels sampled by
both vacuum and wipe techniques, neither sampling technique can
be judged to be significantly more precise. Most data for this
type of comparison were available for floor samples. Here the
standard deviation for duplicate vacuum samples (0.47) was
observed to be larger than that for wipe samples (0.33), but
their confidence intervals overlap considerably. The 95%
confidence interval for vacuum precision was (0.24, 1.35). The
corresponding interval for wipe precision was (0.14, 1.60)
4.8 COMPARISON OF CAP PILOT DATA AND HUD DEMONSTRATION DATA
While conducting the HUD Demonstration project, detailed
environmental data were collected by HUD on all units. Interior
XRF/AAS results and lead loadings from the HUD Demonstration are
presented in Table 4-9 by sample type and room. The tabled
values are geometric mean values over all data collected in a
room.
121
-------
o\
I
0)
rH
X»
n)
<0
£
Q.
W
N Channel
o
0
C
§
10
v
Q.
(0
"5
0
w
s
5
(A
Q.
<55
o
o
u.
£°If
OQ: g s
> =
$ o &£
o il § ^
§§£%
iass
%°l%
5iE|3
o> CM in
§. co m CM
n m co
r-1 CD CN Tf
i^-
CO
CM , o
c c c c nJ o
Solo^S^o)
£0-0-0^-5 §.^
co CD a> co c .ts co .£
mmmmLU^_j_i
CO
CO
r- r-
CD t--'
IT O)
CN CN
O>
<*
CO
un
co o co co
co en ^- 10
r~ CM ^p
cb CD in
CO
*
CN
10 CO h- T- ^3-
co T- I-- tn
. .(DO . CM O>
CN tf5 CO CO
CM IT- CO CN
t
CN
t>-' CN co m cn o co
co CN m co in oo
CD h-
CO CN O * CN O O
co cxi co ^r co co o>
o o o o oo o o
*«- *
T- CN E E E
* * * o § g
|Sifi!*c2
iiif^ll^
SCO CO C CO «- >
m do Q u O * j
co
*
o
CD
t^
O)
»- CN ^f
O CO C)
M" T T
CN
TT
CN
T- in T- CN
. m . «- . o . if
cq CN q CN
t^- CO CO CT>
CO CN
in
CO
CN
Tt T- co oi
T
ot^co omoi^-cn
T--^-in ococoo*-
incJ>o o m co « o
CNCOCO in^-mcoTj-
ooo ooooo
ssl^c! I
iei12!^ ct-s
SgoOl-CcB "°D)
£^ fc C ^* C
mmcoaujox^-j-i
t^
T3
a>
ro
£)
ro
ca
c
o
CO
U)
§.
o
122
-------
0)
H
JQ
(0
in
u
Q.
w
« Channel
o
T3
c
§
in
V
a.
m
"6
0
W
c
§
(A
«>
Q.
CO
w
o
o
ish
5o:|S
£3 if
oil?3
§i.if
xSg^
is if
5i|s
I o &f
o a 5 §
§i£f
ifi$5
E~
|o|%
5ix|3
|S if
o h. § J|
^^
§ lit
=: a) > o)
X Q? ^
c5^
Q |£ |
x|x|
o
o
OL
.e
c
(O
8
CM
1^
en
CM
>n
§h- o
co cb d
T- O)
10
c>
en
< CN i^ o>
CO CM T- IO
'CM '00 * CO
1^ CO IT) IT)
^ Tt CO CD
to
iri
CO
» o o »- o
CO 0 O CM O
O O O (D O
CO «-
m CD uj x * _i
O)
o o>
ci cvi
CM CM
CO CO
*f
O) O O r-
CO T- CM T-
tf> m o> v-
1^ TT i-^ CO
00 T- T- CO
o «-
CO
^
CO
co
o>toooco »- 'incMco
m^-co*fr co T-COCM
oo ^~
co
en in co TJ-
O CO t^ O
o ' r- CM en '
CO CM * T^
CM -^
co
cri
CM
8 88 8 888
m T- o> co t~- OD CM
o> m T- o> in oo *-
in T- to CM co T-
(O
CM
T-orocomo> cj>om
COOOCD^CD T-OCM
mooi^ooo -ooco
comoir-coco T-com
CMO'-'^OO ^00
* *
T- CM CO t £
* * * «: ^ 0 >, 0
isis|filcfiv
iiiiii^ft
njro.£c.*;>ro
mcommmoiiJ^-iQ.
O
co
^- CM
in co
O> CO
CD t
CM
r- CD C3) t^ 00
T- CM T- CO
h- CO - r- CM ' ' O P~ O
CD h- CO CM -- co
CO O » T^
^- ci co in
CD * m ^~
m i- co ^r
CD
CO
o>
*
CM
cooom -oooooo
ocomcM coooinoo
M-COM-CO o> * T- ^-
T T ^ CO CO T
COCOOOOO T- TtOO
mcnoorMco o ID o o
T-T-Ti-r^to -m -moco
inT-cocMS- >- ooo
COCMtfOr- O ^-T-TJ-
CM
* « « .
g . r- CM CO £
*«***>, 0
EEcccra «**S
ggooo§ CMC^^
llses^' *«?g>
££T3TD-D£i = = t3 3~
conjcDa>cpccoro*ito.>
mmcomoQLiJXX^-i-i
LO
0)
CO
.O
CO
(A
ca
E
o
S
T5
I
O
0)
a.
in
o
123
-------
Along with the HUD Demonstration data in Table 4-9, lead
loadings for vacuum and wipe samples from the CAP Pilot are also
reported. These tabled values are also geometric mean values
over all data in a room. As evidenced by the sparseness of Table
4-9, there are very few rooms in which there are both HUD
Demonstration wipe data and CAP Pilot data. The best such
comparative data are found for floor samples where there are 11
rooms that have both HUD Demonstration wipe samples and CAP Pilot
vacuum samples. Data for these rooms are plotted in Figure 4-11
where the CAP Pilot vacuum lead loadings are plotted versus the
HUD Demonstration wipe lead loadings. There appears to be little
agreement between the two sets of measurements.
In the case of HUD Demonstration XRF/AAS measurements, there
are several rooms in the CAP Pilot units where comparisons are
possible. In Figure 4-12, the CAP Pilot and HUD Demonstration
lead loadings are plotted versus the HUD Demonstration XRF/AAS
measurements. Separate plots for floor lead loadings, window
stool lead loadings, and window channel lead loadings are
presented as Figures 4-12a, 4-12b, and 4-12c, respectively. HUD
Demonstration and CAP Pilot floor lead loadings appear to
increase slightly with increasing XRF/AAS readings. For example,
the highest geometric mean lead loading for CAP Pilot wipe
samples (around 2498 /ig/ft2) was in a room with a relatively high
XRF/AAS reading (4.648 mg/cm2) .
Window stool lead loadings (Figure 4-12b) show a somewhat
stronger increasing trend with increasing XRF/AAS readings. This
pattern is evident for all three types of lead loading
measurements. If anything, window channel lead loadings (Figure
4-12c) may show a slightly decreasing trend with increasing
XRF/AAS readings.
