nelie
Technology To Work
REPORT
Draft Final Report
COMPREHENSIVE
ABATEMENT
PERFORMANCE PILOT
STUDY
VOLUME II
MULTI-ELEMENT DATA ANALYSES
To
Office of Pollution Prevention and Toxics,
U.S. Environmental Protection Agency
March 1, 1996
-------
Review Draft - Do Not Cite or Quote
March 1, 1996
REVIEW DRAFT
DRAFT FINAL REPORT
COMPREHENSIVE ABATEMENT PERFORMANCE PILOT STUDY
VOLUME II: MULTI-ELEMENT DATA ANALYSES
Prepared by
. BATTELLE
505 King Avenue
Columbus, Ohio 43201-2693
and
Midwest Research Institute
425 Volker Boulevard
Kansas City, MO 64110
for
Technical Programs Branch
Chemical Management Division
Office of Prevention, Pesticides, and Toxic Substances
Office of Pollution Prevention and Toxics
U.S. Environmental Protection Agency
Washington, DC 20460
-------
Review Draft -- Do Not Cite or Quote
U.S. EPA DISCLAI1 R
This document is a preliminary draft. It has not been
released formally by the Office of Pollution Prevention and
Toxics, U.S. Environmental Protection Agency. It is being
circulated for comments on its technical merit and policy
implications.
This report was prepared under contract to an agency of the
United States Government. Neither the United States Government
nor any of its employees, contractors, subcontractors, or their
employees makes any warranty, expressed or implied, or assumes
any legal liability or responsibility for any third party’s use
of or the results of such use of any information, apparatus,
product, or process disclosed in this report, or represents that
its use by such third party would not infringe on privately owned
rights.
Publication of the data in this document does not signify
that the contents necessarily reflect the joint or separate views
and policies of each sponsoring agency. Mention of trade names
or commercial products does not constitute endorsement or
recommendation for use.
BATTELLE DISCLAThThR
This is a report of research performed for the United States
Government by Battelle. Because of the uncertainties inherent in
experimental or research work, Battelle assumes no responsibility
or liability for any consequences of use, misuse, inability to
use, or reliance upon the information contained herein, beyond
any express obligations embodied in the governing written
agreement between Battelle and the United States Government.
-------
Review Draft -- Do Not Cite or Quote
AUTHORS AND CONTRIBUTORS
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 and key staff
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 team leaders
for the field teams, for auditing the field teams, 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. Key staff included: Bruce Buxton,
Steve Rust, Tamara Collins, Fred Todt, John Kinateder, Nancy
McMillan, Matt Palmgren, Nick Sasso, Robin Hertz, and Casey
Boudreau.
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 designing
and producing a vacuum device for collecting field samples, for
developing training for 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. Key staff included: Gary Dewalt,
Paul Constant, Jim McHugh, and Jack Balsinger.
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 audit reports, for
reviewing the final report, and for arranging the peer review of
the design and the final report. The EPA Work Assignment
Managers were Ben Lim and John Schwemberger. The EPA Project
Officers were Gary Grindstaff and Joe Breen.
-------
Review Draft -. Do Not Cite or Quote
TABLE OF CONTENTS
Page
EXECUTIVE SUMMARY . . . v
1.0 INTRODUCTION . 1
1.1 Study Design 1
1.2 Data 2
2.0 ANALYSIS RESULTS 4
2.1 Characterization of Element Levels 5
2.2 Abatement and Renovation Effects 14
2.2.1 Abatement and Renovation Effects
By Element 14
2.2.2 Abatement and Renovation Effects
Across Elements 25
2.3 Relationships Among the Elements 29
2.3.1 Bivariate Relationships
(Correlations) 29
2.3.2 Multivariate Relationships
(Principal Components) 33
3.0 CONCLUSIONS . . . 36
4.0 RECOMMENDATIONS FOR FUTURE WORK . . . 38
5.0 REFERENCES . . . . 39
LIST OF TABLES
Table 1 Abbreviations for Sample Types Used in
Tables and Figures 4
Table 2 Abatement and Renovation History by House 5
Table 3 Results of Analysis of Variance to Test
for Significant Differences Among
Sample Types, by Element 13
Table 4 Geometric Mean Concentration and Log Standard
Deviation Across Houses by Sample Type 15
1
-------
TABLE OF CONTENTS
(continued)
Page
Table 5 Model Estimates and Log Standard Errors
of Geometric Mean Concentrations in
Unrenovated, Unabated Houses 17
Table 6 Estimates and Log Standard Errors
of Multiplicative Renovation Effects . . . 18
Table 7 Estimates and Log Standard Errors
of Multiplicative Abatement Effects . . . . . . . 19
Table 8 Principal Components for Unit Mean
Log-Concentration by Sample Type 27
Table 9 Principal Components for Unrenovated,
Unabated Home Averages, Abatement Effects,
and Renovation Effects . . . 35
LIST OF FIGUPES
Figure la Lead Concentration vs. Sample Type
(Geometric House Mean) 6
Figure lb Aluminum Concentration vs. Sample Type
(Geometric House Mean) 6
Figure ic Barium Concentration vs. Sample Type
(Geometric House Mean) 7
Figure id Cadmium Concentration vs. Sample Type
(Geometric House Mean) 7
Figure le Calcium Concentration vs. Sample Type
(Geometric House Mean) 8
Figure if Chromium Concentration vs. Sample Type
(Geometric House Mean) 8
Figure ig Magnesium Concentration vs. Sample Type
(Geometric House Mean) 9
Figure lh Nickel Concentration vs. Sample Type
(Geometric House Mean) 9
Figure 11 Potassium Concentration vs. Sample Type
(Geometric House Mean) 10
ii
-------
TABLE OF CONTENTS
(continued)
Page
Figure lj Titanium Concentration vs. Sample Type
(Geometric House Mean) 10
Figure 1k Zinc Concentration vs. Sample Type
(Geometric House Mean) 11
Figure 2a Block Chart of Estimated Average Log-
Concentration in Unrenovated, Unabated
Units for Window and Air Duct Dust Samples . . . . 21
Figure 2b Block Chart of Estimated Average Log-
Concentration in Unrenovated, Unabated
Units for Floor and Textile Dust Samples 22
Figure 2c Block Chart of Estimated Average Log-
Concentration in tinrenovated, Unabated
Units for Soil Samples 23
Figure 3a Plot of First Two Principal Components
of Mean Log-Concentrations for Dust Samples . . . . 28
Figure 3b Plot of First Two Principal Components
of Mean Log—Concentrations for Soil Samples . . . . 28
Figure 4a Window Channel House Mean Correlation
Scatterplot 30
Figure 4b Window Stool House Mean Correlation
Scatterplot 30
Figure 4c Air Duct House Mean Correlation Scatterplot . . . . 30
Figure 4d Floor House Mean Correlation Scatterplot 30
Figure 4e Bedcover/Rug/tipholstery House Mean
Correlation Scatterplot 31
Figure 4f Entryway Dust House Mean Correlation
Scatterplot 31
Figure 4g Entryway Soil House Mean Correlation
Scatterplot 31
Figure 4h Foundation Soil House Mean Correlation
Scatterplot 31
iii
-------
TABLE OF CONTENTS
(continued)
Page
Figure 4± Boundary Soil House Mean Correlation
Scatterplot 32
Figure 5 Key to Relation Between Shape of Ellipse
and Observed Correlation in Figures
4a Through 4i 32
Figure 6 First Two Principal Components for Each
Building Component, Plotted Versus Each
Other for Unrenovated, Unabated Unit Mean
Log-Concentrations, Renovation Effects, and
Abatement Effects 38
LIST OF APPENDICES
Appendix A Summary of Multi-Element Data A-i
Appendix B Outlier Analysis for the CAPS
Pilot Multi—Element Data B-i
iv
-------
E CUTXVE SU)* AR!
This report presents the results of a statistical
analysis of multi—element data collected during the pilot study
that preceded the Comprehensive Abatement Performance (CAP)
Study. The goal of the CAP 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. For the multi-element analysis,
concentrations of lead as well as those of the following ten
elements were measured in the study: aluminum, barium, cadmium,
calcium, chromium, magnesium, nickel, potassium, titanium, and
zinc. The data used in this analysis consisted of these element
concentrations in indoor dust and soil sample types (e.g., floor
dust and entryway soil) collected from six houses. In addition
to lead, the elements barium, cadmium, chromium, titanium, and
zinc were selected for analysis in this study due to their common
use in the composition of paint.
The multi-element approach was undertaken with the
perception that these elements might provide a “tracer” for dust
samples which may have their origin in paint. Aluminum, calcium,
magnesium, nickel, and potassium were selected as tracer elements
for dust samples which may have their origin in soil. Soil has
been suggested as a reservoir holding lead from gasoline
emissions and lead—based paint.
The multi—element statistical analysis presented in
this report addresses the following three study objectives:
(1) Characterize the concentration of lead, aluminum,
barium, cadmium, calcium, chromium, magnesium, nickel,
V
-------
potassium, titanium, and zinc in household dust and
soil,
(2) Characterize the relationships between renovation and
abatement history and the concentrations of these
elements in household dust and soil, and
(3) Investigate the relationship between concentrations of
these elements in the different types of samples
collected (i.e., samples of different media and taken
from different locations), and compare these
relationships across sample types.
An examination of descriptive statistics including
geometric mean measures provided insight into the first study
objective. Analysis of variance was primarily used to address
the second objective, regarding the impacts of abatement and
renovation. Multivariate statistical techniques, including a
correlation analysis and principal components analysis, were
employed to address the third objective.
The highest geometric mean concentrations were observed
for calcium in the indoor dust samples. For the outdoor soil
samples, the highest levels were observed for aluminum.
Titanium and aluminum concentrations averaged higher in soil
samples than in dust samples, and were also higher in entryway
dust samples than in floor dust samples.
Except for lead, there were no strong differences
observed between levels of the different elements measured in
abated and unabated houses.
The strongest correlations in concentrations of
different elements were observed in soil samples. Indoors,
correlations were less clear. Of the elements measured besides
lead, zinc was most frequently correlated with lead, both within
sample types and across sample types. The concentrations of the
elements in the following groups of sample types were found to be
similar: (i) entryway, foundation, and boundary soil samples;
vi
-------
(ii) floor, entryway, bedcover/rug/upholstery, and window stools,
and (iii) air ducts and window channels.
In general, the sparse data available for use in this
analysis prevented conclusive source assessment of lead in
household dust. This being only a pilot study, strong
relationships were not anticipated. The full CAP Study in which
similar and additional data were collected from 52 houses would
provide nine times as much information to address this issue.
vii
-------
COMPREHENSIVE ABATEMENT PERPORMP NCE PILOT STUD!:
MULTI-ELEMENT DATA ANALYSES
1.0 INTRODUCTION
This report presents the results of a multi—element
analysis of data obtained during a pilot study that preceded the
Comprehensive Abatement Performance (CAP) Study. This represents
Volume II of the CAP Pilot report. Volume I dealt exclusively
with the statistical analysis of observed levels of lead (US EPA,
1995) . The goal of the CAP 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.
1.1 STUDY DESIGN
In the CAP Pilot study, six houses of differing abatement
history were sampled. These houses were located in Denver,
Colorado. Two houses were abated by enclosure or encapsulation
methods, two were abated by removal methods, and two were control
houses identified as being free from lead-based paint by previous
XRF-testing. Along with the determinations of lead obtained in
the study, levels of ten other metals were measured within dust
and soil samples taken at these houses: aluminum, barium,
cadmium, calcium, chromium, magnesium, nickel, potassium,
titanium and zinc. Five of these metals (barium, cadmium,
chromium, titanium and zinc) are known to be used in the
composition of paint. The other five elements are known to be
present primarily in other sources. For example, magnesium is
found in clay which is often observed in soil samples. The
purpose of measuring the levels of these other metals in the
samples was to identify groups of sample types that appear to
-------
have come from similar sources with the ultimate goal of
identifying prominent sources of lead found in household dust.
