&EPA
          Protecti
          Pollution Prevention
          and Toxics
                               f Pfi 747-R
                               April 1995
Report on the National
Survey of Lead-Based
Paint in Housing
Appendix  II:
Analysis

-------
         REPORT ON THE NATIONAL SURVEY

          OF LEAD-BASED PAINT IN HOUSING

                 Appendix II: Analysis
             This work was conducted under
           HUD Contract Number HC-5848 and
EPA Contract Numbers 684)9-0174,6S-D2-0139, and 6S-D3-0011
                     June, 1995

-------
The material in this document has been subject to Agency technical and policy review and approved for
publication as an EPA report. The views expressed by individual authors, however, are their own and
do not necessarily reflect hose of the U.S. Environmental Protection Agency. Mention of trade names,
products, or services does not convey,  and should not be interpreted as conveying, official EPA
approval, endorsement, or recommendation.
                                              11

-------
                    TABLE OF CONTENTS
INTRODUCTION	       1-1

1.1    Background	       1-1
12    Report Organization	       1-1
1.3    Reports Based on the National Survey	       1-1
1.4    Objectives of this Appendix	       1-1
NATIONAL ESTIMATES OF PREVALENCE	       2-1

2.1    Private Housing	       2-1

       2.1.1   Prevalence of Lead-Contaminated Paint, Dust,
              and Soil in Private Housing	       2-1
       2.1.2   Amounts of Lead Paint in Private Housing	       2-10
       2.1.3   Levels of Lead in Paint, Dust, and Soil in Private
              Housing	       2-13
       2.1.4   Prevalence of Lead by Degree of Urbanization for
              Privately-Owned Housing	       2-20

2.2    Public Housing	       2-32

       22.1   Prevalence of Lead-Contaminated Paint, Dust,and
              Soil in Public Housing	       2-32
       2.2.2   Amounts of Lead Paint in Public Housing.	       2-35
       2.2.3   Levels of Lead in Paint, Dust, and Soil in Public
              Housing	       2-35


SOURCES OF ERROR IN THE NATIONAL SURVEY DATA	       3-1

3.1    Statistical Concepts and Terminology	       3-1
3.2    Response Rates and Potential for Non-Response Bias	       3-2

       3.2.1   Private Housing	       3-2
       3.2.2   Public Housing	       3-9

3.3    Correcting for Measurement Bias	       3-16

       3.3.1   Adjusting Field Measurements for Calibration Bias	       3-17
       3.32   Adjusting Recalibrated Measurements for Censoring
              Bias	       3-27

3.4    Correcting for Bias in the National Estimate of Lead-Based
       Paint Prevalence	       3-31
                              11

-------
                            TABLE OF CONTENTS (continued)
                    3.4.1  Three Types of Estimates at the Dwelling Unit Level	       3-31
                    3.4.2  A Simulation Approach	       3-33
                    3.4.3  Adjusting the Maximum Conected Measurement to
                           Remove Bias in the Estimate of Lead-Based Paint
                           Prevalence	       3-36

             3.5    Laboratory Measurement Error and the Effects of Small Dust
                    Sample Weights on the Findings	       3-38

                    3.5.1  Trimming the Dust Analysis File	       3-38
                    3.5.2  Trimming me Soil Analysis File	       3-40
                    3.5.3  Laboratory Comparisons by Site for Dust Analysis	       3-40
                    3.5.4  Laboratory Comparisons by Site far Soil Analysis	       3-41
                    3.5.5  Evaluation of me Laboratory Dust Analysis Method	       3-41
                    3.5.6  The Problem of Small Dust Weights	       3-44
                    3.5.7  Quality of Laboratory Data-Conclusion	       3-45


   4         CONCLUSIONS	       4-1

             4.1    Major Conclusions	       4-1

                    4.1.1  Study Findings	       4-1
                    4.1.2  Impact of Measurement Error and Lead Concentration
                           Variations on the Data Analysis	       4-1
                    4.1.3  Use of the Spectrum Analyzer MAP/XRF	       4-2

             4.2    Additional Conclusions and Recommendations	       4-2


                                      List of Appendices

Appendix                                                                           Page

   A    Additional Data Tables for Private Housing	       A-l

   B    Additional Data Tables for Public Housing	       B-l

   C    Dust and Soil Samples Excluded from Data Quality Analysis	       C-l

   D    XRF Validation Data and the Distribution of the Adjusted XRF Readings	       D-l
                                            111

-------
                            TABLE OF CONTENTS (continued)


                                       List of Tables

Table                                                                              Page

   2-1        Estimated Number of Privately Owned Occupied Housing Units
              Built Before 1980 with Lead Based Paint, by Selected Characteristics	       2-2
   2-2        Number and Percentage of Privately Owned Occupied Housing
              Units with Lead in Paint by Lead Concentration, Year of
              Construction, and Location of Lead-Based Paint	       2-4
   2-3        Prevalence of Non-Intact Lead-Based Paint (LBP) by Location in
              the Buflding-Privately Owned Occupied Housing Units	       2-5
   2-4        Rate of Occurrence of Privately Owned Occupied Housing Units with
              Dust Lead in Excess of the Federal Guidelines	       2-7
   2-5        Dust Lead Loadings by Location in Privately Owned Occupied
              Housing Units With or Without Lead-Based Paint (LBP)	       2-8
   2-6        Association Between Lead in Interior Surface Dust and Lead-Based
              Paint (LBP) Condition for Privately Owned Occupied Housing Units	       2-9
   2-7        Association Between LeadinSofl and Exterior Lead-Based Paint
              Condition for Privately Owned Housing Units	       2-11
   2-8        Estimated Number of Privately Owned Occupied Housing Units in
              Selected Lead Hazard Categories	       2-11
   2-9        Amounts of Lead-Based Paint (LBP) on Interior Surfaces by Painted
              Component for Privately Owned Occupied Housing Units	       2-12
   2-10       Amounts of Lead-Based Paint (LBP) on Exterior Surfaces by Painted
              Component for Privately Owned Occupied Housing Units	       2-14
   2-11       Amounts of Lead-Based Paint (LBP) on Interior Surfaces by Selected
              Characteristics for Privately Owned Occupied Housing Units	       2-15
   2-12       Amounts of Lead-Based Paint (LBP) on Exterior Surfaces by Selected
              Characteristics for Privately Owned Occupied Housing Units	       2-16
   2-13       Arithmetic Mean Paint Lead Loadings in Privately Owned Occupied
              Housing Units Buflt Before 1980, By Selected Characteristics	       2-17
   2-14       Arithmetic Mean Paint Lead Loadings in Privately Owned Occupied
              Housing Units Built Before 1980, by Architectural Component and
              Construction Year	       2-18
   2-15       PercentUes and Mean for Pamt Lead (XRF) Measurements for
              Private Housing Units by Sample Location (Unweighted)	       2-19
   2-16       Percentiles and Mean for Lead in Soil Samples from Private Housing
              Units by Sample Location	       2-21
   2-17       Number of Privately-Owned Housing Units in the Sample by Degree
              of Urbanization and Construction Year	       2-25
   2-18       Number of Privately-Owned Housing Units in the Sample by Degree
              01 lrftyiiiy.3HiQp ftnfl RjHHQH Tlir	Tiiirn	T-IIIIHI--I-TII+ iiii	                    2**25
   2-19       Estimated Number COOOs) of Privately-Owned Housing Units in
              the Nation by Degree of Urbanization and Construction Year	       2-27
   2-20       Estimated Number COOOs) of Privately-Owned Housing Units
              in the Nation by Degree of Urbanization and Region	       2-27
   2-21       Percentage of Privately-Owned Housing Units with Lead-Based
              Paint by Degree of Urbanization and Construction Year.	       2-28
                                            IV

-------
                             TABLE OF CONTENTS (continued)


                                   List of Tables (continued)

Table                                                                                Page

   2-22       Percentage of Privately-Owned Housing Units with Lead-Based
              Paint by Degree of Urbanization and Region	       2-28
   2-23       Percentage of Privately-Owned Housing Units with at Least
              5 Sq. Ft. of Damaged Lead-Based Paint by Degree of Urbanization
              and Construction Year	       2-29
   2-24       Percentage of Privately-Owned Housing Units whit at Least
              5 Sq. Ft of Damaged Lead-Based Paint by Degree of Urbanization
              and Region.	       2-29
   2-25       Percentage of Privately-Owned Housing Units with Lead in Dust
              Above Guidelines by Degree of Urbanization and Construction Year	       2-30
   2-26       Percentage of Privately-Owned Housing Units with Lead in Dust
              Above Guidelines by Degree of Urbanization and Region	       2-30
   2-27       Percentage of Privately-Owned Housing Units with Lead in Soil
              Ahnw CnritMmf* fry Degree of TTthaniTatinn and rnngtnirfirm Vaar	       2-31
   2-28       Percentage of Privately-Owned Housing Units with Lead in Soil
              Above Guidelines by Degree of Urbanization and Region	       2-31
   2-29       Estimated Number and Percent of Public Housing Units Built
              Before 1980  with Lead-Based Paint, By Selected Characteristics	       2-33
   2-30       Number and Percentage of Public Housing Units Built Before 1980
               with Lead-Based Paint by Lead Concentration and Location of
              Lead-Based Paint	       2-34
   2-31       Amounts of Lead-Based faint (LBP) on Interior Surfaces by
              Architectural Component and Substrate for Public Housing Units	       2-36
   2-32       Amounts of Lead-Based Paint (LBP) on Exterior Surfaces by
              Architectural Component and Substrate for Public Housing Units	       2-37
   2-33       Arithmetic Mean Paint Lead Loadings in Public Housing Units
              BuOt before 1980, by Selected Characteristics	       2-39
   2-34       Arithmetic Mean Paint Lead Txffldingf by Painted Component t*"^
              Construction Year for Public Housing Units	       2-39
   2-35       Percentiles and Mean for XRF Measurements for Public Housing
              Units by Sample Location	       2-40
   2-36       Percentiles and Mean for Lead in Soil Samples from Public Housing
              Units by Sample Location	       2-41
   3-1         National Response, Contact, and Eligibility Rates at Each Data
              Collection Stage	       3-3
   3-2         Census Blocks Lost by Census Region and Survey Data Collection
              Stage	       3-4
   3-3         On-Square Results for Demographic and Sodoeconomic Variables	       3-5
   3-4         Number and Percent of Census Blocks by Overall Response Rate
              and Census Region	       3-8
   3-5         Association Between Inspected Housing Units in the Same Census
              Blocks with Respect to the Presence/Absence of Lead in Paint, Dust,
              and Soil	       3-9
   3-6         Distribution of Public Housing Family Units by Construction Year	       3-11

-------
Table
                             TABLE OF CONTENTS (continued)


                                  List of Tables (continued)
   3-7        Distribution of Public Housing Family Units by Census Region	       3-12
   3-8        Distribution of Public Housing Family Units by PHA Size	       3-14
   3-9        Paint Lead, Paint Damage aru^ Dust Lead in Public Housing, by
              Occupancy Unweighted Sample Counts	       3-15
   3-10       Standard Deviation of Replicate Field Measurements by Substrate,
              Pooled Across MAP/XRF Instruments	       3-22
   3-11       \Afam and Standard nfyJatJOP of Stmnlat»i B^»afthr»ftx* Vf*aii«t 1 fnt\ 1 neuKngp fiy Pphlic Hnnying^ Tnterinr
              Locations	       2-42
   3-1         RMSE and Bias of the Sample Mean and Median When Estimating
              the Mean of Internal Measurements	       3-20
   3-2         Pooled Standard Deviation of 60 Second Validation Measurements
              by Substrate	       3-23
   3-3         Validation Measurements by Shim Lead Concentration for XRF
              Instrument #2 on Steel	       3-25
   3-4         Histogram of Recalibrated Measurements in Public and Private
              Dwelling Units	       3-29
   3-5         Histogram of Corrected Measurements in Public and Private
              Dwelling Units	       3-30
                                             VI

-------
                                      1. INTRODUCTION
1.1    Background

       Appendix n focuses on the statistical analysis of data collected during the National Survey of Lead-
Based Paint, sponsored by the Department of Housing and Urban Development (HUD).