Table 4-10 is similar to Table 4-9 but contains exterior
XRF/AAS measurements from the HUD Demonstration, and soil lead
concentration measurements from both the HUD Demonstration (post-
abatement) and the CAP Pilot. Data for the sides of the units
where both HUD Demonstration and CAP Pilot soil lead measurements
124
-------
+
SorpBcyx pBG^i mnnoB/^ lojltl dVD
T
S
a
<0
0)
u
H
4J
I
0)
0)
<0
4J
a)
0)
ft
c!
O
2 O
S O
PN
S -u
m -^
3 0)
to =L
M ^
0)
> g)
9^
3 (0
O O
(0 rH
>'0
4J 10
O 0)
rH H
H
ft M
O
0)
125
-------
4- O
O +
+o +
+ + DO
4-
O
0
+ 4-0* E_ *
+
a a +o
O
s
3
q q q q q o
I 9
m
0)
rH 4J
M 3
0 CO
0 0)
rH M
4-1
8 3
-H \
Jj JV,
(0 tf
i>
u a
a o
O -H
S 4->
4J
to
Tj M
(0
0)
g.
H
126
-------
D O +
0 +
+ o
+ o
o o
q
o
9
3
o
05 43
tn
o ^^
H 4J
Q|4-l 0)
^^ G
CU D) (fl
< =t 0)
U 6
*
0)
I
H
127
-------
n +
o +
o+
+ 0
O + D +
n
o o
s
+ O
+ O
+ O
o o o o o
O O O Q «*
fc_fc **
O) M
"H Q)
id o
O a)
d
id ..
0) Q)
iH jj
-------
o
o
o
(8/8n)
o
§
8
I
L
o
3
L S
(0
i
U
-H
M
4J
0) 4J
8 -
0)
0)
at
Q ti
0)
O
m
3 TJ
ra m
n a>
-------
Table 4-10.
Geometric Means for CAP Pilot and HUD Demonstration
Data by Side of Unit: Exterior XRF/AAS Results
(mg/cm ) and Soil Lead Concentrations
Unit
33
43
17
19
80
51
Location
Back Yard
Front Yard
Left Side Yard
Right Side Yard
Back Yard
Front Yard
Left Side Yard
Right Side Yard
Back Yard
Front Yard
Left Side Yard
Right Side Yard
Back Yard
Front Yard
Left Side Yard
Right Side Yard
Back Yard
Front Yard
Left Side Yard
Right Side Yard
Back Yard
Front Yard
Left Side Yard
HUD Demo XRF
of Adjacent Wall
(mg/cm2)
0.1
0.2
6.6
10.8
6.9
5.3
13.5
9.3
HUD Demo
Soil
(pg/g)
288.7
318.9
443.0
1112.8
70.0
120.0
90.0
90.0
558.0
500.0
920.0
539.2
1218.0
1026.4
CAP Pilot
Soil
G^g/g)
108.210
171.237
180.720
287.027
67.514
70.240
49.180
238.390
381.288
941.590
479.202
937.650
130
-------
o +
o
o
o
r4
CO
(8/8n) uopiBiineancQ peaq
0)
M
g,
H
fc
131
-------
were taken are plotted against each other in Figure 4-13. This
plot shows good correlation between the two sets of soil lead
concentrations, with the HUD Demonstration concentrations being
about 25% higher on average. The fitted regression (dashed line)
has a nonsignificant intercept -0.15 and a significant slope
coefficient of 0.99.
In Figure 4-14, the CAP Pilot and HUD Demonstration soil
lead concentration values are plotted versus the exterior HUD
Demonstration XRF/AAS readings. In this figure there is
apossible increasing trend evident in soil lead concentrations
with increasing XRF/AAS readings. This pattern appears for both
the CAP Pilot and HUD Demonstration soil lead concentration
measurements.
132
-------
5.0 STATISTICAL ANALYSIS OF QUALITY CONTROL DATA
In order to assure that the sampling and analytical
protocols employed in the Pilot Study were producing data of
sufficient quality, a number of different quality control (QC)
samples were included in the study design. These QC samples are
designed to control and assess quality in the: (1) collection of
samples in the field, (2) preparation of field samples for
laboratory analysis, and (3) quantitative analysis of samples in
the laboratory. The quality control samples included in the
study may be organized under four major categories:
Blank Samples: Trip blanks, field blanks, method
blanks, and calibration blanks
Recovery Samples: Reference material samples, spiked
samples, calibration verification samples, and
interferant check standards (ICP only)
Duplicate Samples: Side-by-side field samples and
spiked duplicate samples
Interlaboratory Comparison Samples: Side-by-side field
samples to be analyzed by two different laboratories.
In general, analysis of the QC data led to the following
conclusions:
1. Overall, analysis of the blank samples suggests little
if any procedural contamination. Blank contamination
was, noted in the dust wipe blank samples used for
sampling, but not in the dust wipe blank samples used
for field cleaning.
2 . With the exception of one very low percent recovery for
a flame atomic absorption (FAA-W) reference material
sample, the results for all recovery samples indicate
very good method performance.
3. Spiked duplicate samples created in the laboratory
exhibited very good agreement. With the exception of
one pair of soil samples, side-by-side field samples
exhibited good agreement, but also exhibited some
inherent variability as would be expected in field
duplicates.
133
-------
4. Though the estimated ratio of results from the secondary
laboratory to results from the primary laboratory
suggest the primary laboratory lead concentrations are
slightly lower, this difference is not statistically
significant. There appears to be no laboratory bias.
Detailed results of statistical analyses performed on the data
from each of the four categories of QC samples are reported in
the following sections. The quality control samples were assumed
to follow a lognormal distribution and were, therefore, log
transformed prior to analysis. The small number of samples for
each type of quality control procedure precluded an effective
evaluation of their distribution. For the majority of quality
control samples, statistical analysis of the untransformed and
transformed data suggested both could be normally distributed. In
Section 4.1, evidence is cited supporting the log transformation
of the field samples. The log transformation was, therefore,
employed also on the quality control samples.
5.1 BLANK SAMPLES
Blank samples are samples which are expected to contain no
lead or only a very small amount of lead. In the CAP Pilot Study
four types of blank samples were analyzed: trip blanks, field
blanks, method blanks, and calibration blanks. Each type of
blank sample served a specific purpose. Trip blanks were
analyzed to identify any problems with the gravimetric procedures
used to determine the amount of dust collected by the vacuum
sampling method. Field blanks were analyzed to identify sample
contamination anywhere in the normal process of sample
collection, transport, preparation and analysis. Method blanks
were analyzed to examine sample contamination in the normal
process of sample preparation and analysis. Calibration blanks
were analyzed to examine any changes in instrument performance
that may effect estimated lead concentrations reported for
regular study samples.
134
-------
Only gravimetric analysis was performed for trip blanks.
The trip blank data consists of pre-field and post-field weights
(mg) of 52 cassettes sent to the field. The difference between
the post-field and pre-field weights was assumed to be normally
distributed. Unlike other quality control samples, trip blanks
did not involve the measurement of lead content. As a result,
the simple assumption of a normal distribution was utilized. The
arithmetic mean difference was 1.8 mg, with a standard deviation
of 0.2 mg. An estimated 95% tolerance interval for the
difference is (1.2 mg, 2.3 mg). The cassettes, therefore, return
from the field weighing marginally more than they did before
leaving. However, since the estimated bias of 1.8 mg is small in
comparison with the geometric mean dust amounts in Table 3-4, no
adjustment was made to sample weights or concentrations.