The major objectives addressed in the analysis of the
multi-element data from the pilot were to:
(1) Characterize the concentrations of lead, aluminum,
barium, cadmium, calcium, chromium, magnesium,
nickel, potassium, titanium, and zinc in household
dust and soil,
(2) Characterize the relationships between renovation
and abatement history and the concentrations of
these elements in household dust and soil, and
(3) Investigate how the concentrations of these
elements are related, specifically elements in
household dust and exterior soil, air ducts, and
bedcover/rugs /upholstery.
The intention of this examination was to identify
analysis methods for evaluating multi—element data, and to apply
these methods to pilot study data to identify any strong
relationships. With data available for only six housing units,
most relationships were not strongly detectable.
The following subsection describes the data and a
summary of the outlier analysis performed. Section 2 describes
the analyses performed and their results. Section 3 provides
conclusions. As this analysis was performed on data from a
pilot study, Section 4 describes potential future work which
could be done to extend these results.
1.2 IA
A total of 109 regular dust samples were collected in
the CAP pilot study by vacuum methods. By design, 18 dust
samples were to be collected from each house. For each of the
six dust sample types (window channel, window stool, air duct,
floor, bedcover/rug/upholstery, and entryway), two different
2
-------
rooms were targeted for sampling. Forty-eight (48) soil samples
were collected. Soil samples were to be collected at six
different locations at each house: one just outside the front
and back entryways, two at different locations on the foundation,
and two at different locations on the property boundary.
The data provided for multi-element analysis were
obtained for most of the vacuum dust and soil core samples
collected during the pilot study. Only element concentrations
(/2g/g) were analyzed for this report. Element loadings (/Ag/ft 2 )
were also measured for dust samples, but were riot considered in
this analysis.
Table 1 contains a description of the acronyms used in
tables and figures to denote building components (referred to as
“sample types”) . Listings of the raw element concentration data
are displayed in Tables A-la through A-if of Appendix A by house.
For each house, concentrations for each of the eleven elements
are displayed by sample medium, building component, location, and
sample ID. House number and sample ID uniquely identify each
sample. When samples had determinations above the calibration
range for a particular element, the maximum detectable
concentration for the element is reported. (The maximum
detectable concentration depended on the dilution factor of the
sample.) These samples are identified by a superscript “a” in
the Appendix tables.
Univariate and multivariate outlier detection tests
were applied to these data. The outlier analyses and the results
of these analyses are documented in Appendix B. Lists of the
potential outliers were sent back to the laboratory for
verification. Most results for these samples were confirmed as
originally reported. The one instance where an error was
reported is documented in the footnotes to Table A-lb.
3
-------
Table 1. abbreviations for Sample Types Used in Tables and
Figures
Media
Mnemonic
Component
Vacuum Dust Samples
ARD
BRU
EWY (-I)
FLR
WCH
WST
Air ducts
Bed/Rug/Upholstery
Entryway (—Inside)
Floor
Window Channel
Window Stool
Soil Samples
BDY
EWY (—0)
FDN
Boundary
Entryway (-Outside)
Foundation
Results for seven dust samples were excluded from the
analyses. One of these samples was dropped in the laboratory
(sample 12 in house 19) . Sample 7 in house 19 was excluded from
the analysis since the cadmium result was greater than the upper
calibration limit of the measuring instrument. For four of these
remaining five samples, only lead concentrations were available
due to calcium interference: samples 9, 3, and 17 in house 19,
and sample 19 in house 43. Finally, sample 12 in house 51 was
excluded due to sampling problems; the cartridge filled with
sawdust prior to completion of the sample collection.
2.0 ANALTSIS RESULTS
The analysis is divided into three parts corresponding
to the three major objectives introduced above. Section 2.1
contains a characterization of the concentration levels of the
different elements in the various sample types. Section 2.2
describes the estimated effects of abatement and renovation, and
Section 2.3 examines the relationships among the elements and
sample types.
4
-------
2.1 CHARACTERIZATION OF ELE) NT LEVELS
Due to the general lack of room level effects found in
the analysis of the CAP pilot lead data, the basic experimental
unit considered in the multi-element data analysis is the house.
House geometric mean concentrations of the eleven elements were
the basic quantities used in the statistical analyses. These are
tabulated in Table A-2 of Appendix A by sample type and house for
each of the eleven elements. For easy reference, Table 2
displays the abatement and renovation history of each of the six
houses sampled. (Renovation is described later.)
Table 2. Abatement and Renovation History by House
House
Interior
Abatement
History
Exterior
Abatement
History
Renovation
17
19
33
43
51
80
Abated: Removal
Unabated
Unabated
Abated: Removal
Abated: E/E
Abated: E/E
Abated: E/E
Unabated
Unabated
Abated: Removal
Abated: Removal
Abated: E/E
Partial
Full
Levels of each of the eleven elements observed varied
by sample type. Figures la through 1k display geometric mean
sample concentrations by house and building component for lead,
aluminum, barium, cadmium, calcium, chromium, magnesium, nickel,
potassium, titanium, and zinc, respectively. These figures
display all the data considered in the analysis. Mean sample
concentrations for each house are plotted with different symbols.
The grand means over all houses are plotted with a circle and
connected by a solid line across sample types. Note that the
last three sample types in each plot represent soil samples,
while the other sample types represent dust samples.
5
-------
0
..
8
10000
1
100
10
*
z
z
V
x
+ +
,
Sample lype
House L.e er4 + + + 17 ) X 19 S 33 0 0 0 43 V V V 51 Z Z 80 00 99
Figure la. Lead Concentration vs. Sample Type (Geometric House Mean).
100000
jQ 0
C 10000
C
8
E
E
1000 ______________________________
Sample Type
House (.egond + + 17 K K K 19 33 0 0 0 43 V V V 51 Z Z 7 80 °°° 99
Figure lb. Aluminum Concentration vs. Sample Type (Geometric House Mean).
V
V
V
x
V
*
* * +
* +
0
*
0
V
z
6
-------
C
0
.
8
E
x
- x + a
w V
t $ / /
Sample Type
House Legend 17 ‘C ‘C ‘C 19 • S 33 0 0 0 43 V V V 51 2 2 2 80 000 99
Figure ic. Barium Concentration vs. Sample Type (Geometric House Mean).
— 1000
+
o +
100 + I-
1: N
, 4 4 /O
Sarriple Type
Hou9e Legend 17 ‘19 - 33 C 0 43 51 = = 80 99
Figure id. Cadmium Concentration vs. Sample Type (Geometric House Mean).
*
z
+
V
0
0
+
+
x
i o u
z
x +
x
7
-------
1000000
3
C 100000
0
.
icooo
E
0
.
o 1000
4 , ’
Sample Type
House Legend ‘17 x X X 19 33 0 0 0 43 V V V 51 2 2 Z eo - - 99
Figure le. Calcium Concentration vs. Sample Type (Geometric House Mean).
1000
C
0
. * *
100 z x
#
Sample Type
House Legend 17 X X 19 33 0 0 0 43 V V V 51 Z Z Z 80 e-e-e 99
Figure if. Chromium Concentration vs. Sample Type (Geometric House Mean).
V
x
+
V
8
-------
+
Sample e
HouseLegend 17 xxx 9 *5533 0D043 VYY 5 1 22250 e.e-e99
Figure 1g. Magnesium Concentration vs. Sample Type (Geometric House Mean).
1
C
2 100 +
‘ °
Sample Type
House Legend 17 x X X 9 S 33 fl 0 0 43 Y V V 51 2 Z 2 80 99
Figure lh. Nickel Concentration vs. Sample Type (Geometric House Mean).
0
x
z
z
1
z
a
0
9
-------
lcEjocP3
1 xc +
0
+ +
1 _______________________
Sample l .pe
House Legend 17 x X X 19 • S 33 0 0 0 43 Y Y 51 Z 2 Z 80 a-0a 99
Figure ii. Potassium Concentration vs. Sample Type (Geometric House Mean).
C
0
10,, ,t 4 /’° —
Sample jpe
House Legend + 17 X X X 19 * * 33 0 0 0 43 V V V 51 Z Z 2 80 -- 99
Figure lj. Titanium Concentration vs. Sample Type (Geometric House Mean).
0
x
*
0
*
0
+
100
+
10
-------
0
.-
.
100000
10000
1
Ilyt
*
*
a
z
V
F,
Sample Type
House Legend 17 x X x 19 • 33 0 0 0 43 ‘ 51 Z Z Z 0-0-0 99
Figure 1k. Zinc Concentration vs. Sample Type (Geometric House Mean).
z
0
x
z
*
1].
-------
As can be seen in the figures, the highest geometric
mean concentrations were observed for calcium in the indoor
samples. For the outdoor samples, the highest levels were
observed for aluminum. Of the different components sampled,
lead concentrations were highest in air duct, window stool, and
window channel samples (Figure la) . Levels of barium and zinc
appear to be similar to levels of lead across sample types. On
average, titanium was the least variable of the eleven elements
within each sample type. For each of the elements except
titanium and aluminum, dust samples taken from floors and
entryways had similar concentrations. However, the
concentrations for these two elements were higher for entryways
on average than for floors. The levels of aluminum and titanium
observed in entryway dust samples were more consistent with those
observed in soil samples. This could indicate that soil is being
tracked into the homes through the entryways, or it could be a
reflection of the presence of these elements in the construction
of the entryways. Levels of aluminum and titanium were also high
in window channel samples.
The highest mean chromium concentration was observed in
house 33 for regular floor dust samples and entryway dust
samples. The entryway soil samples from this house also had high
chromium concentrations. The individual dust and soil samples
obtained from the back side entrance to this home and a floor
dust sample collected from the adjoining room (kitchen) had the
highest chromium concentrations observed in the study (see Table
A-lc), suggesting a possible relationship between exterior and
interior chromium levels.
In regard to zinc, there were two exceptionally high
zinc concentrations observed on window channels and one high
concentration on a window stool. However, each of these three
measures came from different houses.
12
-------
When grouping the profiles in Figures la through 1k
based on similarity, three groups of elements are formed. Lead,
barium, and zinc seem to have similar contours and comprise one
group. Aluminum and titanium make up a second group, while
cadmium, calcium, and chromium make up the third group.
To quantify the degree of variation in the
concentrations of each element across sample types, an analysis
of variance was performed on the geometric means plotted in
Figures la through 1k. The results of this ANOVA are summarized
in Table 3. For all elements except potassium and chromium, the
differences across sample types were statistically significant.
The strongest differences were seen for magnesium and calcium,
with lower levels observed in soil than in dust.
Table 3. Results of Analysis of Variance to Test for Significant
Differences Among Sample Types, by Element
1enent
Root sean
Squared
Error
E’ value
P value
Comment
Pb
1.14
4.47
.0006
Al
0.48
6.55
.0001
Ba
0.95
3.83
.0019
Cd
1.01
5.54
.0001
Ca
0.67
9.71
.0001
Soil all lower than dust
Cr
0.78
1.59
.1570
Insignificant differences
Mg
0.48
31.27
.0001
Soil all lower than dust,
EWY lower than FDN
Ni
0.77
4.83
.0003
K
0.74
0.55
.8096
Insignificant differences
Ti
0.38
8.44
.0001
Zn
0.76
16.40
.0001
ARD, WST, WSL higher than
rest
13
-------
Also included in Table A—2 are indicators of the
primary method of interior and exterior abatement for each house.