       For background information on prior related studies, surveys and their limitations; lead in the
environment and pathways between paint lead and blood lead; and the effects of lead-based paint abatement
and  refinishfng on lead contamination refer to Appendix I of this report. Detailed descriptions of the
National Survey design and methodology also are discussed in Appendix L  Therefore, they are not
repeated here.


1.2    Report Organization

       There are four chapters in Appendix n, including the introduction. Descriptions of each section are
as follows:

       •     Chapter 2 provides prevalence data of lead-based paint in housing, and the magnitude of dust
            and soil lead contamination in residential environments. It also shows calculated amounts of
            lead in paint and the extent of priority hazards for private housing.

       •     Chapter 3 examines the quality of the data collected during the national survey of lead-based
            paint and the resulting quality of projected national estimates. In order to analyze this, the
            section addresses the effects  of false negatives on the data quality; it analyzes non-response
            rates  and other potential biases in the sample. The section also addresses the MAP/XRF
            performance in measuring lead content in painty the laboratory measurement error inherent to
            all analytical ftyhlV
-------
analytical procedures used to analyze the national survey data.  Readers are referred to the base report of
this document for a summary of the significant findings stemming from these analyses.
                                              1-2

-------
                      2. NATIONAL ESTIMATES OF PREVALENCE
      This chapter presents data cm the prevalence of lead-based paint in private and public housing
units built in the United States before 1980. Also included are the amount of surfaces covered with
lead-based paint and the distributions of lead levels in residential paint, dust, and soil. Unless otherwise
specified, all statistics presented in tins chapter are weighted estimates calculated from the survey data
using statistical sampling weights, as described m Appendix I, Design and Methodology, Section 6.6.

      Throughout this chapter, the concepts of lead loading and lead concentration are used. Loading
applies to the amounts of lead in paint and dust while concentration applies to lead amounts in soil.  For
paint, the loading concept refers to milligrams of lead per square centimeter of surface (mg/cm2), for
dust it refers to nricrograms of lead per square foot (ug/ft2).  Soil is reported in parts per million (ppm),
or the equivalent micrograms of lead per gram of soil (ug/g).  In all three media, appropriate health-
based standards  have not yet been developed for housing.  For ease of presentation we  have used the
current Federal action levels for paint, dust and soil.  It is important to note that these action levels are
not health-based. Because of this, the information presented below does not denote dwelhng units with
"safe" and "unsafe11 lead conditions.  More will be said on the origins of the action levels in the
appropriate sections.


2.1   Private Housing

      The private housing statistics presented below  include  revisions of  data published in the
Comprehensive  and  Workable  Plan for the Abatement of Lead-Eased Paint in Privately  Owned
Housing: Report to  Congress  (CWP Report).  The statistics,  bom point and interval estimates,
presented in the CWP did not incorporate corrections forme effects of paint, dust and soil measurement
bias and variation, and the effects of incomplete testing of surfaces within housmg units. The reasons
for and the impact of the corrections are described in more  detail in Chapter 3 of this  volume.  The
findings presented in this section are all corrected for sources of error, and adjustments are incorporated
into the results of the analyses in bom their point and interval estimates.

2.1.1 Prevalence of Lead-Contaminated Paint, Dust, and Sofl in Private Housmg

      Table 2-1:  Estimated Number of Privately Owned Housmg Units with Lead-Based Paint
by Selected Characteristics - An estimated 64 million (77 million to 81 million)1 or 83 percent of all
privately occupied housmg units in the United States built before 1980 have lead-based paint on some
surface in or around the building.  Housing units included in the above statistic have lead-based paint
"somewhere" in one or more of me following locations: the interior, the exterior, or the common areas of
multi-family structures (Le., hallways, lobbies, mailrooms, laundry rooms, and playgrounds). A surface
with lead contamination is defined here, and by HUD, as having a measured paint lead loading of 1.0
milligram of lead per square centimeter of surface (mg/cm2) or greater.

The data collected  during the National Survey suggests that older homes are more  likely to have lead-
based paint than newer homes.   As displayed in Table 2-1, an estimated 76 percent (64 percent to 88
percent) of the housing units built after  1960 have lead-contaminated paint on their surfaces, but the
percentage increases to 92 percent (84 percent to 100 percent) for houses built between 1940 and 1959.
A slight  counter-intuitive decrease in prevalence  is evident in homes built between 1940 and 1959
compared to pre-1940 homes. Eighty-eight percent (79 -97 percent) were contaminated.  This decrease
The numbers in p«renlhe»ei ire 95% confidence mtervils. See Appendix H, Section 3.43 for the methodology wed to compute the
 confidence intervals.
                                             2-1

-------
                                              TABLE 2-1

         ESTIMATED NUMBER OF PRIVATELY OWNED OCCUPIED HOUSING UNITS
      BUILT BEFORE 1980 WITH LEAD-BASED PAINT, BY SELECTED CHARACTERISTICS
                             (Paint Lead Concentration > = 1.0 mg/sq cm)
Characteristic
Total Occupied Housing
Units Boilt Before 1980
Construction Yean
1960-1979
1940-1959
Before 1940
Housing Type
Single Family
Multifamily
One or More Children
Under Aee 7
Census Region
Northeast
Midwest
South
West
Owner-Occupied (2)
Market Value of Home
Less than $40,000
$40,000 to S79.999
$80,000 to $149,999
$150,000 and up
Renter-Occupied (2)
Monthly Rent Payment
Leas than $400
$400 and up
Household Income (2)
Less than $30,000
$30,000 and up
Total
Occupied Housing
Units (000) (1)
77,177
100%
35,681
46%
20,476
27%
21,018
27%
66,418
86%
10,759
14%
13,912
18%
16,963
22%
19,848
26%
24,967
32%
15,399
20%
50^54
11,885
24%
19,401
38%
11,863
23%
7,405
15%
24,734
16,339
66%
8,395
34%
46,126
60%
31,048
40%
Housing Units
WHh Lead-Based Paint
Somewhere in Building
Percent
83%
(9%)
76%
(12%)
92%
(8%)
88%
(9%)
85%
(9%)
77%
(17%)
89%
(9%)
86%
(13%)
91%
(10%)
82%
(10%)
73%
(18%)
84%
(9%)
92%
(12%)
90%
(10%)
68%
(18%)
85%
(15%)
82%
(11%)
85%
(14%)
81%
(15%)
85%
(10%)
81%
(11%)
Number (000)
64,443
(6,946)
27,275
(4,282)
18,742
(1,638)
18,424
0,892)
56,130
(5,978)
8,308
(1,829)
12,425
(1552)
14,605
(2,205)
18,115
(1,985)
20,393
(2,497)
11,298
(2,772)
42,516
(4,550)
10,888
(1,426)
17,550
(1,976)
8,093
(2,135)
6,276
(1,111)
20,329
(2,721)
13,811
(2287)
6,822
(1259)
39,032
(4,613)
25,121
(3,415)
Number of
Housing Units
in Sample (2)
284
120
87
77
227
57
90
53
69
116
46
179
39
46
45
42
105
59
40
156
127
(1) Total mats data are from the 1987 American Housing Survey.
(2) Some respondents did sot respond to the questions on economic variables. Therefore, counls for
   djsaggregatioD may not add to corresponding aggregate counls.
Note: Numbers ia parentheses are approximate half-widths of 95% confidence intervals for the estimated
percents and numbers. For example, the approximate 95% confidence interval for the percent of housing
units wita joffle lead-based paint is 83% + / -9% or 74% to 92%.
                                               2-2

-------
is not statistically significant,  however, and is probably due to a combination of sampling  and
measurement variation.

       Childhood lead poisoning is one of the most common and preventable public health concerns in
our country today.3  For this reason, EPA also examined the potential exposure of children to lead-
based  paint.   An  estimated  12 million (11 -  14  million) of the  homes projected to  have lead-
contaminated paint are also occupied by families with children under the age of seven,

       Finally, the differences among lead-based paint prevalence by type of housing, market value of
the home, amount of monthly rent payment, and household income are not statistically significant.l

       Table 2-2:  Number and Percentage of Privately Owned Housing Units with Lead-Based
Paint by Concentration, Construction Year, and Sample Location - Because there are no health-
based standards for determining lead paint loadings in housing, the data was examined at varying cut-
off thresholds used by different states and the Federal government. The lowest (which is to  say most
rigorous) loading standard, 0.7 mg/cm2, corresponds to the definition of lead-based  paint employed by
the State of Maryland. This data is presented in the first column of Table 2-2.  The  next most rigorous
loading (as used in Table 2-1) is 1.0 mg/cm2, and corresponds to the Federal definition set by HUD.
The third loading, 1.2 mg/cm2, is the threshold used by the State of Massachusetts.  Lastly, 2.0 mg/cm2
is used.  Although 2.0 mg/cm2 is not a current standard, it is included as an upper-end threshold for
infonnational  purposes.   Expanded tables  (similar to Table 2-1) for 0.7, 1.2 and 2.0 mg/cm2, by
selected characteristics, are presented in Appendix A of this volume (Tables A-2, A-3, and A-4).