The three other types of blank samples (field, method, and
calibration) were all analyzed for lead content. Just as with
the regular study data, the measured amount of lead per sample
was assumed to follow a lognormal distribution.
Data for the three types of blanks were generated for each
of the following four combinations of sample medium, sampling
method, and analytical method:
Dust by Vacuum by GFAA (GFAA-V)
Dust by Vacuum by ICP (ICP-V)
Dust by Wipe by FAA (FAA-W)
Soil by Core by ICP (ICP-S)
Descriptive statistics are reported for the data from blank
samples in Table 5-1.
The descriptive statistics reported include the number of
samples, number of results above the detection limit, minimum,
and maximum. When possible, the geometric mean and logarithmic
standard deviation for the amount of lead per sample are
reported. In addition, a 95% upper confidence bound on the .95
quantile for the amount of lead per sample is also provided. For
135
-------
M
a
(d
iH
ffl
d
-H
0)
rH
e*
CO
^
d
(0
0)
j
di
a.
n
o
M-J
O
§
o
m
0)
o
a
ni
M
0)
H
o
EH
tQ
O
H
V
IQ
H
4->
(0
4J
W
0)
H
Jj
ft W
-H 0)
h rH
U Pi
U 3
(1) n)
Q CQ
I
in
(U
(0
H
S? CD
LO O T-,
0) C c
^ CO ^
(U (U O
5"i~
s c
"o .Q
CO 5
z Q
o
E^ CD
(U
o 2
O
3
'x
CO
5
E
'c
ii
cu c
I'll
*l
CD
O. CD
ro CO
CO
° '5
(U _>>
Q- CO
14
O _!£
>>CQ
s
CD
0
*~
CD
CN
O
CO
CO
§
O
en
en
i
0
P
ii
CO
00
?
en
CO
0
CO
CO
o
CN
o
00
in
CO
co
CO
FAA-W
JN :
fcj
*1
$i
O
''&
n;
g!
°:
«
CO
to
5
IS
o
co
CD
0
CN
T
o
o
CO
o
s
o
CN
CN
GFAA-V
o
o
S2
(U
ss
«
«
0
8
0
g
O)
*"
0
<5i
8
(Q
»*
O
-
CO
Q.
O
fe
-*
V
***
^
^
CN
^
FAA-W
b
«
b
T'
i
3
*-
«.
<">
k
0
CO
,
CO
QL
O
00
co
«-
CD
CO
O
O
,_
1
LJ-
0
c
I
ja
"co
O
fr
o
&
o
g
IS
o
5
o
CO
CN
0
CO
ol
g
*'
o'
o»
o
s
o
X
o
fe
di
CN
CN
FAA-W
O)
8
Oi
IN
§
s
0
s
0
CO
m
O
T3
C
O
.a
co
CO
O
;§'
«
JO
E
£
CD
U)
CO
CD
E
CO
JO
D
0
£
-------
the sake of simplicity, this bound will be referred to as the
estimated 95% tolerance bound. These calculations were possible
only when a sufficient number of results above the detection
limit were obtained in a category.
If all results were above detection, calculation of the
geometric mean and logarithmic standard deviation was routine,
and the estimated upper 95% tolerance bound was calculated using
an exact procedure for lognormal distributions. In instances
where a portion of the quality control data was censored on the
left (e.g., field blank samples), a lognormal model was fitted to
the data and its parameters estimated. The SAS procedure LIFEREG
was utilized in obtaining these estimates. LIFEREG maximized the
log-likelihood function via a ridge stabilized Newton-Raphson
algorithm, thereby providing maximum likelihood estimates of the
log mean and log standard deviation. In these cases, an
approximate procedure was used to calculate the estimated 95%
upper tolerance bound using the detection limit for each sample
as the censoring value. The approximate nature is due to
employing the maximum likelihood estimates in determining
traditional 95% tolerance bounds. Since the traditional approach
does not include an adjustment to the bounds reflecting censored
data, the estimated tolerance bounds are approximate. When a
high percentage of the results were below detection, it was not
possible to calculate a geometric mean, logarithmic standard
deviation or estimated 95% upper tolerance bound, and these
fields are left blank.
When spiked and spiked duplicate cassette and wipe samples
were analyzed, an unspiked cassette or wipe was also analyzed.
These unspiked samples have been included in Table 5-1 as method
blanks.
The data for blank samples are illustrated in Figure 5-1.
The amount of lead (^g) found in each blank sample is plotted by
category. Different plotting symbols are used to indicate
137
-------
o oo
+e o o
-H-
s
g
d
V
O
O
O
0)
rH
to
a
a
0)
D)
M
0
CD
d
o
(D
rH
O
JJ
I
to
a
"0 (d
« H
H A
H
in
ID
H
138
-------
whether the result was above detection or below detection, in
which case the detection limit is plotted. In those cases where
an estimated tolerance bound could be calculated, the estimated
95% upper tolerance bound is illustrated in the figure by a bar
which has the tolerance bound as its upper value.
Dust wipes appear to contain more background lead, or to
become contaminated by routine handling, to a larger extent than
do the other sampling media. This is evidenced by geometric
means of 10.63 /ig lead per sample for field blank wipes, and 2.26
/xg lead per sample for method blank wipes. Analysis of the
vacuum dust and soil field blanks suggests them to be only
marginally contaminated by sample handling. With the exception
of wipes, results from the method blanks suggest that the
laboratory procedures correctly report a negligible amount of
lead when the sample contains none. Similarly, the calibration
blank results provide evidence that the calibration regression
equations remained valid. The GFAA calculated vacuum dust
calibration blanks could not be examined since all were below the
detection limit.
Because it was suspected that the brand of wipes used in the
HUD Demonstration are contaminated with measurable amounts of
lead, pre-field testing of wipes was conducted revealing lead
levels similar to those found in the CAP Pilot blank samples.
Despite this, the same brand of wipe was used in the CAP Pilot to
maintain comparability with HUD Demonstration results, and
because the contamination level was small relative to the
expected amounts of lead in regular samples.
5.2 RECOVERY SAMPLES
Recovery samples are samples which contain a known amount of
lead or have been spiked with a known amount of lead. Four types
of recovery samples were incorporated in the CAP Pilot Study:
reference material samples, spiked samples, calibration
verification samples, and interferant check standards (ICP only).
The reference material samples verify the ability of the
139
-------
laboratory procedure to correctly determine the amount of lead in
samples similar to the regular samples. Spiked samples verify
the ability of the laboratory procedure to correctly determine a
known amount of lead in regular study samples. The calibration
verification samples evaluate the continued viability of the
calibration regression equations. The interferant check standard
samples are a check on the effect of interferences to the ICP
analysis procedure. Again there are four combinations of sample
medium, sampling method, and analysis method of interest.
All spiked samples, including both members of each spiked
duplicate pair, are included in the calculations in this section.
For GFAA, the first continuing calibration verification (CCV)
sample in each batch of samples processed was excluded from the
calculations since this result is simply a repeated recording of
the results for the midpoint calibration standard, relabelled as
a continuing calibration verification sample.
For all but spiked soil samples, the analytical result for
each recovery sample was taken to be the ratio of the measured
amount of lead in a sample to the known amount of lead in the
sample. When multiplied by 100, this value is commonly referred
to as the percent recovery. The percent recovery value is
assumed to follow a lognormal distribution. If the geometric
mean of the lognormal distribution is 100%, this is an indication
that lead is over-recovered half the time and under-recovered
half the time. Percent recovery values over 100% indicate a
measured value exceeding the known amount of lead, and values
under 100% indicate a measured value below the known amount.