A “U” indicates that no abatement was performed in the house
because no significant lead-based paint was present, an “R”
indicates that the house was abated primarily by removal methods,
and an “E” indicates that the house was abated primarily by
encapsulation! enclosure methods. Table A-2 also contains the
number of samples for which concentrations were determined for
all eleven elements. Any sample in Tables A-la through A-if for
which at least one element had a missing value was not included
in the Table A-2 summary.
Grand geometric mean concentrations for each element by
sample type are displayed in Table 4. These were obtained by
taking the geometric mean of the entries in Table A—2 across all
houses for each sample type and element. Thus, each house where
a sample was taken (for a particular sample type) is given equal
weight in these averages. These grand means are plotted in
Figures la through 1k by the circles connected by a solid line.
Each geometric mean is followed by its log standard deviation.
This represents a measure of the between—house variation for that
response without controlling for abatement or renovation effects.
2.2 ABATE NT P ND RENOVATION EFFECTS
2.2.1 Abatement and Renovation Effects by Element
The impact of abatement and renovation on the multi-
element data was assessed by fitting a statistical model
containing both renovation and abatement effects to the data in
Appendix A. The model fitted to data for each element was
C = m + aI + rR + E j = 1,...,6
14
-------
Ui
Table 4. Geometric Mean Concentration and Log Standard Deviation Across
Houses by Sample Type
No of Dust
Sample Sample Units Loading
Medium Type Sampled (mg/fl 2 )
Lead
Aluminum
Barium
Cadmium
Calcium
Chromium
Gao Mean Log Std.
(pglg) 0ev.
Geo Mean Log
(pg/g) Std Dev
Geo Mean Log
(pgFg) Std. Day.
Gee Mean Log
(pg/g) Std Dev
Gao Mean Log
(pglg) Sid 0ev
Geo Mean Log
(pg/g) Std. Dev.
Dust
WCH
WST
ARD
FIR
BRU
EWY-l
4
6
5
6
5
6
73783
4683
35203
5826
4155
7184
2128 097
658 120
771 031
260 081
152 072
314 091
12940 039
6266 036
7136 032
6331 030
6248 047
10761 0.37
1647 1.58
703 116
325 060
295 052
254 045
294 078
19 1 061
239 1.03
263 1.32
93 068
97 0.62
95 049
33730 023
53230 051
40465 061
25042 044
24598 051
32709 103
40.1 0.46
543 054
773 064
48.7 080
550 052
454 079
Soil
EWY-O
FDN
BOY
6
6
6
208 090
209 087
126 079
16058 033
14491 0.40
11373 042
276 021
257 031
166 031
56 085
40 041
2.8 051
9814 040
9812 031
8576 020
40.8 067
287 028
236 031
Sample
Medium
Sample
Type
No of
Units
Sampled
Oust
Loading
(mg/fl 2 )
Magnesium
Nickel
Potassium
Titanium
Zinc
Gao Mean Log
(p9/9) Std Dev
Gao Mean Log
(pg/g) Std. Dew
Geo Mean Log
(p 919) Std. 0ev
Geo Mean Log
(p9/9) Std. Day.
Geo Mean Log
(pglg ( Std 0ev
Oust
WCH
WST
ARD
FIR
BRU
EWY-l
4
6
5
6
5
6
737 83
4683
35203
5626
4155
71 84
5553 0 32
4807 029
3877 042
3222 025
3094 029
4419 040
24.0 035
380 037
407 117
278 060
450 102
207 0.36
2651 0444
2818 067
4260 036
4311 070
4046 089
4045 0.67
496 0 27
376 013
262 038
199 029
191 057
351 0 33
3226 1 07
1939 066
4458 098
770 039
656 070
722 049
Soil
EWV-0
FDN
BDY
6
6
6
574 016
1054 066
636 039
139 074
114 027
97 030
4069 026
3476 032
3504 033
482 023
421 024
372 026
296 037
372 035
178 046
-------
where
C represents the observed (arithmetic) average log-
concentration in house j,
m represents the average log-concentration in
unrenovated unabated houses,
a represents the added effect of abatement,
I = 1 if house j was abated
0 if house j was an unabated house,
r represents the added effect of a full renovation,
R is the degree of renovation house j was undergoing
at the time of sampling (see below), and
E represents house—to-house variation
[ louse 51 was assigned an R) value of 1 indicating “full
renovation” and House 19 a value of 0.5 indicating “partial
renovation”. The other four houses were assigned R values of
zero, indicating that no renovation was being performed.
In the analysis of the lead data, the method of
abatement (E/E or removal) was also considered as a factor in the
statistical model. No significant effect was found; and
therefore, this effect was not included in the above lead model
applied to all elements.
Estimates of the model parameters are reported in
Tables 5, 6, and 7. Table 5 contains estimates and log—standard
errors of the geometric mean concentration of each element in
unrenovated, unabated houses, by sample type. Tables 6 and 7
contain estimates and standard errors of the multiplicative
effects of renovation and abatement, respectively, by sample
type. In Tables 6 and 7, a multiplicative effect of 1.0 implies
no effect. A multiplicative effect less than one indicates that
lower levels were observed in renovated (abated) houses, while a
multiplicative effect greater than one indicates that higher
16
-------
I-i
—3
Table 5. Model Estimates and Log Standard Errors of Geometric Mean
Concentrations in Unrenovated Unabated Houses
Sample
Medium
Sample
Type
,
Houses
Lead
Aluminum
Barium
Cadmium
Calcium
Chromium
Gao Mean Log Std.
ljig/g) Err
Gao Mean Log
(jigIgl Std. Eu
Gao Mean Log
lpglg) Std. Err.
Gao Mean Log
(pg/g) Std. Err.
Gao Mean Log
(pglg) Std Err
Geo Mean Log
lpg/g) Std. Err.
Dust
WCH
WST
ARD
FLA
BRU
EWY-l
4
6
5
6
5
6
7238 0 64
226 1 17
875 041
102 033
117 045
96 019
13346 0 54
5808 039
5341 036
7687 030
11954 039
14146 034
7058 1.95
478 1 11
216 068
313 031
163 049
255 0.47
29 7 0 07
214 089
360 184
191 059
254 034
130 057
34866 0.37
57057 012
53114 068
20998 025
18230 018
25873 035
39 0.72
87 046
46 069
141 036
69 012
109 063
Soil
EWY-O
FON
BDY
6
6
6
63 043
102 0.89
53 081
22668 010
18568 033
11492 0.44
261 015
252 037
128 030
39 081
36 048
21 057
13126 036
13395 027
9977 025
60 068
32 018
21 033
Sample
Medium
Sample
Type
#
Houses
Magnesium
Nickel
Potassium
Titanium
Zinc
Geo Mean Log
(pglg) Std Err
Gao Mean Log
(pgig) Std. Err.
Gao Mean Log
(pglg) Sici. Err.
Geo Mean Log
(pg/g) Sid Err.
Gao Mean Log
(p919) Std Err
Dust
WCH
WST
ARD
FLR
BRU
EWY-l
4
6
5
6
5
6
4237 045
4501 035
2719 050
3337 025
3558 039
4400 0 23
179 050
313 041
352 027
412 069
176 064
24 6 043
2563 072
2784 077
5553 041
4184 046
6723 092
5575 0 70
656 033
370 013
188 044
222 033
387 057
444 0 28
13783 035
1229 037
16504 067
5552 040
4478 094
439 1 0 40
Soil
EWY-0
FDN
BDY
6
6
6
535 011
1175 043
703 045
133 081
149 020
847 021
4955 012
4458 019
3500 033
601 008
443 026
338 022
1831 019
2695 029
1208 052
-------
H
Table 6. Estimates and Log Standard Errors of Multiplicative Renovation
Effects
Sample
Medium
Sample Type
Lead
Aluminum
Barium
Cadmium
Calcium
Chromium
Log Std
Effect Err
Log
Effect Std Err
Log
Effect Std. Err.
Log
Effect Std. Err.
Log
Effect Std. Err.
Log
Effect Sid Err.
Dust
WCH
WST
AED
FIR
BRU
EWY-l
045 062
134 1.57
051 1 32
467 0.12
1708 164
487 004
062 044
096 024
281 1.05
066 0.10
014 120
057 013
0 1 570
0.27 140
737 366
035 011
501 192
025 025
032 001
021 091
043 27 17
059 039
017 092
084 037
084 0.21
283 002
1 72 366
255 007
1045 026
980w 014
061 078
043 024
989 384
044 015
319 011
056 045
Soil
EWY-O
FON
BDY
2 12 021
229 091
1 97 067
0.50 001
049 013
057 0.22
072 0.03
078 016
093 0 10
043 075
078 027
086 037
054 0 15
077 009
081 006
038 053
057 004
071 0 12
Sample
Medium
Sample Type
Magnesium
Nickel
Potassium
Titanium
Zinc
Log
Effect Std Err
Log
Effect Std. Err
Log
Effect Std Err
Log
Effect Std Err
Lag
Effect Std, Err
Dust
WCH
WST
ARD
FIR
BRU
EWY .l
1.14 030
127 014
352 1 98
138 007
088 1 22
221 006
1.10 038
159 019
79 23 0.58
084 054
141 04 323
109 021
072 078
052 067
031 1 34
028 025
004 6 74
039 055
104 016
085 002
348 1 55
074 012
0 18 2 61
058 009
041 018
047 016
001 3.61
145 018
3 35 7 14
191 018
Soil
EWY-0
FDN
BDY
136 001
027 021
195 014
044 074
063 005
062 005
056 002
052 004
063 013
062 001
072 008
065 005
125 004
090 009
124 030
9ndicates effect was signfucant at p 05 level
-------
Table 7. Estimates and Log Standard Errors of Multiplicative Abatement
w
Effects
Sample
Medium
Sample
Type
Lead
Aluminum
Barium
Cadmium
Calcium
Chromium
Log Std
Effect Err
Log
Effect Std Err.
Log
Effect Std Err
Log
Effect Std. Err
Log
Effect Std. Err.
Log
Effect Sid. Err.
Dust
WCH
WST
ARD
FLA
BRU
EWY-l
026 062
445 103
091 022
227 008
0.96 027
325 003
113 044
109 016
1 36 0 17
087 007
047 020
082 009
021 570
293 092
1 42 061
136 007
161 032
208 016
0.82 001
211 060
068 4 53
041 0.26
0.27 0.15
067 024
102 021
06P 001
058 0.61
092 005
1 11 004
0.60 009
119 078
068 016
1 60 064
027 010
056 002
033 030
Sod
EWY .0
FDN
BDY
451 014
213 060
242 044
077 001
0.90 008
114 0.14
122 002
1.13 010
149 007
241 049
130 018
158 0.25
081 010
069 006
092 004
081 034
107 002
123 008
Sample
Medium
Sample
Type
Magnesium
Nickel
Potassium
Titanium
Zinc
Log
Effect Std Err.
Log
Effect Std Err
Log
Effect Std Err
Log
Effect Std Err
Log
Effect Sod Err
Dust
WCH
WST
ARD
FLR
BRU
EWY-l
1.37 030
1 01 009
1 46 0 33
084 005
081 020
075 004
144 038
1 12 012
062 0 10
059 036
210 054
075 0.14
117 078
1 30 044
0 78 0 22
169 016
072 112
088 036
068 016
109 001
1 41 026
095 008
041 044
086 006
019 018
263 010
0 25 060
142 012
155 119
165 012
Soil
EWY-O
FDN
BOY
0 99 001
138 014
088 009
1 46 049
080 003
137 003
0 92 001
088 003
1 17 008
086 000
105 005
123 003
1 89 003
168 006
155 020
9ndicates effect was significant at p= 05 level
-------
concentrations were observed in renovated (abated) houses. Those
multiplicative effects that were significantly different from one
at the 0.05 significance level are denoted by asterisks.