      The analysis reveals  that as the lead loading threshold increases, the number of newer homes
meeting  the criteria decreases faster than the older homes.  In feet, the  prevalence of newer homes
(1960-1979) with lead-based paint "somewhere" in the building fells substantially from 82 percent to 48
percent as the threshold increases from 0.7 to 2.0 mg/cm2.  In contrast, the prevalence decrease for older
homes (pre-1940) is virtually unchanged.  Thus, the data suggests that older homes have higher paint
lead  loadings than newer homes.   Since the amount of lead added  to manufactured commercial
residential paint declined from 1940 to  1980, this observation is not unreasonable. More discussion on
the levels of lead in paint is presented in Section 2.1.3.

      Table 2-3: Prevalence of Nonintact Lead-Based Paint by Location -  Fourteen million or 19
percent of the private dwelling units in the United States are estimated to have nonintact (damaged)
lead-based paint somewhere in the building.  Nonintact lead-based paint is defined as at least five square
feet of defective lead-based paint per unit.  As is evident by the data presented in  the  table, it is
estimated there is about twice the prevalence of nonintact lead-based paint on the  exterior of private
housing units (13 percent) than on the interior (7 percent).
 Test of equality of two independent proportions, in the presence of a complex simple design ami measurement errors. The test statistic
 i« Z = (p, - p,)/(yar p, +var fj"*, where var p = (0 J * confidence interval width/1,96)i. The proportions are significantly different
 from each other if Z > 1.96 (two-tided tett); p, is significantly greater than p, if Z > 1.645 (one-sided test).

 CDC [1991]. Preventing Lead Poisoning in Young Children. U.S. Department of Health and Human Services, Public Health Service,
 Center for Disease Control.
                                               2-3

-------
                          TABLE 2-2
NUMBER AND PERCENTAGE OF PRIVATELY OWNED OCCUPIED HOUSING UNITS
       WITH LEAD IN PAINT BY LEAD CONCENTRATION, YEAR OF
         CONSTRUCTION, AND LOCATION OF LEAD-BASED PAINT
Location and
Construction Year
Unit Interior
1960-1979
1940-1959
Built before 1940
Interior Common Areas
1960-1979
1940-1959
Built before 1940
Building Exterior
1960-1979
1940-1959
Built before 1940
Somewhere in Bnflding
1960-1979
1940-1959
Built before 1940
Percentage of Homes
Paint Lead Concentration (me/sq cm)
>=0.7
70%
58%
76%
84%
7%
5%
5%
11%
77%
68%
84%
86%
87%
82%
94%
88%
>=1.0
63%
49%
69%
83%
5%
4%
5%
6%
73%
61%
81%
86%
83%
76%
92%
88%
> = L2
60%
43%
66%
82%
4%
3%
5%
5%
69%
54%
80%
86%
80%
69%
89%
88%
>=2.0
45%
23%
54%
72%
3%
1%
5%
5%
59%
36%
76%
83%
68%
48%
83%
88%
Location and
Construction Year
Unit Interior
1960-1979
1940-1959
Built before 1940
Interior fy>inmnn Areas
1960-1979
1940-1959
Built before 1940
RnHrimg F.vtvrinr
1960-1979
1940-1959
Built before 1940
Somewhere in BuDding
1960-1979
1940-1959
Built before 1940
Number of Homes (000)
Paint Lead Concentration (me/sq cm)
>=0.7
53,856
20,669
15,537
17,653
5,225
1,913
1,018
2^00
59,258
24,099
17,146
18,018
66,829
29,195
19,210
18,426
>=1.0
48,986
17,483
14,114
17,392
3,597
1,317
1,018
1,261
56,495
21,804
16,675
18,018
64,059
27,278
18,739
18,426
>=1J
45,960
15,234
13,434
17,289
3,103
1,021
1,018
1,066
53,585
19,113
16,454
18,018
61,473
24,770
18,280
18,426
>=2.0
34,499
8,340
10,961
15,200
2,508
425
1,018
1,066
45,644
12,718
15,553
17,373
52,690
17,218
17,046
18,426
                             2-4

-------
                                 TABLE 2-3

         PREVALENCE OF NON-INTACT LEAD-BASED PAINT (LBP)
                    BY LOCATION IN THE BUILDING -
             PRIVATELY OWNED OCCUPIED HOUSING UNITS
Location of
Non-Intact LBP
Building Interior (2)
Interior Common Areas
Building Exterior
Both Interior and Exterior
Somewhere in Bidding (3)
Occupied Housing Units With
Non-Intact Lead-Based Paint
Number (000)
5,596
249
9,657
1,083
14,354
Percent (1)
796
0%
13%
1%
19%
(1)  Base equals all 77,177,000 housing units built before 1980.
(2)  "Building Interior" means that die only non-intact LBP is in the
    interior; there may be intact LBP on the exterior. "Building Exterior"
    tia.Q a similar pifapifip

(3)  A housing unit has some non-intact LBP if there are more man 5 sq.
    feet of damaged LBP somewhere.  Similar definitions apply for interior
    and exterior LBP.  It is therefore possible for a housing unit to have non-intact
    LBP somewhere in the building without having either non-intact exterior LBP or
    non-intact interior LBP (for example, a house with 3 sq. ft of damaged interior
    LBP and 3 sq. ft of damaged exterior LBP).
Note: There was no non-intact LBP in sampled playgrounds.
                            2-5

-------
      Table 2-4 shows the estimated number of housing units with dust lead loadings below and
above the HUD Interim Guidelines are presented.  Specifically, HUD's Interim Guidelines for dust
lead levels  are  200 /*g/ft2 for floors, 500 /ig/ft2 for window sills,4  and 800 us/ft2 for window
wells.  All of these loading standards are set for clearance purposes only (i.e., declaring an abated
residence ready for re-occupancy after lead paint abatement).  Even if HUD's dust standards were
health-based, readers should be cautioned when comparing the National Survey dust lead results to
the HUD guidelines because the former were collected with a vacuum sampling  technique  and  the
latter were developed for a wipe sampling technique.  The sampling recoveries, i.e., percent of dust
lead the sampler picks up from the surface, for both techniques are unknown, and probably differ
substantially. There also is preliminary evidence that wipe samples tend to attain higher dust lead
loadings than do vacuum  samples.

      Table 2-4: Rate of Occurrence of Privately  Owned Occupied Housing  Units With Dust
Lead In Excess of the Federal Guidelines -  This table  shows that an estimated 17 percent of the
privately owned housing units in the United States have dust lead loadings exceeding the relevant HUD
guidelines, with the  highest lead  contamination found in window wells.  This finding supports
conclusions from other dust lead studies mat suggest window wells typically have the highest dust lead
loadings found in a home.

      Table 2-5:  Dust Lead Loadings by Location, With or Without Lead-Based  Paint - This
table examines the prevalence of dust lead l«arimgg above HUD's Interim Guidelines in homes that have
lead-based paint above 1.0 mg/cm2.  Sixteen percent of the homes with no lead-based paint  are
projected to have dust lead loadings exceeding HUD's  guidelines.  In homes with lead-based paint on
bom interior and exterior surfaces, it is estimated that 29 percent will exceed the dust lead  loading
criteria.  Although it appears from the table that the presence of lead-based paint contributes to higher
dust lead loadings, there are additional sources of lead in the environment to account for dust lead in
homes with no lead-based paint. To explore other potential sources of lead more thoroughly, EPA has
further analyzed the National Survey data to examine the associations between dust lead, soil lead, paint
lead,  automobile emissions, and other factors in a report titled Data Analysis of Lead in Soil.
Interested readers are referred to mis document for additional information.

      Table 2-6:  Association Between  Lead in  Interior  Surface Dust and Lead-Based Paint
Condition   This table presents the prevalence of high interior dust lead loadings in relation to  the
location and condition of lead-based paint.  High inaHingg occur more frequently in housing units with
nonintact lead-based paint than on other housing units.  This holds true for units with interior nonintact
lead-based paint, for units with exterior nonintact lead-based paint, a"d for units with nonintact lead-
based paint somewhere on the building. In each case, more homes with nonintact lead-based paint have
dust lead loading above HUD's guidelines than do homes with intact lead-based paint
 Window nils are defined by HUD as the lower part of die window inside the room. In common carpentry terminology this is more
 commonly called a window Bool.

 Window wells are defined by HUI
 this u more commonly called a window sill.
Window wells are defined by HUD as the bottom of a window between the screen and the glass. In common carpentry terminology
 Two examples are Landrigan, P, et al.: Epidemic Lead Absorption Near an Ore Smelter: The Bole of Particulaie Lead, V Eng J of
 Med 292: 123-9 (1975) and Parrel, MJL and Chisoim, JJ.: An Evaluation of Experimental Practices Jar Abatement of Residential
 Lead-Based Paint: Report on a Pilot Project, Eov Res 55:199-212 (1991).


 Data Analysis of Lead in Soil and Dust.  U.S. Environmental Protection Agency. EPA 747-R-93-011, September, 1993.



                                              2-6

-------
                                               TABLE 24

              RATE OF OCCURRENCE OF PRIVATELY OWNED OCCUPIED HOUSING UNITS WITH
                           DUST LEAD IN EXCESS OF THE FEDERAL GUIDELINES
Interior Sorface Dust Lead
TjtfJjfblMl
Unit Inferiors
Somewhere
Window well
Window nil (3)
Floor (?)
Window Only (2)
Interior Common Areas (3)
Federal
Guideline (1)
(ox/soft)

varies
800
500
200
varies
varies
Number (000) of
Housing Units
Abore Goidefine (1)

13,317
11,340
2,684
808
12408
1,249
Percent of
AboTeGudefined)

17%
15%
3%
1%
16%
2%
(1) HUD Interim Guidelines.
(2) Window includes window sill, window well or both.
(3) Categories with small sample sizes should be interpreted with caution.
                                             2-7

-------
                                          TABLE 2-5

             DUST LEAD LOADINGS BY LOCATION IN PRIVATELY OWNED OCCUPIED
                  HOUSING UNITS WITH OR WITHOUT LEAD-BASED PAINT (LBP)
Location of LBP
No LBP at AD
Interior LBP Only
Both Interior and Erterior LBP
Some Interior LBP
?|[mjl fVLlLLLmm A*M£ T f|9 f?\
Some P^Hor LBP
Some LBP
Dust Within Gmde&nMd)
Number (000)
10,681
7,074
16,267
38.855
3,354
45,898
53,181
Percent
84%
88%
71%
80%
93%
81%
83%
Dust Exeeedine Guideline* (1)
Number (000)
2,055
925
6,529
9.682
242
10.598
11.262
Pctccot
16%
12%
29%
20%
7%
19%
17%
(1) HOD Interim Guidelines.
(2) Categoric* with small sample sizes should be interpreted with caution.
                                          2-8

-------
                                              TABLE 2-6

                  ASSOCIATION BETWEEN LEAD IN INTERIOR SURFACE DUST AND
      LEAD-BASED PAINT (LBP) CONDITION FOR PRIVATELY OWNED OCCUPIED HOUSING UNITS
Location of LBP
No LBP
Interior LBP

Exterior LBP

Somewhere

Condition
of LBP
Intact
Non-Intact (2)
Intact
Non-Intact
Intact
Non-Intact
Dust Lead Within
Guidelines (1)
Number (000)
10,681
35,592
3,715
41,369
4,529
45,862
7,319
Percent
83%
82%
66%
88%
47%
91%
53%
Dust Lead Exceeds
Guidelines (1)
Number (000)
2,055
7,801
1,881
5,470
5,128
4,679
6,583
Percent
16%
18%
34%
12%
53%
9%
47%
Total
Number (000)
12,736
43,393
5,596
46,839
9,657
50,541
13,902
Percent
100%
100%
100%
100%
100%
100%
100%
(1) "Within Guidelines" means that surface lead dust does not exceed 200 ug/sq ft on floors, 500
    ug/sq ft on window sills, and 800 ug/sq ft on window wells. See HUD Interim Guidelines.