The analysis of spiked soil samples required slightly
different procedures. Spiked cassette and wipe samples were
created by spiking a known amount of lead into a new cassette or
onto a new wipe. Therefore, the amount of lead contained in
these samples was known. However, spiked soil samples were
created by spiking a regular soil sample with a known amount of
lead. Since the sample already contained some lead, a different
140
-------
calculation of percent recovery was required. For spiked soil
samples, percent recovery was calculated as
measured /xg lead
in spiked sample
measured /xg lead in
unspiked sample
100
lead in spike
As before, the percent recovery value was assumed to follow a
lognormal distribution.
Descriptive statistics for recovery samples are reported in
Table 5-2. The descriptive statistics reported include the
number of samples, minimum, maximum, geometric mean, and
logarithmic standard deviation. Also, an estimated 95% tolerance
interval (upper and lower 97.5% tolerance bounds) was calculated
using an exact procedure for lognormal distributions.
The data for recovery samples are illustrated in Figure 5-2.
The percent recovery for each recovery sample is plotted by
recovery sample category. The estimated 95% tolerance interval
is illustrated in the figure by a bar extending from the lower
tolerance bound to the upper tolerance bound.
The analysis of the recovery samples indicates good recovery
of the lead. The only estimated tolerance interval that does not
contain 100% is that for FAA-W calibration samples. However, all
values in the estimated tolerance interval, (101%, 112%), are
very close to 100%. The estimated tolerance intervals for spiked
samples, calibration verification samples, and interferant check
standards are all narrow, indicating good method performance.
The estimated tolerance intervals for GFAA-V, ICP-V, and ICP-S
reference material samples are wider; however, this is due
primarily to the small number (4) of samples analyzed. Though
more FAA-W reference material samples (10) were analyzed, one
very low value results in a wide tolerance interval. With this
very low value removed, the estimated tolerance interval for
141
-------
ti
H
fr
0)
O
0)
PI
o
M-l
0)
o
w
H
u S
Pi M
H 0)
fc >
O O
CO U
0) 0)
Q tf
CN
in
0)
a
S-> o -r,
O) C
s s 1
D 1-
O*1 (D
si?
» s 1
g
II
co g
5°
O
'£
fl>
o 2
e
OJ
E
E
,c
tB
ra W
W
5-S
a) .&
a. ca
1!
s
CO
CO
5
o
3
m
o
^
co
.
I
O
1 Reference j
r-
CM
co
CO
m
o
o
CO
cn
^
CO
cn
0)
oo
CO
^
QL
O
1 Material
m
CO
in
CM
o
-------
L ^ -si
_
A-dOI 'j -j
Q
<
!
m \ ** d
V
._.
OB
g
M-wa T
J
A ^ni ff
*-V«D "
Effln K-dQI
r -
i 44i*i. . M-W2 ffl
.»
ES3
^
&
\-dOI CO
A-WX)
3
1 1 4. 44. 4. , P dMI "C
,
M VW 1
A-dOI |
i
A-WZ) iS
J
K
8 § 9
°
yCiQAOTKfrr IT^rW^VT
a Ti
a
rl
>1
recover
jj
ti
0)
o
Q)
Pi
O
(4-1
(0
T)
0
a)
o
0)
rH
0
4J
o
CO
4J
C
0)
s .
i-i m
rJ UJ
W 0)
nJ H
rH IQ
nt
tJ rl
rl 0)
H 0
d o
d a)
H 2
OJ
in
Q)
rl
H
143
-------
FAA-W reference material samples (85,153) is also quite
satisfactory.
5.3 DUPLICATE SAMPLES
Duplicate samples are samples which are expected to be
similar either because they were collected side-by-side in the
field (side-by-side samples) or they are created to be similar in
the laboratory (spiked duplicates). In both cases the samples
are analyzed one after the other in the same analytical batch.
Note that the side-by-side soil samples collected for the purpose
of interlaboratory comparison are also included in these batches.
The analytical result for each pair of duplicate samples was
the ratio of the larger measured lead result to the smaller
measured lead result. This ratio has a minimum value of one.
The log of this ratio was assumed to follow the absolute value of
a normal distribution with mean zero and standard deviation a.
Descriptive statistics for duplicate samples are reported in
Table 5-3. The descriptive statistics reported include the
number of samples, maximum ratio, and logarithmic standard
deviation. Also, an estimated 95% upper tolerance bound was
calculated using an exact procedure for lognormal distributions
with known geometric mean.
The data for duplicate samples are illustrated in Figure 5-
3. The ratio for each duplicate pair is plotted by duplicate
sample category. The estimated 95% upper tolerance bound is
illustrated in the figure by a bar extending from a value of one
to the upper tolerance bound.
The duplicate sample results suggest good agreement between
spiked duplicate samples. With the exception of one pair of
side-by-side soil samples, good agreement is also exhibited for
side-by-side samples; however, the inherent variability between
field samples, even when they are collected side-by-side, is
evidenced by the higher ratios and tolerance bounds for these
sample types.
144
-------
n
in
- 3
v
a. >
Q. ra
>> c
1- <
co
ID
in
ID
CM
d
O)
O)
s
>
QL
O
S
CO
o
in
m
CO
d
r_
co
t
?
>
Q_
O
d>
a
I
a
i)
T3
/)
CM
CM
CM
O
CD
C3
CD
O>
Ol
s
CO
0.
g
!J
-
co
CM
O)
-5"
^
CO
Q.
o
co
q
(O
s
[^
o
0)
Q.
F
ra
in
S
O
1
Q.
co
o
CO
m
0
c>
T
T
"5.
F
ra
in
S
>
Q.
O
V
1
Q.
Q
m
CM
M-
o
[s»
^~
a>
a.
F
ra
jn
5
§
^
CO
q
-
o
o
CM
0
0)
Q.
E
3
*
CO
Q.
O
8
CO
en
.g
O
^
CO
co
o.
ca
in
ra
Q.
O)
c
T3
3
15
x
Ol
145
-------
i 1-E^
*
«H
O
0
-H
4J
id
(1)
O
4-1
ra
T)
0)
o
a
m
-------
5.4 INTERLABORATORY COMPARISON SAMPLES
Interlaboratory comparison samples were utilized to examine
possible laboratory bias in the analysis of regular field
samples. Side-by-side vacuum cassette samples and soil samples
from each pilot unit in Denver, as well as six units in Baltimore
(Battelle and Kennedy Krieger Institute, 1992), were randomly
sent to the primary and secondary laboratories. In the case of
soil, the samples were homogenized and split before being sent to
the two laboratories. The analysis results from these samples
were compared to identify any systematic differences between
results reported by the two laboratories.
The data used in the interlaboratory comparison were the
ratios of the secondary laboratory result to the primary
laboratory result. These data are plotted in Figure 5-4. In the
statistical analysis, the ratio data were assumed to follow a
lognormal distribution with a geometric mean of one (1). The
interlaboratory comparison data were analyzed with a general
linear model which included effects for laboratory, city, side-
by-side variation, and unit-to-unit variation.