Figures 2a, 2b, and 2c display block charts of the
estimates in Table 5 (portrayed on a log scale) . A distinction
between sample types was observed in the average levels displayed
in these figures. Therefore, the sample types were purposely
presented in three groups. Figure 2a displays the estimated
average log—concentration in unrenovated, unabated houses for
air ducts, window stools, and window channels. Figure 2b
displays the corresponding estimates for bed/rug/upholstery,
entryway, and floor samples. Figure 2c shows the estimates for
soil samples (boundary, entryway, and foundation) . Air ducts,
window stools, and window channels typically had the highest
baseline levels of lead, calcium, and zinc. Soil samples had the
lowest concentrations of these elements. Notice the relatively
similar behavior of these estimates across the different elements
within each of the three sample groups. For example, the ratio
of lead to aluminum is smallest for soil samples, and largest for
window channels, window stools, and air ducts. Another
distinction was observed in the relationship between lead,
titanium, and zinc. Figure 2a depicts a lower titanium level
than lead or zinc for air ducts and window channels. On the
other hand, levels portray a general rise as one moves from lead
to titanium to zinc for floors, entryways and bed/rug/upholstery
samples in Figure 2b. Finally, in Figure 2c, titanium was the
element with the highest concentration amongst lead, titanium and
zinc, in each of the soil samples.
In trying to identify the source of dust on floors, the
relationship among levels of the different elements for window
stools appears more similar to those for floors, entryways, and
bed/rug/upholstery samples than to those for window channels and
20
-------
N C
Figure 2a. Block Chart of Estimated Average Log-Concentration in Unrenovated, Unabated Units for
Window and Air Duct Dust Samples.
t o
F-a
Pb Ba Cd Ca
cr
Mg NI K 11 Zn
-------
Con 7(I j /1/I li/I iii /I’I iI A /
8.96 5.76/ 2.951 9.95 4.96 8.11/ 3.72 8.34 5.40 6.32 /
III / IU I /1 /
4.76 9.30 5.00 324 9.81 4.23 8.18 2.87 8.81 5.96 6.10
I
4.57 9.56 5.54/ 2.57 102 4.60 8.39 / 3. 8.63 6.10 6.08
Pb M Ba Cd Ca Cr Mg NI K 11 Zn
Figure 2b. Block Chart of Estimated Average Log-Concentration in Unrenovated, Unabated Units for Floor and
Textile Dust Samples.
-------
Con 7 fj / hi / iii j I /
4.14 1110.0 5.56 / 1.35 9.48 / 4.09 6. / a89 8.51 6.40 5.21 /
FON / /1 /1 /
4.63 9.83 5.53 128 9.50 3.45 7.07 2.70 8.40 6.09 5.60
6W I
4.09 9.39 4.86 / 077 9.16 / 3.11 5.38/ 2.18 8.17 5.89 484
Pb PJ Ba Cd Ca Mg NI K 11 Zn
B n 1
Figure 2c. Block Chart of Estimated Avrage Log-Concentration in Unrenovated, Unabated Units for Soil Samples.
-------
air ducts. This is likely a reflection of the general
composition of these dust samples. Figure 2a indicates that the
window channel samples have especially high concentrations of
barium and lead relative to the concentrations of the other
elements. In this manner, window channel samples seem to differ
from the other types of samples.
Close attention should be given to the log standard
errors of the estimates in Tables 6 and 7. Most of these are
very large in comparison to the (logarithm of the multiplicative)
estimates. Note that a total of 198 statistical tests were
performed in the analysis supporting the results in Tables 6 and
7. Each test was performed at the 5% level. Therefore, even if
there were no effects of abatement or renovation on any of these
element concentrations, we would still expect 9 or 10 sample
type/element/factor combinations to be significant. A total of
18 combinations were found to be significant. Of these, calcium
was involved in five cases, lead was the element involved in four
cases, potassium was involved in two cases, while aluminum,
cadmium, chromium, magnesium, nickel, titanium, and zinc were
each involved in one case. Entryways were involved in 9 cases of
significance (5 soil and 4 dust), floors in 3 cases, and window
stools in two cases. One case of significance was observed for
air ducts, bed/rug/upholstery, window channels, and foundations.
Thus, although more cases of significance were observed
than would be expected if there were no real effects, the number
of statistically significant results was small relative to the
number of tests performed. This, along with the limited data set
associated with the pilot study, makes it difficult and perhaps
inadvisable to draw general conclusions from the estimates
reported in Tables 6 and 7.
24
-------
2.2.2 abatement and Renovation Effects Across Elements
A principal components analysis was performed to
determine whether the relationships in element concentrations
among the houses (or houses with a similar abatement/renovation
history) were similar for the different sample types. This
analysis is an attempt to simplify the interpretation of the data
by reducing the number of elements characterized from eleven to
perhaps two or three “element classes”. These element classes
represent weighted averages of the eleven elements. Ultimately,
this may provide insight into the following source-assessment
question: “Where does the lead in household dust come from?”
This analysis, was performed on mean log-concentrations for each
element and house by sample type.
The purpose of this analysis was two-fold. A principal
components analysis provides a mathematical tool for estimating
the approximate dimensionality of the responses. Also, plotting
the higher-order principal components against each other affords
an objective means of identifying clusters of houses with similar
dust and soil element compositions. The ultimate goal is a
reduction in the complexity of the multivariate data analysis.
Principal component analyses can be performed based on
either correlations or covariances. Analyses based on
correlations standardize the range of each of the elements’
concentrations. This prevents the most widely fluctuating
elements from dominating the analysis and gives equal attention
to all variables regardless of their range. Covariance-based
analyses leave all element concentrations in their original
scale. Since the scales observed varied substantially by
element, and a priori there was no reason to weight more heavily
the elements with greater absolute variation, the principal
component analyses were performed based on correlations.
25
-------
Table 8 displays estimates of the first two principal
components (i.e., the two principal components explaining the
most variability in the data) by sample type, followed by the
cumulative proportion of total variation explained by these
components. Figures 3a and 3b display plots of the relationship
between the first and second principal components by sample type.
Figure 3a is for dust samples; Figure 3b is for soil samples.
Houses are distinguished by different plotting symbols in these
figures. Refer to Table 3 for a synopsis of abatement and
renovation history of these houses.
For many of the sample types, more than 70% of the
total variation is explained by the first two principal
components. There was a similar weighting pattern applied to the
elements for floor and entryway dust and entryway soil samples.
The weights are assigned to the different elements so as to
maximize the variation. Therefore, if two elements are
negatively correlated, then the weights of the high-order
principal components of the two elements will likely be of
opposite signs. For example, as will be seen in Section 2.3,
aluminum and calcium concentrations are negatively correlated in
entryway dust. Table 8 shows that their coefficients in the
first principal component are of opposite sign for entryway dust.
Obviously, when considering so many elements, it is impossible
for this relationship between correlation and principal component
coefficients to hold for all pairs of elements.
If there were patterns in the relationships of these
eleven elements affected by abatement or renovation history, then
homes with similar histories would expect to be clustered in
Figures 3a and 3b. The figure allows inspection for such
relationships separately by sample type. However, comparing the
proximity of the houses to each other in Figures 3a and 3b,
there do not seem to be any consistent groupings of houses across
26
-------
Table 8. Principal Componente for Unit Mean Log-Concentration by Sample Type
t’ )
— 3
Sample
Medium
Principal
Component Component
Principal_Comoonent_Coefficients
Pb Al Ba Cd Ca Cr Mg Ni K Ti Zn
Cumulative
Explained
Variability
Dust
ARD
1
2
0 15
-0.14
0.29
0.41
0.11
-0.49
-0 23
0.43
0.37
-0.08
0.45
0.08
0.40
0.25
0.29
-0.10
-0.27
0.46
0.40
0.28
-0.14
0.11
0.44
0.71
BRU
1
2
0.40
0.22
-0 33
0 39
0.25
0.00
-0 17
0.37
0.40
0.26
0.33
0.37
-0.08
0.48
043
0.16
-031
0.13
-0.27
0.37
-0.05
0.21
0.44
030
EWY-l
1
2
-0.32
0.16
0.38
0.28
0.22
-047
0.28
0.34
-0.39
0.24
0.29
0.50
-0.31
0.35
-0.10
-0.16
0.31
-026
0.37
0.19
-0.23
-0.01
0.47
0.67
FLR
1
2
-0.44
0.12
0.32
0.38
0.30
-0.34
0.29
0.37
-0.43
0.24
0 29
0.37
-0.23
0.22
0.01
-0.26
0.26
-0.25
0 24
042
-0.27
0.20
0.36
0.68
WCH
1
2
0.29
0.36
-0.26
0.33
0.38
0.17
0.13
042
0.34
-0.17
0.30
-0.22
-0.40
-0.02
0.17
-0.47
-0.31
028
0 38
0 12
0 24
0.41
0.57
088
WST
1
2
0.41
0.02
-0.37
0.10
0.43
0 14
-0.19
0.41
-006
-0 53
0.33
0 12
-0.34
0.07
017
-0.21
-0.28
0.42
0.11
045
0.35
0.29
0.45
0.74
Soil
BDY
1
2
-0.10
0.46
0.43
-0.03
0.22
0.41
0.08
0.44
0.19
-0 36
042
-0.02
-0.12
-0.24
0.42
-0.02
0.42
000
0.42
-0.02
-001
0 48
0.48
0.84
EWV-0
1
2
-0.36
0.29
0.40
-013
0.15
019
0.22
0.58
0.29
-007
0.33
0.21
-0.25
0.14
0.24
0.52
0.37
004
0.36
-0.16
-0.25
0.40
0.52
0.70
FDN
1
2
-0.33
0.29
0.38
0.13
0.16
0.45
-0.17
0 34
0.32
-0.28
0.36
0.13
0.33
0.25
0.36
-0.18
0.39
000
027
0 37
-0.06
0 50
0 57
0 85
-------
WCHI
vZ
____________ WST I
N +
ARDI
p. +
o a
* x
z
____________ FLRI
S
a “
___________ BRU I
N * X
x
a Z
___________ EWY-Il
a
1 ’
. x
HOUSE +++ 17 XXX IS *** 00043 flY 51 ZZZSO
Figure 3a. Plot of First Two Principal Components of Mean Log-
Concentrations for Dust Samples.
BDY I
z
N
c i +
X
EWY-Ol
+
V
a
X *
FDN I
N 0
V *
0
S I +
X
HOUSE 17 XXX 19 ***33 00043 VVV 5 ZZZ$
Figure 3b. Plot of First Two Principal Components of Mean Log-
Concentrations for Soil Samples.
28
-------
sample types. Substantive conclusions would require data from
more than six houses.
2.3 RELATIONSHIPS AMONG THE ELEMENTS
2.3.1 Bivariate Relationships (Correlations)
Displays portraying the bivariate relationships among
the eleven elements are provided in Figures 4a through 4i. For
each sample type, average log-concentrations for each house are
plotted for each pair of elements. Ellipses are drawn on each
plot that represent 95 percent of the estimated bivariate
distribution. Those plots for which the ellipse is narrow
represent pairs of elements for which there was a strong observed
correlation. Pairs of elements which are negatively correlated
have an ellipse with major axis running from upper left to lower
right. The magnitude of the correlation can be inferred from the
shape of the ellipse by comparing it to the key in Figure 5.