(2) A housing unit has non-intact interior LBP if there are more than 5 sq. ft. of damaged interior LBP.
    Similar definitions apply for "exterior" and "somewhere".

(3) Categories with small sample sizes should be interpreted with caution.

-------
      Table 2-7:  Association Between Lead in Soil and Exterior Lead-Based Paint Condition -
Presented are the estimated number of privately owned housing units with soil lead concentrations in
excess of 500 ppm by the presence and condition of exterior lead-based paint.  In order to examine
associations between varying characteristics measured in this data set, 500 parts per million (ppm) of
lead in soil was used to designate a threshold.

      The table suggests that housing units with exterior nonintact lead-based paint are more likely to
have high (above 500 ppm) soil lead concentrations than are other housing units.

      Table 2-8:   Estimated Number of Housing Units In Selected Lead Hazard  Categories -
Chapter  2 of the  Comprehensive and Workable Plan for the Abatement of Lead-Based Paint in
Privately Owned Housing: Report to Congress discussed the risks to children exposed  to lead-based
paint, nonintact lead-based paint, dust lead, and soil lead.  Table 2-8 displays the estimated prevalence
of these  hazards.  However,  as  noted above, there  are currently  no health-based standards to make
accurate hazard judgments. The standards used here are feasibility based interim guidelines which may
not directly apply to the types of samples collected during the National Survey.  Therefore, the priority
hazard numbers presented in Table 2-8 should be treated qualitatively and should not  be viewed as
definitive values.

      The table shows that of the 64 million homes with lead-based paint, an estimated 12 million are
occupied with families who have children under the age of seven years old (also presented in Table 2-1).
Of the 12 million, an estimated 4 million live in homes with dust lead loadings above HUD's guidelines
or in homes with nonintact lead-based paint present.  Both of these characteristics may pose significant
hazards to children.

2.1.2        Amounts of Lead Paint in Private Housing

      The previous section detailed the estimated prevalence of private dwelling units  in the United
States that have lead-based paint  somewhere on their surfaces.  This section presents national  estimates
of the square footage (interior and exterior) of surfaces covered with lead-based paint.  Painted surfaces
were sampled and quantified in  the National Survey using a number of methods, depending on the
component.  The methodology is described in detail in Appendix I, Design and Methodology, Section
3.7.

      Table 2-9:  Amounts  of Lead-Based Paint on Interior  Surfaces by Component/Substrate -
An estimated 29 billion square feet of painted interior surfaces are covered with lead-based paint.  This
represents 12 percent of die area of painted interior surfaces in pre-1980 privately-owned homes.  On
average,  each home with interim lead-based paint has approximately 601 square feet of interior lead-
based paint.  Although only 9 percent of the paint on walls, ceilings, and floors is lead-bated, those
components account for 62 percent of all interior, surface area lead-based paint.  Conversely, paint on
"shelves/other"  (shelves, cabinets, fireplaces, and closets)  is much more likely to be lead-based, even
though the total surface areas are much less. The separate breakdown by material substrate shows the
Wood and Drywall with the largest amount of lead-based paint.  This would be expected since Walls,
Ceilings, and Floors are typically made from these oubstrate materials.
 Thl» viliw wu derived from the EPA'i Inurim Outdone* on SaaUitMng Soil Lud Cbanup Ltvtli «! Sufutfrnd AIM, September 7,
 19S9 (OSWHR DiiMllv* 19355.4-02).



                                             2-10

-------
                                    TABLE 2-7

               ASSOCIATION BETWEEN LEAD IN SOIL AND EXTERIOR
      LEAD-BASED PAINT CONDITION FOR PRIVATELY OWNED HOUSING UNITS
                 (Numbers Represent Thousand! of Occupied Housing Units)
PrcMnttand
Condition of Exterior
Lead-Bawd Paint

NoLBP
LBP Present, Intact
LBP Present, Non-Intact
Some Exterior LBP
Total
Lead In SoD Somewhere
Within
Guideline! (1)
Number
17,719
35,914
3,815
39,729
57,448


Percent
91%
80%
44%
74%
79%
Exceeding
Guidelines (1)
Number
1,660
9,215
4,821
14,035
15,695


Percent
9%
20%
56%
26%
21%
(1) The guideline is 500 ppm.  See EPA, Interim Guidance.
                            TABLE 2-8

      ESTIMATED NUMBER OF PRIVATELY OWNED OCCUPIED
     HOUSING UNITS IN SELECTED LEAD HAZARD CATEGORIES
            (Numbers Represent Thousands of Housing Units)
I ftad VnTnird Categories
Lead-baaed Paint Present (1)
Lead-baled Paint Present and
Paint Non-Intact (2)
Lead-bawd Paint Present
and Lead Duit Present (3)
Lead-bated Paint Present : Paint Non-Intact,
OR Lead Duit Present
All Occupied
Housing Unite
64,443
14,354
11,262
19,030
Housing Units
With Children
12,427
3,321
1,676
4,025
(1) Lead-based paint concentration of at least 1.0 mg/sq cm.

(2) At least 5 squat* feet of defective lead-based paint.

(3) Lead in dust exceeds 200 ug/ sq ft for floors, or 500 ug/sq ft for window sills, or
   800 ug/sq ft for window wells.
                                        2-11

-------
                                           TABLE 2-9

             AMOUNTS OF LEAD-BASED PAINT (LBP) ON INTERIOR SURFACES
     BY PAINTED COMPONENT FOR PRIVATELY OWNED OCCUPIED HOUSING UNIT
                              (LBP Concentration  > = 1.0 mg/sq cm)




Component/Substrate
Components:
Walls/ceiling/floor
Metal component (2)
Non-metal component (3)
Shelves/other (4)
Totals
Substrates:
Wood
Metal (5)
Drywall or plaster
Concrete
Undetermined
Totals

National Total Amount of LBP

(millions of
sqft)

18,148
107
7,172
4,011
29.437

11,672
141
17,113
225
287
29.437
(percent of aB
D&JQDt OP COttlPOPy^Tf

9%
4%
24%
36%
1296

26%
4%
9%
3%
5%
12%
Amount LBP
Per Homing
Unit With
LBP a>
(square feet)

371
2
146
82
601

238
3
350
5
6
601
(1) Base equals the estimated 48,986,000 units with lead-based paint on interior surfaces.
(2) Includes metal trim, window sffls, molding, doors, air/heat vents, and radiators.
(3) Includes non-metal trim, window sills, molding, doors, and air/heat vents.
(4) Includes shelves, cabinets, fireplace, and closets, on any substrate.
(5) Metal substrate refers to any architectural component on metal substrate.

Note: Because of rounding, totals may not be exactly the same as the sum of the numbers.
                                      2-12

-------
       Table   2-10:     Amounts   of  Lead-Based   Paint   on   Exterior   Surfaces   by
Component/Substrate - Table 2-10 presents data on the prevalence of exterior lead-based paint by
architectural component and material substrate categories.  An estimated 49 billion square feet of
painted exterior surfaces are covered with lead-based paint.  This represents 44 percent of the area
of painted exterior  surfaces hi  pre-1980 privately-owned homes.   On average,  each home with
exterior lead-based paint has approximately 869 square feet of exterior lead-based paint.  The data
indicates there is more exterior surface area painted with lead-based paint man interior surface area
(see Table 2-9 above).  As expected, the component breakdown shows mat the walls account for 78
percent of all exterior lead-based paint.  Similarly, the breakdown by exterior material substrate
shows wood with the largest amount of lead-based paint (63 percent).

       Tables  2-11:   Amounts   of  Lead-Based  Paint  on  Interior  Surfaces  by  Selected
Characteristics - The table clearly shows that older homes, on average, have more lead-based paint on
interior surfaces than do newer homes.  The table also radicate that the distribution of lead-based paint
in single and multifamily dwellings is approximately the same (11 percent -12 percent).

       Tables  2-12:   Amounts  of  Lead-Based Paint  on  Exterior  Surfaces  by  Selected
Characteristics - The trend seen on interior surfaces (Table 2-11) is repeated on the exterior surfaces,
but is even more pronounced. On average, 70 percent of all exterior painted surfaces  on pre-1940
housing have lead-based paint.


2.1.3  Levels of Lead in Paint, Dust and Soil in Private Housing

       Table 2-13:   Arithmetic Mean Paint Lead Loadings by Characteristics - For both interior
surfaces and exterior surfaces, a clear trend is apparent in paint lead loadings (mg/cm2) from newer to
older homes.  Old lead-paint has more lead in it man newer lead-based paint  This is consistent with the
paint manufacturing trends, where the amount of lead added to paint has dropped since the 1940's.

       Two additional observations from the table are mat Northeastern homes  contain statistically
significantly higher lead  loadings on interior surfaces man on the rest of the nation's housing stock and
exterior lead-based paint contains  statistically significantly more lead than interior paint Because these
numbers are arithmetic means, however, they may be influenced by large values in the data and may
give misleading results.  For this reason, tables with geometric means (which approximate the median)
are given in Appendix A of this volume and should be consulted for supplemental information.  Table
A-8 (geometric means) in Appendix A of this document shows the same general trend as does Table 2-
13, but the differences between characteristics are not as dramatic.

       Table 2-14:   Arithmetic Mean  Paint Lead  Loadings  by Component/Substrate and
Construction  Year  - This table further breaks down Table 2-13's construction year category by
component and characteristic.  The  data shows the same trend, i.e., an increase in paint lead area
concentration from newest to oldest, especially on exterior walls. Geometric means of this data are also
presented  in Appendix  A, Table A-9, which provides additional useful  information.   Again, the
geometric means show the same general trends as does the arithmetic means.