The geometric mean ratio for the vacuum cassette samples was
1.07 for Denver units and 0.95 for Baltimore units. Since the
hypothesis tests of equal variance and equal geometric mean
ratios were both accepted, the data were pooled. The pooled
cassette data had a geometric mean ratio of 1.01 with an
estimated 95% tolerance interval of (0.24, 4.26).
For the side-by-side soil samples, the hypothesis test of no
laboratory bias was accepted for both the Denver and Baltimore
units. The data had equal variances and no significant
laboratory-by-city interaction effect, so the soil data from both
cities were pooled. The pooled soil data had an estimated
geometric mean of 1.09 and an estimated 95% tolerance interval of
(0.82, 1.50) .
147
-------
o
9
(S/8n) uot
S
H
O
CD
ni o
H -H
Pi id
_ &
wl Fj
to u
h
o
^ -U
O>
M M
i> O
fl «H
0)
o ra
C d
o o
O -H
JJ
d rt
id h
0) 4J
>t O O
^ c! fl)
O 0 H
U O Pi
o id m
X» 0)
Id rH ID
rH 4J
>i M 0)
H O CQ
Id 4J BJ
*O id id
a n o
o o
O >O *0
a) id a
in
a)
g,
H
148
-------
The geometric mean ratios are similar for both cassette and
soil samples, although the soil data suggest that the primary
laboratory results are slightly lower than those of the secondary
laboratory. This difference, however, is not statistically
significant.
149
-------
6.0 REFERENCES
Battelle Memorial Institute and Kennedy Krieger Institute, 1992,
"Final Report for the Lead-Based Paint Abatement and Repair and
Maintenance Pilot Study--July 9, 1992", drafted under Contract
No. 68-DO-0126.
Battelle Memorial Institute and Kennedy Krieger Institute, 1992,
"Quality Assurance Project Plan for the Kennedy Krieger Institute
Lead-Based Paint Abatement and Repair and Maintenance Study--July
22, 1992", drafted under Subcontract No. 257-9801-1 (WA 11-78)
and Prime Contract No. 68-DO-0137.
Battelle Memorial Institute and Midwest Research Institute, 1991,
"Quality Assurance Project Plan for the Abatement Performance
Pilot Study--April 12, 1991", drafted under Contract Nos. 68-DO-
0126 and 68-DO-0137.
Farfel, M., 1987. "Evaluation of Health and Environmental
Effects of Two Methods for Residential Lead Paint Removal",
Doctoral Dissertation, Johns Hopkins School of Hygiene and Public
Health, Baltimore, Maryland.
Midwest Research Institute, 1991, "Engineering Study to Explore
Improvements in Vacuum Dust Collection--December 9, 1991",
drafted under Contract No. 68-DO-0137.
Reeves, R., Kjellstrom, T., Dallow, M., Mullins, P., 1982, New
Zealand Journal of Science, Vol. 25, p. 221-227.
150
-------
50272-101
REPORT DOCUMENTATION
PAGE
1. REPORT NO.
EPA 747-R-93-007
3. Recipient's Accession No
4. Title and Subtitle
Comprehensive Abatement Performance Pilot Study Volume I: Results of Lead Data Analysis
5. Report Date
February 1995
6.
7. Author(s) Bruce Buxton, Steve Rust, John Kinateder, David Burgeon, Fred Todt, Gary Dewalt
8 Performing Organization Rept. No.
9 Performing Organization Name and Address
10. Project/Task/Work Unit No.
0301104-05
Battelle Memorial Institute
505 King Avenue
Columbus, Ohio 43201-2693
and
Midwest Research Institute
425 Volker Boulevard
Kansas City, Missouri 64110
11. Contract(C) or Grant(G) No.
(C) 68-D2-0139
(G)
12. Sponsoring Organization Name and Address
U.S. Environmental Protection Agency
Office of Pollution Prevention and Toxics
401 M Street, S.W.
Washington, D.C. 20460
13. Type of Report & Period Covered
Final Report
14.
15. Supplementary Notes
16. Abstract (Limit 200 words)
This report presents the results from the pilot study that preceded the Comprehensive Abatement Performance Study. The goal of the Comprehensive
Performance Study was to assess the long-term impact of lead-based paint abatement. The pilot study was conducted to test the sampling and analysis
protocols that were intended for the full study. These protocols called for determining the levels of lead in dust and soil samples collected at residential
units. The pilot study was conducted at six houses, and all steps that were planned for the full study were included in the pilot.
The major finding of the pilot was the difference between wipe and vacuum methods for collecting dust. The choice of method had a noticeable
impact on the level of lead associated with the collected sample In addition, an inter-laboratory comparison of dust and soil samples indicated no systematic
difference in lead levels between the two laboratories. Also, intra-laboratory comparisons of sample results by inductively coupled plasma-atomic
absorption spectrometry (ICP) and the more sensitive graphite furnace atomic absorption spectrometry (GFAA) indicated good agreement within the common
domain of instrument detection limits.
Estimates of random house-to-house, room-to-room, and side-by-side sample variability were obtained for most of the sample types in the study.
These estimates were used for determining the number of houses and number of samples per house for the full study.
17 Document Analysis
a. Descriptors
Lead, Lead-Based Paint Abatement, Statistical Analysis
b. Identifiers/Open-Ended Terms
HUD Abatement Demonstration, Encapsulation, Enclosure, Removal
c. COSATI Field/Group
18. Availability Statement
Release Unlimited
19. Security Class (This Report)
Unclassified
20. Security Class (This Page)
Unclassified
21. No. of Pages
158
22. Price
(SeeANSI-239.18)
OPTIONAL FORM 272 (4-77)
(Formerly NTIS-35)
Department of Commerce
-------
APPENDIX A
CAP Pilot Study Data
A-l
-------
CO
CO
o
f
(/)
0
a!
Q_
s
X
o
CO
c_
_J
Dl ~
C CM
"O 4-*
CO 73 **-
CJ CO ^
_1 O O)
C
o
O 4J
o
cj cD
CO 4-*
CD
a
i
4-t *>
CO t_ ^.
«g §>
CD »-*
U
C
o
o
"S
4-1
0) U
4->
'£!
4-»
C Z
^
t
CO
10
1
-
z
CD
CO
CO
4J
C
CD
C
o
u
CO CD O
c §"0!
CD CO O
U CO O
x tr
'L '!§'!§ r 1 r 1 1 r 1 f
O> (M N- CO IO «-
O* O CO «- O O
xo «- o m ^o CNJ
fv -d- ro ro
V
vO co o o o co -o cxi >o co ro *~
N^ OQ N- »«J' "J1 ^ %4" O*> O *~" O f*"-
o ro cxi ON co m r>» *o r** rn co *«^
N. *o r^- *o *o ^o « «o m r^ r\j (\i
ro f- ro ro (M
0*0*00*0*0*0*00*0000
s4-N±-^->4-roro^-rorororororo
*X.^<^^^*^Nls.'s»s*1V,s«s.s*s.vs4sX
inminininmmininmininm
ooooooooooooo
a.a.o « ro
ro vt «-
c>S PS
ro CM c\j
o o o o
ro ro ro ro
^ *. X. V^
in in in in in
o o o o o
a.
u
CO CO (D
333
O) D) O)
CD CD CD
III
333
(D CO CO
i i i
333
0 0 £
333
rosrro
a. a. a. a.
u u cj cj
CO (0 CO O3
3333
O) O) D1 O)
CD CD
M 5 5 3
3 u- u- UJ
sf in (-.-
X
i_
E E E 0
T 'Z T a;
N. *O ir\
ru «- «-
in
38 3§
ro
ON O^ ON O^
ro ro ro ro
""X *"X "*X, *^s
m in if\ in
o o o o
cxSo.^
LJ u. U u-
C
O
t/i
L.