On the plots in Figures 4a-4i, each house is identified
with a different symbol. This allows one to determine whether
certain houses have similar characteristics with respect to the
various elements and/or sample types.
The strongest relationships among the elements across
houses were observed in foundation and boundary soil samples
(Figures 4h and 4i) . These correlations were strongest among
aluminum, chromium, nickel, potassium, and titanium. Strong
relationships were also observed among lead, calcium, chromium,
and nickel in samples taken from bedcover/rug/upholstery (Figure
4e) . Lead concentrations were most frequently correlated with
zinc concentrations. High correlations were also observed on
window stools between lead, titanium, barium, and zinc (Table
4b).
A slightly different categorization of sample types
could be made based on the patterns observed in these
29
-------
6
?
6’
c
N
6
Bs
6J
g
g
c
‘
9
€
9
9
c
c
c
;:
c
c
ç
9
E
c
3
N
E
7
c
c
/
c
i:
K
9
q
9
6’
c
€
®
g
7
c
3
Q
Il
\
@
/
&
9
c ;
3
?
c
6P
/
6
;
c
c ’
/
c
c
c
y
\
c
M
ç
y
\
c
,
g
N
,
c
c
£
K
6
/
\
,
,
c ?
2
Q
?
Figure 4a.
Window Ch2nnel House Mean
Correlation Scatterplot.
Figure 4b. Window Stool House Mean
Correlation Scatterplot.
c
c
i
c
c ;
N
9
c
/
c
c
c
/
Q:
6
/
9
c
,
c
c
/
c
9
c
ct
&
Q:
c
c
9
/
c
Figure 4c. Mr Duct House Mean
Correlation Scatterplot.
Figure 4d. Floor House Mean Correlation
Scatterplot.
House Legend: * = 17
0=19 +=33 X=43 Z=51 Y=80
— v
;
c &%
c; p
P1
K
J
—7
30
-------
/
6’
g
;
ftJ
c
‘
c
3
3
9
c9
c
9
c ;
9
,
6’
G
6’
c)
a’
6’
a’
c
3
c 7
;
c
9
6
6’
6’
()
P1
(
c
3
£
K
g
9
c
3
G
c
£
c
Figure 4e. Bedcover/Rug/Upholstery House
Mean Correlation Scatterplot.
Pb
(
(9
(
( )
(
N
(9
6’
(
I
Q
9
,
,
c 5
°.
c
;
ci:;
c
Cd
/
,
D
/
c;;’
c; 3
c
9
,
c
P1
ct
I
c 9
3
:
K
!
Figure 41. Entryway Dust House Mean
Correlation Scatterplot.
X=43 Z=5i. Y=80
c
%
&
—
N
9
3
s
c 3
6’
/
9
3
c13
D
£
;
(9
,
&
cV
&
?
c
®
2
4)
D
%
c
E
G
/
; ‘
9
N
6’
e
6’
Y
V
6’
K
?
,
9
3
s
!3
6’
CD
:D
c
%
Figure 4g. Entryway Soil House Mean
Correlation Scatterplot.
Pb
c
9
\
\
c
N
c
‘
‘
I
/
@
a
©
©
/
( 7
Cd
)
)
(
(
\
c
Cd
)
if
(
(
c
/
if
@
(
if
)
)
if
I
3
(
if
)
P1
if
( 9
(
if
9
if
ci
‘
if
g
if
K
(
(
;,
c
c
Figure 4h. Foundation Soil House Mean
Correlation Scatterplot.
House Legend: * 17 0 = 19 + = 33
3].
-------
Pb
(9
(9
(
(?
( )
6’
;
c
&
/
6’
/
6’
cQ
7
L
I
‘
cV
3
g
I
/
c
cc
c
ç
c9
cc
c
,
c
,
6’
/
c
c
ç
(
6’
(g?
( 7
6’
(
F
/
( 3
/
( )
9
6’
@
6’
K
6’
( )
,
cg
cD
c
/
/
6’
g
g
,
c
Figure 4i. Boundary Soil House Mean
Correlation Scatterplot.
HouseLegend: *j7 019 +33 X43 Z51 Y80
90%
60%
30%
0%
0
0
0
0
Figure 5. Key to Relation Between Shape of
Ellipse and Observed Correlation in
Figures 4a Through 4i.
scatterplots. Relatively consistent sets of bivariate plots were
observed for the following groups of sample types: floor,
interior entryway, and exterior entryway; boundary soil and
32
-------
foundation soil; and window stool and window well. Air ducts and
bedcover/rug/upholstery samples do not appear similar to any of
the groups mentioned nor to each other. These groupings of the
sample types do not appear particularly surprising although one
might have expected exterior entryway samples to be more like the
other two soil samples than like interior floor dust samples.
The floor, interior entryway, and exterior entryway
group displays consistent, strong bivariate relationships between
aluminum and titanium; cadmium and chromium; and barium and
potassium. As introduced above, the boundary and foundation soil
group displays the strongest bivariate relationships, suggesting
consistent correlations between lead and calcium; aluminum and
chromium; nickel, potassium, and titanium; chromium and nickel;
potassium and titanium; and nickel and potassium. The window
channel and window stool group has consistent bivariate
relationships between lead, barium, and zinc; chromium and
magnesium (negative correlation); and titanium and zinc.
2.3.2 Multivariate Relationships (Principal Components)
For the estimated model parameters displayed in Tables
5, 6, and 7 (average log-concentrations, in unrenovated unabated
houses, increments in log-concentration associated with
renovation, and increments in log-concentration associated with
abatement), a second principal components analysis was performed
on these estimated model parameters across the nine sample types.
The purpose of this analysis was not only to identify consistent
patterns in the composition of dust across different sample types
(unrenovated, unabated house analysis), but also to determine
whether abatement or renovation impacts different components in
different ways.
The numerical results of the principal components
analyses and plots of the first two principal components are
33
-------
displayed in Table 9 and Figure 6. Table 9 displays estimates of
the coefficients for the first two principal components followed
by the cumulative proportion of total variation explained by
these components. Figure 6 displays the relationship between the
first two principal components (the orthogonal directions in
which the greatest variability were observed)
The first two principal components generally accounted
for at least 68% of the total variability in the model parameter
estimates. This means that although there were eleven elements
measured (lead, aluminum, barium, cadmium, calcium, chromium,
magnesium, nickel, potassium, titanium, and zinc), most of the
variation among the nine sample types occurred within a two-
dimensional space (i.e., two linear combinations of the eleven
element concentrations)
For averages in unrenovated, unabated houses it can be
argued that the three soil sample types are grouped into one
cluster; floor, entryway, window stool, textile, and air duct
dust sample types form another cluster; and window channels stand
alone. For the renovation effect, all samples are grouped into
one cluster except for air ducts and textiles, which stand alone.
One must recognize that air ducts and textiles were not sampled
in the fully renovated house. Therefore, the estimated impact of
renovation on these sample types is less meaningful than on the
other sample types which were sampled in the fully renovated
house.
For the abatement effects, there were no clear clusters
or outlying sample types, but the three soil sample types appear
close together in Figure 6, and interior entryway samples were
right on top of exterior entryway samples, with floor dust
samples nearby. This may be an indication that even after
abatement, the composition of the soil near the foundation and
entryway is similar to that of the soil at the boundary.
34
-------
U ’
Table 9. Principal Components for Unrenovated, Unabated Home Averages,
Abatement Effects, and Renovation Effects
Coefficients are applied to the estimated parameters for each sample type to obtain maximum spread among sample types in two dimensions
Response
Principal
Component
Principal_Component Coefficients
Cumulative
Explained
Variability
Pb
Al
Ba
Cd
Ca
Cr
Mg
NI
K
Ti
Zn
Unrenovated Unabated
Unit Means
1
020
-037
0.17
0.43
0.41
0 15
037
036
-009
-017
0.32
040
0 71
2
048
0 20
0 48
0.04
-O 00
-0 32
-000
-0.28
-0.27
043
0 25
Abatement Effect
1
034
-0 37
0 30
0.11
007
-0.43
-0 43
0 16
009
-023
042
0.36
068
2
035
029
0.31
0.31
-0.46
0.07
006
-0.34
0 10
044
0 22
Renovation Effect
1
002
040
0.43
-0.13
0.03
046
034
030
-022
040
-010
043
0 83
2
047
-0 22
0 13
-0 15
0 40
-0 01
-0 15
0 33
-0 37
-023
0 45
-------
+
, x*
a
V
*
MTh
* + 06
Oon oriiifl +‘ WOH xxx *** aaa D ORU
ZZZ EWY—I * EWY’_O 060 FflJij 666 V’
Figure 6. First Two Principal Components for Each Building Component, Plotted versus
Each Other for Unrenovated, Unabated Unit Mean Log-Concentrations,
Renovation Effects, and Abatement Effects.
Notice that the clusters apparent in Figure 6 (for the
unrenovated unabated houses) are consistent with the groupings
identified in Figures 2a, 2b, and 2c. One notable exception to
this is that the window channel sample type appears
to be somewhat removed in Figure 6 from the grouping identified
in Figure 2.
3.0 CONCLUSIONS
Based on the statistical analyses of the multi-element
data, the following conclusions were drawn:
36
-------
Characterization of multi-element concentration
1. Of those elements analyzed, aluminum, calcium, and
potassium had the highest concentrations in indoor dust
and soil.
2. After controlling for abatement and renovation effects,
concentrations of the elements provide for the
following groupings of sample types:
- air duct, window stool, and window channel dust
- floor, entryway, and textile dust, and
- foundation, entryway, and boundary soil.
Effects of abatement and renovation on multi—element
concentrations
3. Lead concentrations in dust and soil near the entryways
of abated houses were three to five times the levels in
unabated houses. The concentrations of lead in dust
from floors and entryways of renovated houses were
about five times those in unrenovated houses.
4. Calcium concentrations in the dust of renovated houses
were significantly higher on window stools, floors,
textiles, and entryways. This difference was tenfold
for textiles and entryways.
5. Several other statistically significant effects were
estimated for the remaining elements, but with little
consistency across elements or across sample types.
Relationships between element concentrations for sample types
6. Of the ten elements measured besides lead,
concentrations of zinc were most positively correlated
with lead, both within sample types and across sample
types.
7. The strongest bivariate relationships among the
elements were observed in boundary and foundation soil
samples; three groups of sample types were identified
as having similar bivariate relationships among many of
the elements; floor, interior entryway, and exterior
entryway; boundary and foundation soil; and window
channel and stool.
37
-------
8. A principal component analysis of estimated element
concentrations in unrenovated control houses by sample
type suggested similarities in dust and soil
composition within the following groups: 1) exterior
entryway, foundation, and boundary soil, and 2) floor,
interior entryway, window stool, air duct, textile
dust.
The sources of lead in household dust still remain to
be identified. The sparcity of data available for this analysis
proved to be a limitation of this study. The following section
suggests the potential value in performing a similar analysis on
data collected in the CAP full study.
4.0 RECO * ENDATIONS FOR uxu E WORK
In the pursuit of a solution to the problem of
childhood lead poisoning, it is important to identify the sources
of lead in order to best target its containment. The data used
for this analysis were collected in a pilot study for a large
scale investigation of the long-term efficacy of various
abatement methods based on samples collected from six houses.
The full CAP study, initiated one year later, involved 52 houses,
35 abated and 17 unabated, and would provide a greater amount of
data to address this issue. Although fewer samples were
collected at each house in the full study than in the pilot,
samples were collected from similar locations in the full study,
with additional dust samples collected outside entryways to
provide a better means of relating interior dust lead levels to
exterior lead levels.