       Table 2-15:  Unweighted Percentiles and Mean for Paint Lead Measurements by Sample
Location - Arithmetic means, standard deviations, m^iang  and other selected percentiles are provided
for the actual Scitec Metals Analysis Probe X-ray fluorescence device (MAP/XRF) measurements taken
at privately owned housing units.  The descriptive statistics (all unweighted) are grouped by sample
location (interior of unit, exterior of unit, and all common areas).  Two important findings outlined by
the table are that a substantial number of samples had no detectable lead, and the highest measurements
were recorded from exterior painted surfaces.
                                            2-13

-------
                                 TABLE 2-10

    AMOUNTS OF LEAD-BASED PAINT (LBP) ON EXTERIOR SURFACES
PAINTED COMPONENT FOR PRIVATELY OWNED OCCUPIED HOUSING UNITS
                    (LBP Concentration  > = 1.0 mg/sq cm)




Component/Substrate
Components:
Walls
Metal component (2)
Non-metal component (3)
Porches/other (4)
Totals
Substrates:
Wood
Metal (5)
Drywall or plaster
Concrete
Undetermined
Totals

National Total Amount of LBP

(minions of
sqft)

38,447
403
9,530
726
49.106

30,930
5,486
1,969
7,426
3,296
49.106
(percent of aD
paint on component)

49%
8%
41%
12%
44%

46%
33%
46%
40%
60%
44%
Amount LBP
Per Housing
Unit With
LBP a)
( square feet)

681
7
169
13
869

547
97
35
131
58
869
(1) Base equals the estimated 56,495,000 units with lead-based paint on exterior surfaces.
(2) Includes only metal windows, doors, soffit and facia, columns, and railings.
(3) Includes non-metal windows, doors, soffit and facia, columns, and railings.
(4) Includes porches, balconies, stairs, etc., on any substrate.
(5) Metal substrate refers to any architectural component on a metal substrate including
   aluminum siding OD
Note: Because of rounding, totals may not be exactly the same as the sum of the numbers.
                                    2-14

-------
                                  TABLE 2-11

         AMOUNTS OF LEAD-BASED PAINT (LBP) ON INTERIOR SURFACES
BY SELECTED CHARACTERISTICS FOR PRIVATELY OWNED OCCUPIED HOUSING UNITS
                     (LBP Concentration > = 1.0 mg/sq cm)
Characteristic
Construction Year
1960-1979
1940-1959
Before 1940
Housing Type
Single Family
Multi-family
One or More Children Under Age 7
Total pre-1980 boosing
National Total Amount of LBP
(milHniK of
sqft)

5,279
8.247
15,912

27,001
2,436
4,290
29,437
(percent of all
paint)

5%
13%
22%

12%
11%
10%
12%
Amount LBP
Per Housing
Unit With LBP
(square feet)

302
584
915

645
343
471
601
Number of
iPfmsiiiff t/Fii^y
With LBP
(OOOs)


17,483
14,113
17,392

41,884
7,104
9,112
48,986
                                      2-15

-------
                              TABLE 2-12

         AMOUNTS OF LEAD-BASED PAINT (LBP) ON EXTERIOR SURFACES
BY SELECTED CHARACTERISTICS FOR PRIVATELY OWNED OCCUPIED HOUSING UNITS
                     (LBP Concentration > = 1.0 mg/sq on)
Characteristic
Construction Year
1960-1979
1940-1959
Before 1940
Housing Type
Single Family
Multi-family
One or More Children Under Age 7
Total pre-1980 bousing
National Total Amount of LBP
sqft)

10402
12,635
25,969

46,216
2,890
6,127
49,106
(percent of all
paint)

23%
41%
70%

45%
31%
26%
44%
Amount IrRP
Per Housing
Unit With LBP
(square feet)

482
758
1,441

924
446
581
869
Number of
With LBP
(OOOs)

21,803
16,675
18,018

50,014
6,482
10,548
56,495
                                  2-16

-------
                              TABLE 2-13

ARITHMETIC MEAN PAINT LEAD LOADINGS IN PRIVATELY OWNED OCCUPIED
   HOUSING UNITS BUILT BEFORE 1980, BY SELECTED CHARACTERISTICS
                 (Paint Lead Concentration > = 1.0 mg/sq cm)
Characteristic
Total Occupied Housing
Units Buffi Before 1980
Construction Year:
1960-1979
1940-1959
Before 1940
Homing Type
Single Family
MuMfamily
One or More Children
UnderAze?
Census Region
Northeast
Midwest
South
West
Interior Surfaces
(mg/sq. cm.)
0.7 ( 0.4 .

0.3 ( 0.2 ,
0.5 ( 03 ,
1.4 ( 0.7 ,

0.7 ( 0.4 ,
0.4 ( 0.2 ,

0.7 ( OJ .

1.5 ( 0.6 ,
0.5 ( 0.2 ,
0.3 ( 0.2 ,
0.4 ( 0.2 ,
0.9 )

0.4 )
0.7 )
2.2 )

0.9 )
0.7 )

1.0 )

2.5 )
0.8 )
0.5 )
0.6 )
Exterior Surfaces
(mg/sq. cm.)
1.9 ( 1.3 ,

0.6 ( 0.3 ,
1.5 ( 0.9 ,
4.6 ( 2.6 ,

2.0 ( 1.3 ,
1.0 ( 0.3 ,

1.6 ( 0.5 .

2.4 ( 1.4 ,
2.1 ( 1.2 ,
1.7 ( 0.4 ,
1.2 ( 0.4 ,
2.5 )

0.8 )
2.1 )
6.5 )

2.7 )
1.6 )

2.7 )

3.4 )
3.0 )
2.9 )
2.0 )
 Note: Numbers in parentheses are 95% confidence intervals for the respective arithmetric means.
                               2-17

-------
                                            TABLE 2-14
ARITHMETIC MEAN PAINT LEAD LOADINGS IN PRIVATELY OWNED OCCUPIED HOUSING UNITS
      BUILT BEFORE 1980, BY ARCHITECTURAL COMPONENT AND CONSTRUCTION YEAR
                             (Paint I/pad Concentration > = 1.0 mg/sq cm)
Owrartwistir
Wafls/ccflmgs/floor



Metal (1)



Non-mettlCZ)



Other (3)




1960-1979
1940-1959
Before 1940

1960-1979
1940-1959
Before 1940

1960-1979
1940-1959
Before 1940

1960-1979
1940-1959
Before 1940
lulctJM1 Suffices
(ne/sq. cm.)

0.3
0.5
1.3

0.2
02
03

OA
OJ
2.7

0.1
02
\A

( 0.2 ,
( 0.2 ,
( 0.5 .

( 0.1 ,
( 0.1 ,
( 0.2 ,

( 0.0 ,
( 0.4 ,
( 1.7 ,

( 0.1 ,
( 0.1 ,
( 0.7 .

OA )
0.7 )
2.1 )

QA )
0.4 )
0.4 )

0.9 )
1.3 )
3.7 )

0.2 )
0.3 )
2.1 )
Extcnor Surfaces
ODDC/SQ. cm.)

0.5
1>1
6.2

02.
OA
1.1

0.8
2.6
5.0

02
0.8
2.1

( 02
( 0.6
( 2.6

( 0.0
( 0.1
( 0.1

( OA
( 1-2
( 2.7

( 0.0
( 0.1
( 02

, 0.8 )
, 23 )
, 9.9 )

, 03 )
, 0.7 )
, 21 )

, 13 )
, 3.9 )
, 7-2 )

, 0.3 )
, 1-5 )
. 4.0 )
        Note: Numbea in poeztheae* are 95% confidence intervals for the req>ective •fM"iirTik'
         (1) Include* metal trim, window sflb, molding, doon, an/betf venti, and ndiaton.
         CQ Include* noo-metal trim, window nDs, mokfing, dooa, and as/heat veati.
         (3) T""1*"^" ibehrea, cabinffi, fireplace, and cloaeU, on any nbatzate.
         (1) Include* only metal windowi, dooa, aoffit and facia, ctdnnmi, and nflingi.
         0) bebdea non-metal windows, dooa, aoffit and fteia, cofamna, and oufings.
         (3) Tr*1™1*" poxcfaea, balcooiet, atain, etc., on any mbitntc.
                                                   2-18

-------
                         TABLE 2-15
         PERCENTELES AND MEAN FOR PAINT LEAD (XRF)
MEASUREMENTS FOR PRIVATE HOUSING UNITS BY SAMPLE LOCATION
                       (UNWEIGHTED)
                (Paint Lead Concentrations in mg/sq cm)


*K^^«^^^«^^M
1%
5%
10%
25%
Median
7556
90%
95%
99%
Maximum

Mean

mid&ra Deviation
No. of Samples
Location
Interior
0.00
0.00
0.00
0.00
0.03
0.19
0.60
1.66
4.49
10.18
21.82

0.81
1.95
4,273
Exterior
0.00
0.00
0.00
0.00
0.05
0.42
1.85
5.81
9.30
27.71
53.81

2.07
4.64
1,047
^r^mi^pipfpfi nhrpjuB
0.00
0.00
0.00
0.04
0.18
0.70
2.20
5.54
8.74
19.69
19.69

2.10
3.70
218
                         2-19

-------
      Table 2-16:  Percentiles for Lead in SoQ Samples by  Sample Location - Unweighted
descriptive statistics, including arithmetic means, standard deviations and selected percentiles, are
presented for lead concentrations in soil samples.  Soil samples were collected from three locations on
the property of each dwelling unit a drip-line sample, collected at a point of about one foot from an
exterior wall of the dwelling, potentially contaminated with deteriorated lead-based paint; an entrance
sample collected near the most commonly used entrance^ to measure the potential for tracked-in lead;
and a remote sample, intended to measure background lead from sources other than lead-based paint
These locations are analyzed separately in this table, and overall statistics are presented.

      The concentrations at the locations near the housing unit (drip-line and entrance) are similar to
each other and are higher than the remote samples when the medians are compared.  This would be
expected if lead-based paint is contributing to sofl lead contamination. The arithmetic means which,
reflecting the skewness of the distribution, are substantially larger than the median, and reflect the same
trends (higher concentrations near the structures).  For more detailed analysis of me soil lead data
collected during the National Survey, readers are referred to EPA's report entitled Data Analysis of
Lead in Soil.

      figure 2-1:  Box Plot Example - Dust lead loading data is presented in box plot form.  For
readers not familiar with boxplots, a descriptive discussion is given below.