(D (D (D O
3)g>3>^
CD CD CD CO
a: o: of _i
| | | |
3 D 3 3
> > > £
1 1 1 1
3333
S***
< u. u- u.
>0«- «-
XXX
ili
ro
in
V
« tn o
CM
O O ON
ro ro ro
^*N *X ^
in in in
o o o
a. < a.
CJ u- U
c c
CO CO
CO
3^2
D) 41 01
Q) -r- -^
iS§
> > >
1 1 1
333
Q Q Q
5 * *
CM
X
t_
O
r^
0
0
ro
£
o
ro
x^
in
o
a.
CJ
CO (D CD
333
D) O) O)
o a> a>
OC. 0£ C£.
Ill
U O O
1 1 1
333
Q Q Q
Sou
Z> 3 3
in ro >t
X
L.
CM
ro
S
o
CM
CM
O
ro
x^
in
o
a.
u
ro co
3 3
D) O)
* O
CM ro CM o
CM >o o ro
in in r in
C^ ON O* O1*
ro ro ro ro
>^ *X X^ s^
in in in in
0 O O O
a. a. a. a.
CJ CJ U U
!2
CO
z
1
ro co CD
D) CD O) D)
CJ >^ CD QJ
a: u- DJ ae
|
3
O
CO
O O O 3
CO CO CO Q
Q Q 2 J
in in ro ro
X X X X
i- (_ c_ L.
CO CM CM
>0 >* ^
co co o
«" ro
o
-o
S
o o o o
ro ro ro ro
*x ^ ^x *x
in in in in
o o o o
Q. < < <
C
CO
(0 (D flJ fD CD
333^33
D) O1 Ol CD O> O>
CD CD CD - -O m m co co m co m
Of I
o i
at.
CD
8-2
A-2
-------
*J XXX
CO L. t- t-
ID TJ H-
0) co v»
_i o o>
IN- co o
ro -j- rvi
it" It" £" L" X X X ^ b
0.0.0. 0.0.0.0.0. o. a. 0.0.0.0.0.0.0.0.0.0.0. o. a. a.
L. I_ L.
a. a. a.
in
< CM
in
o -o co
O o~ CO
ro x ra x x x x
a L. ~D t_ i- L. t_
o o
0) L. fl) L. t_ 1_ L.
co a. co a. a. a. a.
NO
' rvi
c
o
CO 4-1
0)
a
O)
o>
N-I in co
eo m t^-
in«-^inin
co
co
00*0^00*
g
?
o> u
4-1 0)
CD
0 o
o
NO
in
o
- TJ
CA O
X.C
3
i_ CD
a. O)
0) 0) 0)
cc or of
CO
i
L- L. L. [_
0)
D
£
(U
C.L.COt-1-t-I-t-
Sc
ro
t- co co
c c
o o
at to
_. _. _ _i _/ ~O .
3 3 3 3 3
QJ CU >333333>>>3>>>>
i i i i i i i i i i i i i i i i
(A CO Ul
a a o
"-(ACAC/)(A(ACAC/)tf)(ACAtA(ACA"-"-"-"-(ACAC/)tOtA(ACOCACO-aocjaia:e£z3:i i a:Q£QS>->->-ZQ:orc<:oif>r<<3:3:zii 1-1 t-i >->->->- z < a
C/7COC/) 3a£OOIIIC->Ul/>(/)1I-JQ33OtIIJ'^ CJOCJCJCOCOC/>LOC/)OOaOQ^vQ£
CD 0) O
0) O +J
c E ro
0) CD O
U CO O
L. E
O -D
o >- o _i_i_i mcQcacacacacacamcami
o o
ce. i
;OOOOl-l I 1 I- I I I I I I- I I- I (-1 l-» l-l-f-l O
o ?o ocs
-2 g
CO
0)
I « |V.
A-3
-------
X X
XX XXXXXXXXXXX
X
<5 t
4J XX(OXXXXXX
(D i.t_"nt_t-t_(-t_(_
-Q o - - O .- .^ .~ .^ .^ .^ .
(0 (-L.CUC-UC-U1_(_ l_ L. t_ i_t.l-a>t.(-E.t_C-L. £_£_ L.l_f_(-i_L-t_C_l_C_L.
i a.o.c/)a.a.a.a.a.a. 0.0. a. 0.0.0.0)0.0.0.0.0.0. 0.0. 0.0.0.0.0.0.0.0.0.0.0.
><> I-- eo
rvi in ro
o
ro
§o-
in
o o
rvi m >o r>- o>
r»- o o o> o>
ON-
<\l -O
N. «- «- o O
o
~
3
C
O
CJ
vx
(0
ia
o
t
4-*
u>
o
."^
a.
u
CD V
CO 4-1
0
§
(0 u -x
0) *J O)
i C 3
U
g
U
"8
4) u
4J V
o
o
u
M O
1*
1
(U
"a.
i
ro o o
O> f\l «-
8:83
«- «- OJ
o- o> o>
o -o -o
^ ^ ^s
in in in
o o o
3o.5
a ~ u-
c
o
(A
l_
a
(0 (0 O
"3 "3 u
en OI.Q
ro o> s- o
irt r\j -o >o
Sr«- »* o
o o o-
(M
o> o- o- o>
>o >o -o >o
"-. -X X, »>.
in in in in
o o o o
S2c,2
CJ C3 - 0
C C C
to (0 (0
(0
"5* 2 5 5
O) 0) (U 01
V
§8;
O T-
00
O o>
o -o
x^ *^
in in
0 0
0. Q.
c
(0
(0
5 13
U O)
ro -o
co ro
O (M
o- o*
-o ^
^, ^,
in in
o o
So.
lJ^ tj
(D (D (D CD CD
D D D D D
D) O) O) O) D)
g
in
ro
o-
^
in
V.
in
o
o.
ID
3
in in N. o
in N. ro in
O >* -O ro
r»- ro ro o
«*«-«- (M
o o o- o-
in in in in
^ ^ *v ^x
in in in in
o o o o
o.§o.3
- 0 - C3
c
o
01
t_
(0
ra ra ra ra o
3333
(M ro
O in
(M in
co co
0> 0-
in in
v^ ^
in in
0 0
3c,
ro ro
3 3
(M N- ro ~*
O> in >o ^1-
«- o< o o-
O> (M o<
^ --S
in in
S^ N^
in in
0 0
So.
u -
to to co to
3 3 D D
V V
(MfNJw-OsO^-N-O1*
o-*mruo*fMOT-
OJCO'O^Ou^O* CO
O^O^O^ONO^O^OtO^O"^^
inmmiAinmmmintnm
NX*^N-s<^"X'*Nss^%V.>V.1^'*ss
mtninu^tnirtmir\inirtLn
ooooooooooo
c.0. 0.0. 0.0.333
M^.^«^^^MU.U_U_
O
i~
C/}
t -g
"O (D
{0(0(0(0(0 (0(0(0(0(0 (0(0
3 D~3"D 3^ DD D~3 3-2 33
O> O) D) O> D) (U 0)O)O)O)O)d> O) O)
Sii
8 8. S. 8.8.8.