The data used in this analysis were collected from only
a small handful of houses (six) in the Denver area. Analysis of
these data was not expected to demonstrate significant
relationships. Instead, it was used to determine sample sizes
and test sampling protocols for the subsequent full CAP study.
38
-------
Use of the data collected in the full CAP study to perform
analyses similar to those in this report would provide nine times
as much data (52 houses) to characterize the source of lead in
household dust, both in abated and unabated houses.
5.0 REFERENCES
US EPA, 1995, “Comprehensive Abatement Performance Pilot Study,
Volume I——Results of Lead Data Analyses”, EPA 747—R—93—007.
Morrison, D., 1976, Multivariate Statistical Methods , Second
Edition, McGraw-Hill.
39
-------
APPENDIX A
SU) ARY OF )WLTI-ELE NT DATA
A-i. Multi-Element Data Listing
A-2. Geometric Mean Concentrations by Sample Type and Unit
-------
Table A-la. CAP Pilot Study Multi-Element Data, House 17
m
C
is
a
-I
is
I-I
Sample Identification
Cc centratiori, WglgI
Medium
Component
Location
Sample ID
Pb
Al
Ba
Ca
Cd
Cr
K
Mg
Ni
Ti
Zn
Dust
WCH
WST
ARD
FIR
BRU
EWY-l
KIT
KIT
BD1
BD1
LVG
LVG
LVG
KIT
BD1
KIT
KIT
BD1
BD1
BD1
LVG
LVG
601
EWY
EWY
07
06
14
16
36
39
40
09
19
01
03
11
12
13
31
32
18
20
21
114131
22083
726 85
337.99
50643
26955
33678
36344
71746
50
25391
37318
32803
22471
153
6369
6692
2816
25873
2677724
659556
16274 87
1252425
4475.87
1254895
976988
896632
8663.88
1694
694983
728784
926141
609359
5165.24
6462 19
513931
1024255
1022246
91487
43983
627 26
72482
37746
182144
217222
187 13
17304
742
184107
741.58
87534
6981177
44163
165.42
4338
3667
110304
22699.03
48002.71
39088.01
417287
29715
21150.79
2721952
163533
16881.18
14246
23113.58
15383.03
8765.93
33722.36
1389872
7081.12
19032.89
1225382
16548 39
b
1137
197 81
19129
3952
30686
14556
6572
61527
31
1339
2614
1457
874
1061
371
997
1958
11 48
4502
2357
35 77
3858
4287
4331
5083
4033
6485
1619
2935
4366
4271
2926
2596
2459
4385
365
3465
480513
3185688
3817 92
499182
880283
4923.04
6291.52
77438
578907
14419
1715868
1001167
1146905
1486837
987072
4597 17
809704
811838
8421 34
487308
845697
8035 08
735533
1091543
397775
637633
378481
373365
2724
395011
279445
224061
4176246
249427
1595 8
321003
228621
30933
2048
15916
23 06
222
18811
1684
27 12
2268
1963
13
1627
12044
455
3387372
22248
186
7684
273
2789
956.57
32289
551 82
36847
24343
626.99
50508
24546
29633
5507
10436
20734
18784
24252
1586
20868
849
2852
331 86
149234
173207
10045 28
422307
252174
130890
190958
516 53
3986045
502
133825
51621’
28448
174851
48600’
229 40
572.3
42553’
620 35
Soil
EWY-0
FDN
BOY
FRO
BAC
LFT
BAC
BAC
IFT
BAC
IFT
22
23
24
26
29
26
21
28
704
36388
7024
694
6568
5222
7046
5636
2041234
1958558
2077997
17951 05
1821279
2670432
2021482
2505986
195.83
43975
198 82
261 98
1713
22099
18278
20569
1284944
14160.18
12213.25
11275 17
1173242
13088.59
825867
133344
275
24107
281
262
251
268
233
261
377
26894
4089
39 16
3796
4456
3848
4381
536142
45706
5407 94
4415 52
446055
639675
593693
587071
53998
61415
667 53
2568 3
296125
9843
500
102935
1507
23811
15 72
13 93
1429
171
1547
1642
48649
58248
421 58
39086
3849
69191
45442
6432
18052
49930’
279 23
34543
29892
11563
17678
10848
analysis result was greater than upper calibration limit for zinc, reported value is a lower bound on the true Zn concentration assuming an instrument sample concentration
of 5 pg/mL
analysis result was greater than upper calibration limit for cadmium, sample excluded from data analysis
C
0
z
0
C,
CD
0
-------
Table A-lb. CAP Pilot Study Multi-Element Data, House 19
• Analysis result was greeter than upper calibration limit for zinc; reported value is a lower bound on the true Zn concentration assuming an instrument sample
concentration of 5 pg/mL
b ICP analysis hampered by calcium interference; no multi-element date reported
sampla dropped in lab, therefore, no multi-element data reported
Titanium Concentration received by Battelle on July 24, 1995 via fax from MRI Originally, concentration reported as 0 38 pg/g
Sample Identification
Concentrations_(pn/g)
Medium
Component
Location
Sample ID
Pb
Al
Ba
Ca
Cd
C i
K
Mg
Ni
Ti
Zn
Dust
WCH
WST
ARD
FIR
BRU
EWY-I
BD I
LVG
BD1
KIT
LVG
BDI
IVG
LVG
BD1
BD1
BD1
KIT
KIT
LVG
801
EWY
EWY
17
04
16
36
09
19
01
03b
11
12’
13
31
32
08
18
20
21
36788
7083
21527
17682
69 53
62441
18993
69 53
30115
.
4023
99.5
6794
481 99
48457
20107
18441
412958
776183
418735
.
894803
45633
.
5504.92
568566
425268
43303
680906
2901.15
8659.46
674203
7414
28117
20885
.
5851
17872
59805
.
83081
10288
53.14
694.67
19046
27511
5681
148864.4
7419087
9268482
6961023
177240
1997404
58480.23
927994
8142.32
9380468
37037.35
1395138
94780.78
414
3741
1696
.
2372
615
.
1949
1362
571
324
12 73
8.51
6.16
1008
5025
77.8
30.28
145.95
3606
.
113.58
15706
4494
41.9
186.6
81.44
4075
40.06
12028
24483
169326
309705
288897
.
247267
213681
22913
226931
1900 86
102431
539783
205037
.
1242288
405361
261933
.
510354
7937.28
.
336526
398515
291028
289751
4602 61
242759
689201
598681
.
1906
11631
4727
3129
3145
•
15209
30646
4322
4072
389 22
11161
3061
4764
41619
38527
16558
35142
129.78
157.26
16632
13606
14307
26468
10443
28962
24122
23135
205174
94377
146594
706.47
•
68325’
152083
31554’
26691’
1967 25
34126
55109
58253
Soil
EWY .0
FDN
BDY
FRO
SAC
FRO
FRO
LFT
FRO
LFT
LFT
22
23
28
24
25
26
27
29
4969
4035
19653
4918
23839
98 17
4329
4416
12841.36
928405
31300.7
1020946
10529.04
10937.94
8343.79
80294
130.93
12833
40863
11556
22761
120.94
11587
110.3
1223328
13372.92
1513247
1260323
1245127
8323.19
1116584
11658.17
227
2.04
323
202
485
2.3
2.3
163
2366
1789
3425
1974
2781
24 6
1598
1526
343163
284254
698002
300539
318934
3432 85
34851
295139
49092
37028
98468
40253
37827
43045
151102
150823
1026
11.68
138
801
2097
8 91
658
649
3829
28455
75313
29468
37389
379 06d
25652
22281
16104
27844
281
14331
46134
160 67
10729
13044
0
C
is
0
-I
is
0
0
z
0
C ’)
P.
0
0
0
0
—
CD
-------
Table A-ic. CAP Pilot Study Multi-Element Data, House 33
Sample Id e
ification
Concentrations_(ugig)
Medium
Component
location
Sample ID
Pb
Al
Ba
Ca
Cd
Cr
K
Mg
Ni
i i
Zn
Dust
WCH
WST
ARD
FIR
BRU
EWV-I
LVG
BD2
LVG
LVG
LDY
BD2
LVG
BD2
602
LVG
LVG
IVG
KIT
KIT
LVG
EWY
EWY
17
04
14
16
36
09
19
01
03
11
12
13
31
32
18
20
21
7238 25
57488
17452
56192
58072
47655
160502
13475
18292
18867
12835
10682
11607
8819
11693
12802
8842
13345 86
704392
97397
804521
395714
803089
3552
490884
488063
1313019
1244336
13414.99
1361978
1323038
1195433
21715 21
1791001
7057 8
48848
59445
183116
5104017
20649
22492
35711
13885
29987
452.52
16689
28809
30061
162.54
225.8
29809
3486607
3734122
2687453
5583219
1552299
7669884
3678182
42255
4181808
2078874
2154272
2385941
1903319
20173.65
1822988
2095709
1585119
2966
1959
2474
1101
1006
1978
656
1305
4089
8891
6606
2081
3571
3295
2544
1294
3001
3905
1352
10078
8699
8583
5297
4067
9674
8523
18048
19026
14588
51597
67648
6906
94 18
52319
256344
59598
373399
335462
1512946
366559
841354
1830.23
120962
5104.15
5707.61
584624
598561
560286
6722.99
5796 35
582801
423708
414932
322294
443562
677991
337843
218899
325346
293791
317231
294908
406177
349118
366808
3558 28
5175 63
387388
17 86
5223
2146
2457
1714285
277
4461
3348
1507
1863
1959
2287
209
168
17 55
21 51
1274
655 72
62537
37272
48041
28313
29663
11962
16479
19509
38912
31419
3245
38586
35506
38739
57207
55832
13782 89
118328
149916
161062
117743
261921
1039904
42599
64609
93914
86568
60819
60868
57675
44783
45839
48172
Soil
EWY-0
FDN
BDY
FRO
BAC
FRO
IFT
FRO
LFT
IFT
FRO
22
23
28
24
25
29
26
27
632
13578
5696
16685
10821
17574
4412
16151
228306
2617844
2153998
2195418
22731 44
25525 21
1090828
1316162
25195
401 46
27996
35624
3093
369 12
12092
16131
812595
12471 77
809303
1241096
12923 18
12303 17
1203103
526873
252
1443
175
351
327
4 17
218
201
2939
951 74
2691
3134
2842
36 76
2699
1992
418877
6241 22
352544
4964.74
361552
5539 57
297687
306188
49469
84889
49383
305737
616 18
3354 59
47446
49714
1079
1306
678
1591
11 9
13 25
959
758
73025
66737
57476
42279
601 36
49783
32138
44276
13993
243 15
12208
25819
26327
26501
16482
11154
0
0
D
a ’
a
0
z
0
C)
a’
0
0
C
0
a’
-------
Table A-id. CAP Pilot Study Multi-Element Data, House 43
a
C
a
a
S I
a
C
z
C
-t
C ,
‘4
is
C
0
C
0
a
a
Sample Idrilication
Conc—ntrations (p9/9)
Medium
Component
Location
Sample ID
Pb
Al
Ba
Ca
Cd
Cr
K
Mg
Ni
Ti
Zn
Duat
WCH
WST
ARD
FLA
BAl i
EWY-l
LVG
KIT
LVG
DIN
K IT
LVG
DIN
LVG
LVG
DIN
DIN
DIN
KIT
KIT
LVG
DIN
EWY
EWY
05
38
04
16
36
09
19’
01
03
11
12
13
31
32
08
18
20
21
96283
143363
96439
37835
39618
1137.67
611 38
14728
2048
23442
25576
14905
30191
309.13
102.12
19548
26347
58933
1371603
3536838
516815
1051271
917231
915241
659581
782678
691686
863223
7486 ,79
1039615
1341074
6497.66
11463.49
134452
1425466
38386
36137
52056
51198
44322
24313
22036
28836
41968
39332
21042
87281
593.49
2087
30442
33058
211325
5639876
13100
4743145
20181 33
3378892
6353554
15095
4332359
3013819
2190083
1496863
17116 36
250368
2797508
2214325
18154.49
2328933
893
7226
1817
2059
2206
1103
471
126
1.73
812
459
823
879
6
615
526
691
2381
325
8274
2855
4434
16475
.