      Each boxplot shows a univariate data distribution, for example, the dust samples collected from
a specific location (e.g., entrance floor).  The box in the boxplot represents the middle 50 percent of the
data; the bottom of the box gives  the  25th  percentile; the top gives the  75th percentile;  and the
horizontal line inside the box gives the median. The vertical lines extending from the top and bottom of
the box reach to the largest and smallest observations, respectively, except for outliers.  Outliers are
plotted separately, as shown in Figure 2-1.  Data sets  approximating a normal distribution will produce
a symmetrical boxplot, and fewer than 1 in 100 observations will be classified as unusual.

      Figure 2-2: Boxplots of Dust Lead Loadings by Location - Because the data is approximately
log normal (skewed to the right), it is plotted on a log scale. In doing so, the data approximates a
normal distribution, reflected in the symmetry of the boxplots.  From this display of the data, it is
possible to visually compare lead loadings  in  all  of  the  sample locations  inside the dwellings
simultaneously.  Generally, the highest lead loadings are found in window wells, as discussed in Table
2-4.


2.1.4 Prevalence of Lead by Degree of Urbanization for Privately-Owned Housing

      The objective of mis section is to examine the association between the degree of urbanization and
the prevalence of lead in paint, damaged paint, dust, and sofl in privately owned housing.

      To accomplish the objective of mis analysis, the 150 census blocks in the 30 counties surveyed
(see Appendix I: Design and Methodology ) were each assigned to one of four urbanization categories
based on their 1980 population as reported by the Bureau of Census.  All housing units in a single
census block were assigned the  same urbanization category.  The four urbanization categories are
defined as follows:
                                            2-20

-------
                            TABLE 2-16

       PERCENTTLES AND MEAN FOR LEAD IN SOIL SAMPLES
       FROM PRIVATE HOUSING UNITS BY SAMPLE LOCATION
                   (Soil Lead Concentration in ppm)

Statistic
MrntTTnrni
1%
5%
10%
25%
Median
75%
90%
95%
99%
\fa-ri miim

Mean
Standard Deviation
Number of Samples
All
Locations (1)
1
3
6
12
23
54
152
519
1,188
4,127
22,974

324
1,207
768

Drip Line
1
1
6
11
23
60
201
810
1,476
10,674
22,974

448
1,766
249

Entrance
3
4
10
17
30
65
201
792
1,376
5,123
6,828

260
894
260

Remote
1
2
5
7
19
44
119
279
545
2,968
6,951

204
691
253
(1) Includes 6 samples taken on playgrounds which are not listed separately
  due to the g""!! sample size.
                              2-21

-------
                                               FIGURE 24

                                            BOXPLOT EXAMPLE
K)
 LOADING
(ug / sq ft,
log scale)
10000 •
1000
100 -
10
1
.1 •
On 4
.01 n
The box covers th«
center 50% of the
dale. — ,
\

. ^ 	 VBiy UIIU!
observat
^•M
•MM
. - 	 Unusual
T _
1 . i .
J
J^ -r-
rA
\
median
«••

»uw
ion *
X
•w*
X
B Whiskers show the
3 range of the data
/excluding me unusual
observations

.
•- 	 	 •-•• • 'i1-1 	 	 	 -•• • • n 	 1 	 — 	 1
First Second Third Fourth
Location Location Location, Location,
Symmetric Skewed
Data Data
                                                   LOCATIONS

-------
                  FIGURE 2-2

       BOXPLOTS OF DUST LEAD LOADINGS
   FOR PRIVATE HOUSING, INTERIOR LOCATIONS
100,000. •
10,000. •
1.000. •
inn .
1UU.
LOADING

(uglsqll, 10.
log scale)
I. -
0.1 -
0.01 -
nnni

<


•
L
<
	 1

> <



J |
> <
i 	

> s



I 1
>
i 	

>


{
•w
<
C
<
*"f




> ^
i 	
>
>
•* ^




X
<
i 	





7
>
i 	
I




-1





••
i 	 1
Dry Floor   Wot Floor
Entrywoy    Dry SKI

        LOCATIONS
W.tSffl
Dry WoK
Wot WoN

-------
       •     An area was considered a large city if it was located in a central city9 of a Primary
             Metropolitan Statistical Area (PMSA)/Metropolitan Statistical Area (MSA)10 with  a
             population was over one million.

       •     An area was considered a suburb of a large city if it was located in a PMSA/MSA with a
             population of over one million^ but was not located in a central city.

       •     An area was considered a small city if it was located in a PMSA/MSA with a population
             under one million

       •     An area was considered a rural area if it was not located in a PMSA/MSA.

       After each of the 284 sampled private housing units were assigned one of the four urbanization
categories, EPA studied the relationship between the prevalence of lead in paint, damaged paint, dust,
and soil and the degree of urbanization of the counties, along with region and construction year.

       Tables  2-17 and 2-18 present the number  of housing units in the sample,  by degree of
urbanization, construction year, and region. These tables show mat subdividing the sample by degree of
urbanization with construction year and region, the majority of cells have less than 30 housing units.
This is important when interpreting the results presented in the following urbanization tables.  For
example, in Table 2-27, 88 percent of pre-1940 small city homes have soil lead above 500 ppm, while
only 36 percent pre-1940 large city housing units have high soil  lead concentrations.  By examining
table 2-17, it can be seen that the 88 percent was based on only IS homes.  Because 15 is a small
number to scale-up results to national levels, no firm conclusions can be drawn.

       Although only cells with 10 or more housing units are presented in Tables 2-19 through 2-28,
caution is recommended when interpreting the results — point estimates, confidence intervals, and tests
of significance — for cells with few housing units represented.

       In many of the results discussed below, the difference between two  proportions is tested for
statistical significance at the .05 level  Thus the probability of a false positive (Le., finding a significant
difference when the proportions are the same) is .05 for any single comparison.  Since in each table
there are multiple comparisons that can be made, the probability of at least  one false positive, or the
probability of a false positive when comparing the largest observed proportion to the smallest, is greater
than 0.05 (the exact value depends on the number of comparisons and  on the correlations among the
various proportions).

       The analysis showed that mere were no significant differences in the prevalence of housing units
with nonintact paint, lead in  dust, and lead in soil by the degree of urbanization.  Differences were
noticed in the prevalence of lead-based paint between housing units located in large cities and suburb
areas versus housing units in small cities and rural areas.
9
 The largest city in each MSA is designated a "central city"; in addition there may be additional central cities if specified requirements
 are m** A more complete definition of "central city" can be obtained from the U.S. Office of Management and Budget.

10
 U.S. Office of Management and Budget current standards provide that an MSA is an area that includes at least one city with 50,000 or
 more inhabitant*, or a Census Bureau-defined urbanized area of at least 50,000 inh«hit«iit« and a total MSA population of at least
 100,000. OMB 1980 standards provide that within metropolitan complexes of 1 million or more population, separate component areas
 are defined if specified criteria are met. Such areas are designated PMSAs.  More complete definitions of "MSAs" and "PMSAs" can
 be obtained from OMB.
                                              2-24

-------
                                           TABLE 2-17

                        NUMBER OF PRIVATELY-OWNED HOUSING UNITS
                         IN THE SAMPLE BY DEGREE OF URBANIZATION
                                   AND CONSTRUCTION YEAR
Degree of Urbanization
Large City
Suburb of Large City
Small City
Rural Area
Total
Construction Year
1960-1979
28
30
34
28
120
1940-1959
33
19
19
16
87
Pre-1940
32
17
15
13
77
Total
93
66
68
57
284
                                          TABLE 2-18

                        NUMBER OF PRIVATELY-OWNED HOUSING UNITS
                         IN THE SAMPLE BY DEGREE OF URBANIZATION
                                         AND REGION
Decree of Urbanization
Large City
Suburb of Large City
Small City
Rural Area
Total
Re
Northeast
29
16
8
0
53
Midwest
14
13
10
32
69
pon
South
29
18
44
25
116
West
21
19
6
0
46
Total
93
66
68
57
284
 Hate: - "Luge cty* mctodes hoasmg unite located in cental abet where the PMSA/MSApopalitiaa is over
lonffiaa. 'SabaA of bage eay* includes bommg units looted ootsife ca^ cities btt in PMSAs/MSAs with popalttkn
overl mZbaQ. 'Smifl city' nclndet housing aria looted in PMSAs/MSAs wittt popnlitioo under 1 nrifficn.
•BnalAita* mctodeithoiehousiiijnnJuttjjtirenottaatedinaPMSAyMSA.
                                    2-25

-------
      The analysis also showed that there were significant differences in the prevalence of all the lead
characteristics by construction year and Census Region.  In most cases, these differences were not
dependent on the urbanization of the housing units.

      Tables 2-19 through 2-28 present the results of the analysis.  They are discussed below.

      Tables  2-19 and 2-20:  Estimated Number of Housing Units in the Nation  by Degree of
Urbanization, Construction Year, and Region - The tables show that the housing stock is relatively
evenly distributed across the four urbanization categories.  The percentages range from 21 percent for
housing units located in rural areas to 28 percent for units located in large cities.

      Tables  2-21 and 2-22: Percentage of Housing Units with Lead-Based Paint  by Degree of
Urbanization, Construction Year, and Region > Two sets of numbers  are presented in each cell.  The
first number is the estimated percentage of housing units nationally with the lead characteristic.  The
numbers in parenthesis are the 95 percent confidence limits. Results are presented only for cells with 10
or more housing units in the sample.  The  tables  show that the survey data indicated there is no
statistical  difference  among the four urbanization categories in the percentage of housing units with
lead-based paint

      Tables  2-23 and 2-24: Percentage of Housing Units with Damaged Lead-Based Paint by
Degree of Urbanization, Construction Year, and Region - Table 2-23 shows mat a higher percentage
of pre-1940 housing units have damaged lead-based paint when compared to post-1940  housing units.
Though a clear gradation in the  percentage of housing units with,  damaged  lead-based paint by
construction year is evident by visual observation, mere are no significant differences in the presence of
damaged lead-based paint by urbanization.

      Table 2-24 shows mat a higher percentage of housing units  located in large cities in the
Northeast and Midwest have damaged paint than in housing units in the South and the West  This is
probably due to the fact mat much of the housing stock in the cities in the Northeast and Midwest is
older man the housing stock in the cities in the South and West For housing units in suburbs of large
cities, there is no significant variation by region. Small sample sizes for the remaining two urbanization
Categories mal^ any conclusions impossible
      Tables 2-25 and  2-26:   Percentage of Housing Units with Dust Lead  Loadings above
Guidelines by Degree of Urbanization, Construction Year, and Region - Displayed are the results by
urbanization with respect to high levels of dust lead loading (see note on HUD's Interim Dust Lead
Guidelines in Section 2.1.1, Table 2-4).  Analysis of the two extreme urbanization categories-large
cities (25 percent) and rural areas (9 percent) shows a marginal difference at the .05 significance level.