ouuuoooooououu uuuuuuuouuoooouoou __
(D QJ O
0) Q. 4J
c e (/)COtf>C/) -r-n W CO W 0) CO U) W
33333333333333 O333333333333333333OOOO3333333
. , , ,OCJC/)C/)C/)-JIO2IXOI1J_I^^UO
:zZ23£3£333;u-u_miULiju-Li_u_u_u-ZD:3>
u S
O T3
I- -M
§(0 (0
>- u
o o
a; i
ororororororororororororororororo>->-oOOOi ii >-i ii
CD
|-£
(0
co
0)
U)
§s
A-4
-------
a.
a.
t
o
X X
L. 1-
5 J.5
ra "- L.
i a. a.
xxxxraxraxxxxxxxx xxxx xxxxxxxxxxxx
C_l_C.C.'Dt_'ni-t_t_Ul.t_L.l- 1-t.i-l. l-C-L.UC_i-C_i_L.C_C_(_
. -o i. i. L. i. I.
i C ni CD (o (0 nj
i O E E E E £
ii
0 (Q CD
O .~ O '~~ ~~ ~~
a.a.a.a.c/>a.c/>a.a.a.a.a.a.a-o. 0.0.0.0. 0.0.0.0.0.0.0.0.0.0.0.0.
a. a. a. a. a.
o> ^
c CM
13 *j
ra "O «*-
-
oo
£ co ro
N- in in
*- oj CM
ru
o r^ «- m co
oj »» r»-
> ro
o in <
o* in *» ro ro ro 0^*^*00
o >o -4- oo ro
o in -o o> o
-j- oj -j* o <
oj
oo
SO r- CO
'
ra t- o
o >o ro
o co vj- m' vd co in ' in co co o ^o
oooo«r
co ro o o> oj
r- o
o
o
"8
t->
a> o
4J 0)
(0
0 o
u
.,,
ininininininininminintAininin
o^a o> o^
in in in in
ininininininintninininmininin
ooooooooooooooo
c
o
o
to o
ra 0)
33
^
M
o
(A
§8 x
(A nI(D
2 3 3 > 3 > >
>>>>>>>>>>>
>>>>>>>>>>>>>>
Ul(A(/)^<'->^>^ o
cu Q. -M
c e ra
0) ra o
u - O
o
>->->->- H- I I 1 I t UUUUUUUUUIUU (JUUUZZ
>->-OO
0)
IA
32
ro
ro
A-5
-------
XX X
(OXCOXXXXXXXXXXXXXXXXXXXCOXXX XX
-Ol--OC-l-[_l_l_I-[_l_l_l-l_i_L_t-l_i_i_I_[_-Ol-.C-L. L- L.
C(OCCOCOCOCOCOCOCOn]COCOCDCOCOCDCOCOCOCOCOCCOCO(D COCO
O OEOEEEEEEEEEEeEEEEEEEeOE66 E E
.2 O 0 - o £ £
CO . >. X
coxcoxxxxx x co XX
gt--b«_t.c_i-L. t_ -g t-t-
1 § 1 1 .1 i .1 i o i i
^OItZuZElZ tZ 01 Ufl
a.u>a.a.a.o.a. a. v> a. o.
OI
ta
(0 "O H-
0) rtJ ^v.
O r- <\J r-
r- in * CM
co«-o>inroQorocoo>c\i CM «- r«-
rocoroo*i-r--coroinco«-'- co «- ro
m>oco«-r«->o'-'-
ro *! in ro CM <- ro co
ro r in - «-
rvj
o
CO t- **.
01 \*f
O
§
CJ
o co o r «- ro
*O in co N- O «~
feg
M ro
co «- o o <\i ro
ry co NO eo «- eo
:§!
^. r^ co ro ^ CM CM ^f o
ro^-ro>*oo>o & aa
r-«-«-fMCM*-O. t>O
NO CN c\j o >d* co ^
eo CM %o o o ro CM
CM ro CM ro in in vt
CM r- CM «-
g
^ ~s «V ^, «* --N "^ ~X -, ^ -X ^ >V X, X.
ininmininininininmininininininininininininmininin inin
oooooooooooooooooooooooooo oo
,
minininintninin intn inin
oooooooo o o oo
ca o
S
<
< Q- < Q. Q- Q. < Q-
50. < a. a. a. a. a.
cj < cj u cj o u
Q. < < a.
o < u- u
o
TP
g§
c
o
U)
-se
CO
CO CO
^ It
c c
CO CO
" 1- DO CO
O) O)J3 .
01 0) CO
O)O1D)O)QJ O)O)O)O)O)O)O)O)Q)D)D> O) ^ O) O> O> O) O) O> O)
od)cja)>>~ajQ>
t- CO CD L- 1-
CO CO CO CO O CO
D)O).Q.Q O)D)^ <^ d) CJ CO 4) *^ '^
CKCtJJKQ£U-U-CKQ£ IO£IA-U-
OOD.Q-Q.Q-OO.Q-Q-
CO CO - '- - - CO "- '*- "-
-UUOOOUOUUOO
CO CO CO CO CO CO CO CO CO CO CO
u u u u u o
ID CO CO CO CO CO
&
CO
I I I I I
o o o o -5 o '5 o S S 3 S S 3
3333
-------
73
C
I-*
g
CJ
+^
o
a.
t_
o
CD
t-
_l
O) ^
C CM
CO "5 £
Q) CO ^
_1 O O)
c
O *J
o
01 CU
CO 4J
o>
-l C 3
U
"S
4-J CJ
Q "o
U
(/)
-o
CO O
. 4.1
511
!
CU
I
C/)
i
i?
ro
^
in
o
o
ID CD ID CD
3333
O) O) U) D)
ci) CD OJ CD
§111
3333
CD ID CO ID
illi
S 5 5
in ro -J- ro
CO CO CO CO
gssfes
xxxxxxxx
iiiiilli
r-. CM ro ro Is- t> »rorororo ^t>t'-
ssggggggg
g^pjro^roN-co^
xxxx xxxx
gggg gggg
O^tm^O ro c\J « co
muScotNJ -^ooo*
r\j^-\oo c\j(\j(\JO
sJ-O«-0 «-OOO
«-ir»c\J*O ro>OOJ^-
-J- ^ in m oo
O^ « in in Nt
o NO »o « *O
S C> 5 N ^0
ro
OOOO O*O*«OI'O
rororom rorororo
^ *x. ^s *v. 1*x,'^11^>sv.
mmmm IA tn ir» in
oooo oooo
< < 5 LJ CJOCJCJ
*
t
T3
(GfQ(fJ(Q{U(!J(QtQ(D
333333333^
D>a>o)cncnoiDiO)Di(i>
acafoccea^occcocafu.