3378
3019
515
4202
4411
47 04
456
3991
3162
3531
262
534436
463556
458629
663451
355389
41003
701708
3173751
861411
62741
680485
7385 11
690821
920171
111228
919123
605799
1403996
4535.1
4450
401788
421003
672457
.
294076
80882
372116
345041
2911.18
314581
443063
385689
410081
453116
446419
17.3
1818
2525
1745
2256
287
.
2313
2614
4487
215
1501
204
6153
2559
2579
1887
1641
50813
2443
440.36
311 83
35268
40836
.
198
25124
23053
23722
26214
291 45
42212
19796
3439
48626
46697
253841
111925
134371
694971
115981
780599
16441
98924
286697
116095
13154
94868
98063
298949
124749
76315
206626
Soil
EWY-0
FDN
SD ?
Ff10
SAC
SAC
Ff10
SAC
Ff10
Ff10
SAC
22
23
26
24
25
29
26
27
62277
20461
30405
33658
18072
24477
28961
6075
13817 34
1939988
1512008
1854827
2157086
1935304
1251732
533662
37445
37351
28312
46003
33868
33715
203
8322
1206842
1305533
1296142
1000294
1583413
824321
125337
579139
6.58
283
245
539
38
429
453
094
2865
3209
2588
4181
3642
3463
2833
1375
381292
455404
387982
380351
414451
319713
478004
173611
493 6
506 18
49309
306692
6104
240935
49145
30056
11 76
11 66
1071
1264
1406
1202
1212
951
32631
74013
49612
60108
72299
51659
41348
31351
491 55
30034
21117
81166
56124
48190’
22139
8810
analysis result was greater than upper calibration limit for zinc, reported value ia a lower bound on the true Zn concentration asauming an instrument aample concentration
of 5 pg/mL.
ICP analysis hampered by calcium interference, no multi-element data reported
-------
Table A-le. CAP Pilot Study Multi-Element Data, House 51
Sample Identification
Concentrations tug/a)
Medium
Component
Location
Sample ID
Pb
Al
Ba
Ca
Cd
Cr
IC
Mg
Ni
Ti
Zn
Dust
WCH
WST
FIR
EWV-l
BAT
6D3
803
BAT
8D3
BD3
BD1
BAT
BD3
803
BD3
BD1
BD1
BD3
EWY
EWY
07
15
17
06
14
16
40
01
11
12
13
31
32
44
20
21
27335
42097
49256
636701
77393
67018
358094
244616
96616
46655
71155
178438
176035
64648
64015
40262
483191
1334632
124968
402183
794958
916425
695302
440799
633688
115958
505573
5689.56
60863
328844
8493.4
711383
119476
28846
29992
67853
278.13
31415
74631
9302
4324
8619
13508
143281
32504
2704
23409
7553
1226355
1353355
1562875
1535659
9231311
7734641
7752348
133542
2643023
1483808
1132687
91261.26
3928109
1769777
129846
1272934
132
611
521
199
473
608
7
878
75
172
53
719
644
4.37
698
116
2609
3355
3376
3112
22.88
3032
2611
2065
2581
559
1683
2667
2289
1406
2245
3776
90078
327529
340675
90472
216789
3113.23
278275
1864.38
208407
81457
192261
169294
20503
175782
231994
1629
1454626
438673
455774
929264
473466
482295
512268
858939
301353
101629
559211
368986
313713
200657
721804
743466
525
2198
1946
165.25
9039
1886
2441
3598
1533
34
1349
1275
1192
802
1371
1978
36181
48532
56974
25934
34525
40651
48646
14892
18787
4414
17511
22554
25957
1171
2935
21149
319873
7528
54891
411193
83542
86583
217436
338741
96637
30397
78238
143514
147076
65703
74326
275947
Soil
EWV .0
FDN
BDY
FRO
BAC
FRO
BAC
SAC
FRO
BAC
BAC
22
23
24
25
28
26
27
29
8992
50471
93765
53896
42607
34581
32934
30049
8711.72
912617
916965
920729
931742
776156
819419
738651
231.52
26885
25838
26178
25677
20656
17685
1781
410207
579736
544899
796009
752254
593411
655566
707242
451
374
413
381
316
386
255
24
2237
2313
1592
2252
2017
2457
1922
1689
2287.08
264947
16075
242953
23088
222418
259753
243224
18991
30233
38385
151691
29547
30399
14909
168919
69
77
751
71
69
1118
614
583
3423
32398
378
34325
34636
3064
30505
27129
43252
37643
53306
37739
33984
31377
23503
21725
• Analysis result was greater than upper calibration limit for zinc; reported value is a lower bound on the true Zn concentration assuming an instrument sample concentration
of 5 pg/mI
During initial sampling attempt, cartridge filled with sawdust prior to completion of sample collection Sample was excluded from lead analysis and multi-element
analysis An additional sample (ID 44) was taken to compensate for the loss of this sample
I - r i
‘ 5
C
‘5
0
5 ,
0
0
z
0
(1
‘5
0
0
C
0
‘5
-------
Table A-if. CAP Pilot Study Multi-Element Data, House 80
Sample Identification
Concentrations (pg/g)
Medium
Component
Location
Sample ID
Pb
Al
Ba
Ca
Cd
Ci
K
Mg
Ni
Ti
Zn
Dust
WCH
WST
ARD
FIR
BRU
EWY.I
BD3
KIT
KIT
BAT
BD3
PAN
KIT
KIT
BAT
BD3
KIT
BAT
BAT
BD3
BD3
003
KIT
KIT
BAT
BD3
EWY
EWY
15
41
42
06
14
36
39
40
09
19
45
01
03
11
12
13
31
32
08
18
20
21
93817
4550
579419
6157385
68012
53494
78807
46576
169936
96514
38948
1211.33
64918
17977
175.4
24278
1817
22333
344 3
6632
34228
2216
1161247
813736
113925
60989
611906
519784
383469
625781
58125
52732
3608.13
6872.02
8731 69
371825
480607
643044
4950.09
550859
7776.B8
209904
1176211
744436
84607
2246622
1088835
3031504
137666
6579447
2940219
655833
163709
36633
46969
100552
571.9
18597
17637
24019
322.95
35011
26282
101.32
303.49
25664
510127
6835811
2945978
2125135
3819057
1052487
2927588
4586667
4969503
3223993
13403.01
50977
32751 16
1394134
179657
9710.26
1820009
15063.71
41062 18
7622 978
25044.11
9819924
1758
2307
3037
3083
1715
785
2329
201
665
779
552
537
4.37
925
509
533
423
798
569
479
861
4
4971
9482
977
15136
6602
601
10421
20568
8399
7882
1692
3115
3238
4608
5942
4426
2535
5669
3607
3333
3327
2887
33384
95886
180519
153603
34805
2471 875
74495
315438
22064
348149
341719
384785
43785
351587
505449
384208
254197
3838.51
25092
113596
499187
67004
503225
405676
374558
508089
315553
2740865
242798
270906
376306
227704
181777
299118
2857 74
150536
194274
171868
188888
2293.53
295272
118064
370709
235301
1561
21.48
1473
42.43
9928
1591346
3535
13964
3764
1244
1007
1456
1851
5116
1954
1477
1055
21.54
15 21
42 51
2782
882
43949
71513
56778
1813
42613
62996
49368
46068
22536
20941
10344
22591
197.7
15545
17748
22446
23915
24299
272
11735
38857
30073
184876
4826.26
450938
3512127
1632.77
25875
7559.21
3467.03
596348
1171 51
123995
163863
117944
43573
50802
326.29
43631
51433
66444
13609
70344
46774
Soil
EWY.O
FDN
BOY
FRO
BAC
BAC
IFT
BAC
BAC
FRO
BAC
22
23
28
24
25
29
26
27
37956
34975
411 94
941 59
45886
31683
30797
34253
1642786
1515385
1764231
1727246
88089
888523
12974.08
134267
281.68
28832
34022
41378
201 67
19759
24626
27856
6962 53
1053447
823478
6943.95
5155 78
74332
831715
7260 13
988
769
829
1398
606
756
93
6 19
31 33
319
31 81
3287
2299
2377
2402
2479
496823
470839
522336
444321
247422
256888
422044
4661 5
4892
50188
48699
77246
1512 5
149713
48946
1231
1151
1383
1382
743
805
951
48568
50074
52791
56427
322 17
28829
43714
385 15
41746
492 42
972 71
34504
37723
39431
• Analysis result was greater than upper calibration limit to, zinc, reported value is a lower bound on the true Zn Concentration assuming an instrument sample concentration
of 5 pg/mL
0
0
z
0
*
(1
( 5
0
0
C
0
C D
C D
‘C
CD
0
CD
-------
Table A-2. Geometric Mean Concentration by Sample Type and Unit
Sample
Type
House
Interior
Abatement
History
Exterior
Abatement
History
Renovation
Samples
Taken in
Unit
Ba
Geometric Mean Concentrations_ija
Ig)
Pb
Al
Ca
Cd
Cr
K
Mg
Ni
Ti
Zn
WCH
33
43
51
80
U
A
E
E
U
A
A
E
Full
1
2
3
3
72383
11749
827 6
29136
133459
22025 3
9305.9
10248r
70578
3755
4693
59152
348661
27181 3
29601 8
461391
2966
2540
773
2309
3905
2782
30 92
7722
25634
49774
2158 1
17945
4237.1
7979 5
6625 4
42444
1786
1802
28 21
3669
655.7
352S
464.2
5630
137829
2089 1
1097.5
34266
WST
17
19
33
43
51
80
R
U
U
R
E
E
E
U
U
fl
R
E
Partial
Full
6
3
4
3
4
5
368.3
1392
425.4
525 1
18544
38283
95057
51200
6836 3
7928 2
67183
34165
8176
1633
721 8
4907
4586
55561
332013
1007821
643059
31861 3
960192
409179
14020
1380
1522
4354
796
1810
3812
49.10
100.43
47 14
2740
105.18
74100
17084
32600
47642
20303
12544
71919
5090.4
4478 2
4222 3
5742.0
31059
43.74
4715
26.21
21.50
5121
5058
414.4
2983
422.0
3645
3648
4063
27815
7652
13543
2212.6
15947
52235
ARD
17
19
33
43
80
A
U
U
A
E
E
U
U
A
E
Pertial
2
1
2
1
3
510.6
6244
874.6
11377
861.2
88138
89480
5340.9
91524
48000
1799
585 1
2155
2431
6655
166151
696102
53114.2
635355
27795 2
20109
2372
3602
1103
659
5114
145 95
4641
16475
48 20
66955
3097 1
5553.4
41003
2971 9
37591
5103 5
27194
67246
24974
2110
312 90
3515
2870
16 77
2697
351 4
1884
4084
1696
45375
14659
16503.7
78060
20538
FIR
17
19
33
43
51
80
A
U
U
A
E
E
E
U
U
R
A
8
Partial
Full
7
5
7
7
6
7
1655
173 1
1307
2209
12271
3048
55489
48306
99214
8504 5
50240
5668 4
6426
2174
2672
3800
1378
3388
146863
274705
256223
22451 7
542046
19605 1
917
787
3589
688
643
571
2883
65 56
20343
41 41
2063
4040
109740
23984
38547
88280
18900
37896
27265
3911 9
33415
3844 8
38705
21099
3995
76 26
2042
2729
1444
1903
1511
1459
2905
2642
1796
2066
5687
573 1
6476
1310 1
12352
6097
BRU
17
19
33
43
80
A
U
U
R
E
E
U
U
A
E
Partial
1
2
1
2
2
669
4833
1169
1413
151 1
51393
44446
119543
86305
4040 3
4338
3637
1625
2521
1632
190329
589430
182299
248889
176923
997
1041
2544
607
522
43.85
12327
6906
3878
34 67
80970
13954
67230
84568
16883
32100
33426
35583
39770
1867 1
7684
20842
1755
2569
2543
849
1663
3874
2609
178 7
5723
8194
4478
19312
3007
EWY-l
17
19
33
43
51
80
R
U
U
R
E
8
8
U
U
R
A
E
Partial
Full
2
2
2
2
2
2
2699
1926
1064
3940
16054
2754
102325
76408
19721 0
138440
77731
93574
6360
1250
2594
8358
1330
2791
142400
1149923
182262
205622
1285634
155217
1499
788
1971
603
900
587
3556
4040
221 98
3042
2912
3099
82685
33268
58122
77016
19440
18289
26593
64235
4477 1
44976
73255
29534
2759
3819
1655
1763
1646
1566
3076
2643
5652
4765
2491
3418
5138
5666
4699
1255.7
14321
573.6
—3
0
5 1
0
z
0
C,
51
0
0
C
0
p.