      Table 2-25 shows a clear increase in lead loadings by construction year for all four urbanization
categories. Table 2-26 shows that the percentage of housing units from large cities and suburbs in the
Northeast with high dust lead leadings is significantly higher than the other three regions.

      Tables 2-27 and 2-28:  Percentage  of Housing Units with Sofl Lead above Guidelines by
Degree of Urbanization, Construction Year, and Region - Fewer dwelling units with soil lead levels
above the Federal guidelines are found in large cities man in other areas (Table 2-27), with most of the
difference evident in pre-1940 homes.  It is difficult to explain this  observation, especially since other
                                            2-26

-------
                                              TABLE 2-19

                                    ESTIMATED NUMBER 0000s) OF
                         PRIVATELY-OWNED HOUSING UNITS IN THE NATION
                      BY DEGREE OF URBANIZATION AND CONSTRUCTION YEAR
Degree of Urbanization
Large City
Suburb of Large City
Small City
Rural Area
Total
Construction Year
1960-1979
7,938
9,602
10428
7,618
35,686
1940-1959
6,613
5,305
4,914
3,641
20,473
Pre-1940
7,038
4,826
4,205
4,951
21,020
Total
21,589
19,733
19,647
16,210
77,179
Percent
28%
26%
25%
21%
100%
NOTE: Column totili are from tbe 1987 American Hooang Survey. Other entries tic cstmutci from the Nftkxul Survey of Lead-Based Brim in Hooting.
                                              TABLE 2-20

                                    ESTIMATED NUMBER ('OOOs) OF
                         PRIVATELY-OWNED HOUSING UNITS IN THE NATION
                             BY DEGREE OF URBANIZATION AND REGION
Degree of Urh"n«*»««»n
Large City
Snbnrb of Large City
Small City
Rural Area
Total
Re;
Northeast
7^61
6499
_ _
. _
16,963
Midwest
2,718
3,034
3,664
10,432
19,848
ejon
South
5,000
3,763
10,422
5,783
24,967
West
6416
6^42
_ .
. -
15399
Total
21495
19,738
19,647
16,215
77.179
Notes:
1. luge city'include* boosing null located m coOnl ctticc wbere the PMSA/MSA popohtion ij over 1 mfflSon. 'Suburb of tage city * mctadci housing
nuts located outride cental otiesbtrt in PMSAi/MSAj win popotation over InaOiaa.  'Small city* mctndo boosiDC nnii looted in FMSAs/MSAs wini
popobrioo mder 1 mflUoo.  Ibitil Area* klcndet 1ba*e hearing onij tint «e notkxated n iPMSA/MSA.
2. A'--repn»ertiicenwjlhfcMth«nlObottiintunki.
3. Cokmm look are firom the 1987 America Homing Survey. Other entriei ire eitnmlM from the Nitkml Survey of Lad-BuedPtint in Hooang.
                                       2-27

-------
                                                TABLE 2-21

                         PERCENTAGE OF PRIVATELY-OWNED HOUSING UNITS
                                WITH LEAD-BASED PAINT BY DEGREE OF
                               URBANIZATION AND CONSTRUCTION YEAR
Degree of Urbanization
Large City
Suburb of Large City
Small City
Rural Area
Total
Construction Year
1960-1979
84%
(63% -9756)
78%
(56% - 94%)
75%
(54% -91%)
69%
(45% - 88%)
76%
(63% - 87%)
1940-1959
85%
(66% - 97%)
98%
(83% - 100%)
91%
(69% - 100%)
94%
(72% - 100%)
92%
(82% - 98%)
Pre-1940
86%
(67% -98%)
67%
(37% -91%)
100%
(90% - 100%)
100%
(89% - 100%)
88%
(76% - 96%)
Total
85%
(73% -94%)
80%
(65% - 92%)
84%
(70% - 94%)
84%
(69% - 95%)
83%
(73% -91%)
                                               TABLE 2-22

                         PERCENTAGE OF PRIVATELY-OWNED HOUSING UNITS
                               WITH LEAD-BASED PAINT BY DEGREE OF
                                      URBANIZATION AND REGION
Degree of Urbanization
Large City
Suburb of Large City
Small City
Rural Area
Total
Re)
Northeast
86%
(66% -98%)
79%
(50% - 97%)
—
—
86%
(71% -96%)
Midwest
80%
(49% - 98%)
100%
(89%- 100%)
84%
(49% - 100%)
94%
(79% - 100%)
91%
(79% - 98%)
eion
South
97%
(84% - 100%)
79%
(52% - 97%)
83%
(66% - 95%)
68%
(43% -88%)
82%
(79% -98%)
West
75%
(49% - 94%)
72%
(44% -93%)
-
—
73%
(54% - 88%)
Total
85%
(73%-94J
80%
(65%-92J
84%
(70%-94»::
84% '
(69% -955
83% :
(73%-9lrf
 Note: — 'Ltzfc city' jnchidr^ bouimg units looted in centnl cities where the PMSA/MSA population is over
1 million. 'Suburb of luge city' includes bouimg units looted outside central "t"*« but in PMSAs/MSAs with population
over 1 million. 'Small city' include* bousing units looted in PMSAs/MSAs with population under 1 million.
•Run] Area'  mr.lndc* those housing units that are not located in a PMSA/MSA.
       — Numbers in parentheses are 95% ••"" intervals for the r*imarA percents.
For example, the 95 % confidence interval for the percent of bousing units with lead-based located in a
large city and contracted between 1960-1979 is 62% - 98%.
       — A *—* represents a cell with lets than 10 bousing units in the sample.
                                        2-28

-------
                                          TABLE 2-23
                      PERCENTAGE OF PRIVATELY-OWNED HOUSING UNITS
                   WITH AT LEAST 5 SQ. FT. OF DAMAGED LEAD-BASED FAINT
                   BY DEGREE OF URBANIZATION AND CONSTRUCTION YEAR
Decree of Uri»»n«»t*>n
Luxe City
Suburb of Large City
Small City
Rural Area
Total
Construction Tear
1960-1979
4%
(096 - 8%)
10%
(196-2856)
896
(096-2496)
896
(096-2596)
896
(296 - 17%)
1940-1559
1396
(256-3196)
556
(096-2496)
1396
(196-3796)
3896
(1256 - 6896)
1696
(6% - 29%)
Pre-1940
47%
(25% -69%)
15%
(1%-41%)
51%
(21% -81%)
46%
(16% - 78%)
40%
(25% -56%)
Total
21%
(10% -35%)
10%
(2% -22%)
19%
(8% -34%)
27%
(13% -44%)
19%
(11% -29%)
                                          TABLE 2-24

                      PERCENTAGE OF PRIVATELY-OWNED HOUSING UNITS
                   WITH AT LEAST 5 SQ. FT. OF DAMAGED LEAD-BASED PAINT
                          BY DEGREE OF URBANIZATION AND REGION
Degree of Urbanization
Large City
Suburb of Large City
Small City
Rural Area
Total
Re
Northeast
41%
(19% -6596)
11%
(0%-36%)
—
—
28%
(13% -46%)
Midwest
26%
(4% - 58%)
15%
(0%-45%)
88%
(55% - 100%)
31%
(13% -53%)
33%
(18% -50%)
Ekm
South
6%
(0%-22%)
11%
(0%-35%)
8%
(l%-22%)
19%
(4% - 42%)
10%
(3% -20%)
West
8%
(0% - 28%)
4%
(0% - 22%)
—
—
5%
(0% - 17%)
Total
21%
(10% -35%)
10%
(2% -22%)
19%
(8% - 34%)
27%
(13% - 44%)
19%
(11% -29%)
 Note: - T:«q«gy fretade« homing mriti looted m cental catiei where the PMSA/MSA popabttJoq it over
InriffioQ. 'Stab of tap cfy'incliidetboasinE into looted oolite
ovcrl ™ffi""  "Soon city* include* baaing vat* looted m FMSAt/MSAi wife popalnicn tmder 1 nffliaa.
•RanlAm*
      — Hmnbtn inpmmfan-t ttr 9*P <
For ample, the 95% confidence iatavil for the patent of hoQaae unin wife kaMaed loctfcd ia >
luxe city ad contracted between 19W-1979 is 0« -19*.
      — A'-'lepTMenti i eefl wilh leo thm 10 boDriDt noil in t
                                   2-29

-------
                                             TABLE 2-25

                       PERCENTAGE OF PRIVATELY-OWNED HOUSING UNITS
                      WITH LEAD IN DUST ABOVE GUIDELINES BY DEGREE OF
                             URBANIZATION AND CONSTRUCTION YEAR
Degree of Urbanization
Large City
Suburb of Large City
Small City
Rural Area
Total
Construction Year
1960-1979
6%
(096-33%)
0%
(056-5%)
4%
(0% - 17%)
2%
(0% - 14%)
3%
(0% - 10%)
1940-1959
12%
(2% - 30%)
8%
(0%-29%)
30%
(8% -58%)
6%
(0%-28%)
14%
(5% -26%)
Pre-1940
60%
(37% -81%)
60%
(31% -86%)
44%
(16% - 75%)
22%
(2% - 54%)
48%
(25% - 64%)
Total
25%
(13% -39%)
17%
(6% -32%)
18%
(7% -33%)
9%
(2% -22%)
18%
(10% -28%)
                                             TABLE 2-26
                       PERCENTAGE OF PRIVATELY-OWNED HOUSING UNITS
                              WITH LEAD IN DUST ABOVE GUIDELINES
                            BY DEGREE OF URBANIZATION AND REGION
Degree of Urbanization
Large Cky
Suburb of Large City
Small Cky
Rural Area
Total
Rei
Northeast
63%
(39% -84%)
47%
(19% - 76%)
—
—
55%
(37% -73%)
Midwest
19%
(l%-49%)
4%
(0%-27%)
27%
(3% -64%)
12%
(25% -30%)
15%
(5% - 29%)
pon
South
6%
(0%-22%)
0%
(0% - 15%)
8%
(l%-22%)
4%
(0% - 19%)
5%
(1%-13%)
West
2%
(0% - 16%)
3%
(0%-20%)
4%
(0%-41%)
-
3%
(0% - 13%)
Total
25%
(13% -39%)
17%
(6% -32%)
18%
(7% -33%)
9%
(2%-22%l
18%
(10%-28%1
 Note: -
ngumtslo
Imfflico. •Suburb of buje city' mctadef hi
tied in ccntni cities where the FMSA/MSA population is over
units located outside ctntnl cftiei but in FMSAi/MSAs wJh popalnkm
aval minion. "SmiD city" inclndei boosing units located in PMSAi/MSAi wilh popoliticn under 1 minion.
•BmlAra* ncfadei noie bonoc nab flntoe not located in* FMSA/MSA.
      - Number* in pjuuilhue* «te 95% coofi«taiceiGterv»lifortbee»timite
-------
                                             TABIE 2-27
                       PERCENTAGE OF PRIVATELY-OWNED HOUSING UNITS
                        WITH LEAD IN SOIL ABOVE GUIDELINES BY DEGREE
                           OF URBANIZATION AND CONSTRUCTION YEAR
Dfpf* of Urb>"»»q**"a
Large City
Suburb of Large City
Small City
Rural Area
Total
Construction Year
1960-1979
5%
(0% - 2096)
10%
(l%-28%)
1%
(Oft - 10%)
0%
(0% - 1196)
496
(096-1196)
1940-1959
5%
(096 - 19%)
0%
(0% - 8%)
18%
(2% -44%)
12%
(0%-38%)
8%
(2% - 18%)
Pre-1940
36%
(16% - 39%)
76%
(47% - 96%)
88%
(61% -100%)
55%
(25% - 85%)
60%
(44% - 75%)
Total
15%
(6% -27%)
23%
(10% - 39%)
24%
(17% -47%)
20%
(8% -36%)
20%
(12% -30%)
                                            TABLE 2-28