§ §§§§§§
30)0)0)333333
UQ.Q.Q.U U U U O L)
>333»»»
C/){/l(/)(AC/)(/)(/)(0(/)Ul
3333333333
owO«-N-OJ«
r^Qr^QQQQQf^^t
M^^aoo-.-jMro.j
xxxxxxxxx
g.il.ii.iili
*j- o ro »4- m oo
inc>2^°coSiS
ofyrooo>oinpNN.
f\lO»r*-N--4-CM'«4iO1>rO
^ >^ r*- ^ *Q o ro co o^
^
ooooooooo
roforororoior^ii'r^
~,^^^^N.V^^-x,
ininininminininin
ooooooooo
OOCJUUUUULJ
IDIDIDIDCOCDCOIDCOCD
3333333333
o>u)D)D)cno)cnD)cnD>
(DQjajOQJCPOQJtUQJ
Q£ y O£ (* G£ ^ tt fv e^ F*
§§§§§§§
U U U U O U O
>>>>5>>
3333333000
CLuocncnfjfjoalo
in ho *^ ro ""tf1 *"~ *~ in ^~ ro
ro ro ro ro ro ^~ >~ o O O
CD CD CO CD CO LLJ UJ U- M^ U-
«-t?2SSiSRicM
XX X
t_ L. C.
EEEOEOEEEEO
r»- in
r-. o
in
!n|0c>oco°oo rorocM
cMOvrocor^^ o«->o
ro ro 5 c\i 5 ro ro S rvl «
« T
ooooooooooo
«coooooooeo«eooocooo
~.1^~v-^~v~v1xvx'xs^^
ininininininininininin
ooooooooooo
U(ju5<->5uuuu5
S
1 §§ §
J3 (AC/) (A
1 .r- It
0) 1- t- C I-
cora cocooOC°cocoO
3353D0(J533°
o>o)cu cna>^2^ d-^-N± «%O«-
LJ CJ CJ CJ CJ CJ CJ CJ t~ H" h~
r»-Kicoinmc>ooo«-(M
CMCMCMCMfMCMCMrOOOO
en
o
A-7
-------
^
"g
g
4-1
O
(_>
xw
ID
ID
O
t
4-*
cn
O
o.
X
o
4-»
ID
1
__l
?£
~O 4^
ID TJ <«-
a
0
§
£ XN
Si. *x,
4-> O>
_1 C 3
0) %_/
o
§
u
"S
*J
2 8
ID '
Q
O
U
(A
« O
+3
C £
<
«
fl
1-
11)
1
X X X X
L. L. L. t_
SEES
a. o. a. a.
o- co
r^ m
f^ -o
(M (M
CO «- ON O
«- >* ~» ro ^*
~* <\i ro ^t
vo «- >* ro
o o> ox o
S** ^ **N *^
co co co co
s^ Ss^ ^ *^
in in in in
0000
<
O. ^ O_ Q.
(J U- O CJ
*~* ««
Sc
ro
(D ID ID (
O) CD « %O>4'O*^~i/^o tri PO ""tf1 "tf1 >* o in in >O -sjfvji/ifvj
«~«j-mN.^omc\ioo N- K-N-m N-m-^o^o* oiooo^o
N-ro« « oooroo "41 s^oo coc\i^>ofvj ^~N-N-ro
coo «-fNjr\j«-f\j st rorM N^o-^co
o -4- m m K- «- t-
M ro >o
IO ^ (\| f^ fx. o OQ N- f\J 00 O ^~ N- *Q 00 O rO 00 O C3 O^ O O O^
IOl/>vOlOO>lACMCO O CM^ tON-O>O*^~tO ^OCOONa-ON-*"
irt ^ *~ ^~ CNJ ^ ^o ro w ^"~ ro to ro ^~ c^ «~~ « ir\ t^ co ^ o*
0
oooooooo o^ o-o-o-oooooooo-ooooooo.
m oo co co co co co a co co eo oo eo eo co eo « eo co co co oo co co eo oo «
^X^^^^^X^X^X^ ^x ^XxXxXx^Xx^xX.*XxsXx*Xx*xs*Xx^ XXx'XxXx^
ininintnininmin in ininininininininininminintninininin
oooooooo o oooooooooooooooooo
3333 3
O U " LJ U O C_) "-i CJ " " CJ CJ LJ U CJ U ^ 3 3 U- CJ CJ CJ LJ CJ U
<1)
.2
C/J
A-
i) c c c
"XJ (D (D (D
Ji)i,SSxii)lSc?,i;roi?)Sc?,i,i?,i,gi>i,i,o>??li,i,i,i1i,i)
X X
1§
^ J>
ro ro
>o ro
S
tn
o* o^
co co
*x, -Xx
in in
o o
3 u
ID ID ID
0.0, a,
^
§11
ooooouooooooooooouoou
rJDDOaJ(DDDDZ3D3 CD 3 OJ D
OUUQ.Q.Q.UUUUUU Q.UQ.O
tO (O tO < «f to (O tO to tD tO '^ to '^* CD
(D ttl O
gfr'S
E a n
o >- o
o o
a: i
10.0.0.0.
i2
o
00
A-8
-------
50272-101
REPORT DOCUMENTATION
PAGE
1. REPORT NO.
EPA 747-R-93-007
3. Recipient's Accession No.
4. Title and Subtitle
Comprehensive Abatement Performance Pilot Study Volume I: Results of Lead Data Analysis
5. Report Date
February 1995
6.
7. Author(s) Bruce Buxton, Steve Rust, John Kinateder, David Burgoon, Fred Todt, Gary Dewalt
8. Performing Organization Rept. No.
9. Performing Organization Name and Address
10. Project/Task/Work Unit No.
G301104-05
Battelle Memorial Institute
505 King Avenue
Columbus, Ohio 43201-2693
and
Midwest Research Institute
425 Volker Boulevard
Kansas City, Missouri 64110
11. Contract(C) or Grant(G) No.
(C) 68-D2-0139
(G)
12. Sponsoring Organization Name and Address
U.S. Environmental Protection Agency
Office of Pollution Prevention and Toxics
401 M Street, S.W.
Washington, D.C. 20460
13. Type of Report & Period Covered
Final Report
14.
15. Supplementary Notes
16. Abstract (Limit 200 words)
This report presents the results from the pilot study that preceded the Comprehensive Abatement Performance Study. The goal of the Comprehensive
Performance Study was to assess the long-term impact of lead-based paint abatement. The pilot study was conducted to test the sampling and analysis
protocols that were intended for the full study. These protocols called for determining the levels of lead in dust and soil samples collected at residential
units. The pilot study was conducted at six houses, and all steps that were planned for the full study were included in the pilot.
The major finding of the pilot was the difference between wipe and vacuum methods for collecting dust. The choice of method had a noticeable
impact on the level of lead associated with the collected sample. In addition, an inter-laboratory comparison of dust and soil samples indicated no systematic
difference in lead levels between the two laboratories. Also, intra-laboratory comparisons of sample results by inductively coupled plasma-atomic
absorption spectrometry (ICP) and the more sensitive graphite furnace atomic absorption spectrometry (GFAA) indicated good agreement within the common
domain of instrument detection limits
Estimates of random house-to-house, room-to-room, and side-by-side sample variability were obtained for most of the sample types in the study.
These estimates were used for determining the number of houses and number of samples per house for the full study.
17. Document Analysis
a. Descriptors
Lead, Lead-Based Paint Abatement, Statistical Analysis
b. Identifiers/Open-Ended Terms
HUD Abatement Demonstration, Encapsulation, Enclosure, Removal
c. COSATI Field/Group
18. Availability Statement
Release Unlimited
19. Security Class (This Report)
Unclassified
20 Security Class (This Page)
Unclassified
21. No. of Pages
158
22. Price
(See ANSI-239 18)
OPTIONAL FORM 272 (4-77)
(Formerly NTIS-35)
Department of Commerce
------- |