C D
-------
Table A-2. (Continued)
Sample
Type
House
Interior
Abatement
History
Exterior
Abatement
History
Renovation
Samples
Taken in
Unit
Geometric Mean Concentrations (4
Pb Al Ba Ca Cd Cr K
lg)
Mg
Ni
Ti
Zn
EWY-0
17
19
33
43
51
80
ft
U
U
R
E
E
E
U
U
ft
ft
E
Partial
Full
2
3
3
3
2
3
1601
733
788
3384
673 7
3796
199947
155108
234370
16152.0
89165
16376.5
2935
1901
3048
3411
2495
3023
134889
135278
93606
126871
48766
84530
25.75
246
399
357
4 11
857
10069
2438
9097
2876
22.75
3168
49502
40834
45171
40690
2461 6
49622
575.9
5636
5919
4976
757 7
4926
5990
1183
985
1139
729
12.51
5323
4345
6543
4934
3330
5045
3002
2327
1607
342.3
403.5
4294
FDN
17
19
33
43
51
80
R
U
U
H
E
E
E
U
U
ft
ft
E
Partial
Full
3
2
3
3
3
3
684
1083
1469
2460
5994
5164
189396
103680
233545
19783.7
9231 2
110573
207.4
1622
3439
3745
2590
2545
117340
12527.0
125429
109297
68844
64322
264
3 13
363
445
368
862
3932
2343
31 99
37 50
19 34
2619
47402
30960
46328
4092 1
2081 4
30453
17187
3902
1848 8
1652.2
566.2
12049
1463
12.96
1359
12.88
7 17
938
3988
331 9
502 1
6304
355 5
3742
3066
257 1
2686
605 7
4089
502.1
8DY
17
19
33
43
51
80
ft
U
U
H
E
E
6
U
U
ft
ft
E
Partial
Full
3
3
2
2
3
2
592
57.3
860
132.6
3247
3248
238274
90156
11982 1
8192.7
17738
131984
2025
1156
1397
1300
1867
2619
112960
102708
7961 7
85198
65040
77707
254
205
209
206
287
759
4219
1817
23 19
1974
1998
2440
60637
32807
3019 1
28807
24131
44355
7972
9936
4857
3843
9148
4913
1632
725
853
1077
737
1025
5870
2788
3772
3853
2938
3775
1304
1310
1356
1401
2521
3952
U unabated unit. ft = removal unit, and E = encapsulationienclosure unit
C D
CD
D
Si
0
0
z
0
CD
0
0
CD
cE
-------
Revtew Draft - - Do Not Cite or Quote
APPENDIX B
OUTLIER ANALYSIS FOR THE CAPS
PILOT MULTI-ELEMENT DATA
-------
Review Draft -- Do Not Cite or Quote
APPENDIX B
OUTLIER ANALYSIS FOR THE CAPS
PILOT MULTI-ELEMENT DATA
B-]. INTRODUCTION
This appendix documents the statistical outlier analysis
performed on the CAPS Pilot multi-element data. The statistical
approach employed, the outliers identified, and the findings of
MRI’s review of the outlier data are discussed.
Two outlier tests were applied to the multi-element data.
The first was a univariate outlier test, which evaluates one
element at a time. This is the same test that was previously
applied to the lead data. The test was applied to the natural
logarithms of the concentrations for lead, aluminum, barium,
cadmium, calcium, chromium, magnesium, nickel, potassium,
titanium, and zinc. The second test was a multivariate outlier
test, which evaluates measurements for all eleven elements
simultaneously. The multivariate test detects measurements which
for a single element may not be an outlier, but when viewed in
combination with the other elements is inconsistent with the
majority of the data. Before performing the outlier tests,
groupings of the data were defined.
B-2 DATA GROUPING
The following homogeneous groups of data were identified
for each indicated sample type:
• Vacuum Cassette 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);
• Soil Samples (4 groups): boundary, foundation, exterior
entryway, and all exterior samples combined.
B-].
-------
Review Draft -. Do Not Cite or Quote
Initially, data for all six units in the Pilot Study were
combined before performing the univariate and multivariate
outlier tests on these groups. When there were sufficient data,
subsequent univariate outlier tests were also performed by
segregating the data in each group by abatement method and by
housing unit. Segregating by abatement method and unit were not
done for the multivariate test due to the need for larger sample
sizes with the increase in dimensionality.
B-3 METHODS
The details of the univariate and multivariate outlier
tests are given in the following sections.
B-3-l Univariate Outlier Test
Formal statistical outlier tests were performed on the
natural logarithms of the concentrations for lead, aluminum,
barium, cadmium, calcium, chromium, magnesium, nickel, potassium,
titanium and zinc. Data were placed into groups of comparable
values, and a maximum absolute studentized residual procedure was
used to identify potential outliers. The SAS procedure GLM (SAS
PC, ver. 6.08) 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 student-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.
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 more outliers were detected.
B-2
-------
Review Draft -- Do Not Cite or Quote
B-3-2 Multivariate Outlier Test
The multivariate outlier test is based on the Hotelling T-
squared statistic, with one major difference (the Hotelling T-
squared statistic is discussed in most multivariate statistics
texts, such as Multivariate Statistical Methods. Second Edition ,
by Donald F. Morrison, copyright 1967, 1976 by McGraw-Hill, Inc.,
page 131) . The difference in the statistic used here is that, in
computing the statistic for the i tt observation, that observation
is excluded from the computation of the mean vector and the
variance-covariance matrix. This yields estimates of location
and covariance which are unaffected by the observation in
question, and leads to a more robust outlier test. This is a
multivariate extension of the univariate studentized residual
used for the univariate outlier test. Under assumptions of
normality, the resulting statistic has an F distribution, with
numerator degrees of freedom equal to p (the number of elements)
and denominator degrees of freedom equal to a function of p and
the sample size, N. In this case, p was equal to eleven (the
eleven elements are: lead, aluminum, barium, cadmium, calcium,
chromium, magnesium, nickel, potassium, titanium, and zinc)
The observation corresponding to the maximum value of the
statistic in a data group was declared a potential outlier if the
statistic exceeded the (1- .10/N) quantile of the F distribution
with appropriate degrees of freedom. When a potential outlier
was identified, that sample was excluded from the group, and the
outlier test was performed again. This procedure was repeated
until no more outliers were detected.
B-4 RESULTS OF OUTLIER ANALYSIS
The potential outliers identified by these two tests were
screened by a statistician to eliminate those which were merely
numerical anomalies due to very small sample sizes. The
B-3
-------
Review Draft -. Do Not Cite or Quote
remaining outliers identified by the univariate test are listed
in Table B-i, and those identified by the multivariate test are
listed in Table B-2. These lists of the remaining outliers were
sent back to the laboratory for verification. However, the
laboratory reported a few errors for several samples. The
corrected values appear in Tables A-la through A-if while the
original reported values can be found in the footnotes to these
tables.
B-4
-------
Review Draft -- Do Not Cite or Quote
Table B-i. Univariate Outliers Detected by Univariate Methods
Sample
Processing
Batch
House ID!
Sample ID
Concentration IM
Igl
Al
Ba
Cd
Cr
Ni
Ti
Zn
CLS
33/20
94 18
CRS
33/21
523 19
SSS
33/23
14.43
951.74
CSS
33/31
51597
CSS
33/32
676 48
SSS
43/22
6 58
SSS
43/26
_________
4.53
SSS
43/27
094
1375
CSS
43/11
2866 r
CSS
43/32
422 12
CKC
43/36
220 60
CKC
17/01
16.00
5500
50200
CLS
17/03
104.36
SSS
17/23
241 07
26894
238 ll
CLS
19/04
231 35
CLS
19/08
18660
CLS
19113
152083
SSS
19125
485
SSS
19126
038
CLS
17119
61527
SKI
43(24
5 39
SSS
19(28
753 13
CLS
19/36
165.58
CRS
80/06
609.89
30315 04
181 30
35121 27
SSS
80(24
1398
564.27
972.71
SSS
80126
9 30
SSS
80(27
6 19
CLS
80109
5963.48
CLS
80(45
16.92
CSS
80(39
29402.19k
CSS
80141
22466 2?
CRS
51(12
1.72
5.59
4414
CLS
51(20
22.45
SSS
51/24
53306
SSS
51(26
386
CRS
33(19
99.07
CAS
43/16
306 14
These values were sent to MRI in a letter dated February 1 7. 1992.
These values were confirmed by MRI via a fax received by Bettelle on July 24. 1995
All other values were sent to MRI in a letter dated January 14. 1992.
B-5
-------
Table B-2. Outliers Detected by Multivariate Methods.
Sample
Processing
Belch
House
Sample
ID
Concentration (jigig)
Pb
Al
Ba
Cd
C
Cr
Mg
Ni
K
Ti
Zn
CLS
17
03
253 9V
6949 83
1841.07
13 39
23113 58
29 35
3950 11
16 27
17158 68
104 36
1 338.25
CKC
17
01
50
1694
742
3 r
14246
16 19
2724
13
14419
55.07’
502
CRS
80
06
6157385
60989
3031504
3083
2125135
16136
508089
4243
153603
1813
35121.27
SSS
51
26
345 8V
7761 56’
206 56
3 86
5934.11.
24 57’
303 99’
11 18
2224 18
306.4
313 77’
SSS
17
23
36388
1958558
439.75
241 07
1416018
268.94
61415
23811
45706
582.48
49930
SSS
33
23
135 7
2617844
401 46
1443
12471 77
961 74
84889
1306
6241 22
66737
243 15
‘These values were confirmed by MRI in a fax received by Battelle on July 28, 1995. All other values were confirmed by M I II in a fax received by Battelle on July 24. 1995
Performing the multivariate outlier test on only the elements lead, cadmium, chromium, titanium and zinc also identified the following samples as outliers
(identified by House/Sample IDI 17/22. 19/23, 19/26. 33/21 and 51121
t j
CD
C
CD
0
CD
-I
0
0
z
0
c - I
CD
0
-I
0
C
0
CD
------- |