                       PERCENTAGE OF PRIVATELY-OWNED HOUSING UNITS
                        WITH LEAD IN SOIL ABOVE GUIDELINES BY DEGREE
                                  OF URBANIZATION AND REGION
Degree of Urbanization
Large City
Suburb of LareeCky
Small Cky
Rural Area
Total
Re)
Northeast
24%
C7%-46%)
47%
(19% - 76%)
23%
(l%-46%)
—
33%
(17% -51%)
Midwest
4%
(0%-26%)
20%
(2% -52%)
60%
(23% -91%)
27%
(10% -49%)
43%
(27% - 66%)
?ion
South
6%
(0%-22%)
0%
(0% - 15%)
16%
(4% -33%)
7%
(0%-25%)
11%
(4% -21%)
West
17%
(2% -41%)
16%
(2% -41%)
4%
(0%-41%)
—
15%
(4% -31%)
Total
15%
(6% - 27%)
23%
(10% -39%)
24%
(17% -47%)
20%
(8% -36%)
20%
(12% -30%)
 Note: — Tngc city" includes twang otto looted in cental dtiei where (he niSA/MSA population it over
1 miiiMi 'Sidmifc of ta^ci^iiKtodesbaoii^ into located oiiside central
overt "-IK™ -Snail city' include* bouse noils located in PMSA*/MSAi with population noder 1 nriffion.
         mcladea time honavinili that ate oat located m a F14SA/MSA.
      - Nmnben in vaiuitbtau ate 95% confidence imavaU for the estimated pereena.
For aaapk, the 95% confidence ioterral fat *e percent of bonone onto widx kMMMfedkxatedina
lute ci^y aid coMtractod between 1960-1979 it 0* - 21*.
      - A--"itpre»oiti«ceDwilJifc«»th«nl01i
-------
of studies cite soil in large cities with the highest levels of lead hi soil.11 This finding may be an
artifact the  survey methodology, hi which soil samples were taken by core samples.  No surface
scrapings were taken from paved surfaces.   Many of the sampled homes in large cities had  no
unpaved surfaces suitable for core sampling.

       Table 2-28, which  displays the estimated  percentage of housing units with lead in soil  by
urbanization and region shows that -the south has a significantly lower percentage of housing units with
lead in soil when compared to the Northeast and Midwest  Again  this may be related to the relative
ages of the housing stock in the different regions of the country.
2.2          Public Housing

       Below are the results of the public housing data analysis.  The presentation parallels the private
housing presentation.

2.2.1        Prevalence of Lead-Contaminated Paint, Dust and Soil in Public Housing

       Table 2-29:  Estimated Number and Percent of Housing Units with Lead-Based Paint by
Selected Characteristics - An estimated 86 percent (ie., about 782,000 units) of all public housing
units in the United States built before 1980 have lead-based paint somewhere in the building.

       "Somewhere" refers to lead-based paint on one or more of the following locations: the interior of
the unit- the exterior walls; or the common areas of multi-family structures. As with private housing, a
surface with lead-contamination is defined here, and by HUD, as having a measured paint lead loading
of 1.0 nag/cm2 or greater.

       Although the data collected during the National Survey suggests that older public imfa* are more
likely to have lead-based paint man newer units, the differences are not as great as those predicted for
the private dwelling units (see Table 2-1).  Because the sample sizes are small (Le., only 97 were units
were sampled) and stratified by construction  year, conclusions may not truly represent all public
housing units in the United States and readers are cautioned in their interpretation.

       Table 2-30:  Number  and  Percentage of Housing  Units with Lead-Based Paint by
Concentration and Sample Location - Table 2-30 shows the impact  of four different paint lead
loading thresholds on the prevalence of lead-based paint in public housing.  Included thresholds are the
Maryland, Federal, and Massachusetts standards. Also included are dwelling units with concentrations
of lead at 2.0 mg/cm2 or higher.  Similar to Table 2-2 for private hfni modifying  the threshold
concentration substantially changes the number of  dwelling units characterized as having lead-based
paint on interior painted  surfaces. However, the different thresholds have less of an  effect on the
prevalence of exterior lead-based paint man on interior paint

       Prevalence of Nonintact Lead-Based Paint  - Less than 10 public housing units in the sample
had  more than five square feet of nonintact lead-based paint   Because the sample size  was small,
projecting meaningful national estimates and analyzing relationships between nonintact lead-based paint
  One such study reporting higher sofl lead levels in inner cities was reported at the Trace Substances in Environmental HeaUh-XXV
 conference held in Columbia, Missouri, May 20-23, 1991 entitled Dun Control as a Means of Seducing Imer-Cfty Childhood Pb
 Exposure by H.W. Miclke et al.
                                             2-32

-------
                                            TABLE 2-29
                  ESTIMATED NUMBER AND PERCENT OF PUBLIC HOUSING UNITS
           BUILT BEFORE 1980 WITH LEAD-BASED PAINT, BY SELECTED CHARACTERISTICS
                                  (Paint Lead Concentration >= 1.0 mg/sq on)
Characteristic
Total Public Housing Units Built
Before 1980
Construction Yean 1960-1979
1950-1959
Before 1950
Total
Public Housing
Units (000)
910
100%
455
50%
273
30%
182
20%
Housing
WithLead-Bi
OUUICWUU& C 1
Percent
86%
( 78% - 94% )
79%
(66% - 92% )
90%
( 77% - 100%)
97%
( 88% - 100% )
Units
tsed Paint
2 RiriMfaip
Number (000)
782
(705 - 858 )
359
(299 - 419 )
246
(209 - 273 )
177
(160 - 182 )
Number of
Housing Units
in Sample
97
43
24
30
Notes: (1) Numbers in parentheses are 95% confidence intervals for the eam*mt percents and numbers.
       (2) Categories with small cample sizes should be interpreted with camion.
                                             2-33

-------
                               TABLE 2-30

     NUMBER AND PERCENTAGE OF PUBLIC HOUSING UNITS BUILT BEFORE
        1980 WITH LEAD-BASED PAINT BY LEAD CONCENTRATION AND
                     LOCATION OF LEAD-BASED PAINT
Location
Unit Interior
Interior Common Areas
BnQdine Exterior
Playgrounds
Somewhere in BnOding
Percentage of Homes
Paint Lead Concentration (mg/sq cm)
>=0.7
80%
44%
71%
12%
90%
>=1.0
75%
38%
68%
12%
86%
>=L2
70%
36%
67%
11%
85%
>=2.0
46%
31%
59%
77%
Location
Unit Interior
Interior Common Areas
RniMino F.vtprinr
Playgrounds
Somewhere in BnOding
Number of Homes (000)
Paint Lead Concentration (me/sq cm)
>=0.7
730
401
647
112
821
>=1.0
685
347
623
112
782
>=1.2
633
331
612
99
774
>=2.0
417
279
540
697
Note:
   A"—" indicates that there were less man 10 housing units in the sample
   with the lead characteristic.
                                     2-34

-------
 with dust and soil lead was not possible.  Furthermore, the dust lead data for the public housing
 units was  suspect because a large number of vacant apartments  were sampled.  Since dust lead
 loadings are a function of total dust present,  and because the unoccupied units were thoroughly
 cleaned prior to the sampling visits, the representativeness of the data is unknown.  Thus, scaling
 the results to project national estimates is not advisable.

 2.22  Amounts of Lead Paint in Public Housing

       The previous section reported the prevalence of public housing units in the United States that
 have lead-based paint somewhere on their surfaces. This section presents national estimates of how
 much surface area is covered with lead-based paint

       Table 2-31: Amounts of Lead-Based Paint on Interior Surfaces by Component/Substrate -
 Table 2-31 presents data on the prevalence of interior lead-based paint by architectural component and
 material substrate categories. An estimated 12 percent or 252 million square feet of all painted interior
 surfaces are covered with lead-based paint Twelve percent is also estimated for private dwelling units
 (Table 2-9). On average, each public housing unit with lead-based paint has approximately 447 square
 feet of interior lead-based paint  Although painted walls, ceilings, and floors account for more area,
 painted metal  components are much more likely to be lead-based.  The component breakdown shows
 that the ^alls/cefln^floors" component has 193 million square feet of lead-based paint accounting for
 approximately 76 percent of all interior lead-based paint  However, only  10 percent of the paint on
 walls, ceilings, and floors is lead-based.  Paint on  "metal components" (e.g., radiators, doors, air heat
 vents) is much more likely to be lead-based, even though the total surface areas covered with lead-based
 paint are far less.  The metal component only has one tenth the area of lead-based paint, but this
 represents 33 percent of all painted metal components.  The separate breakdown by material substrate
 shows the "Drywall" category with the largest amount of lead-based paint with 136 million square feet,
 or 54 percent of all interior lead-based paint

      Table 2-32: Amounts of Lead-Based Paint on Exterior Surfaces by Component/Substrate -
 Table 2-32 presents data on the prevalence of exterior lead-based paint by architectural component and
 material substrate categories. The data indicates there  is less exterior surface area painted with lead-
 based paint than interior surface  area. This is the opposite of private housing findings (Table 2-10).
 An estimated  S3 million square  feet of lead-based paint covers exterior  surfaces  (7 percent of all
 exterior painted surfaces on public housing), with an average of 214 square feet per public housing unit
 The component breakdown shows mat the non-metal components have 44 million square feet of lead-
 based paint accounting for 53 percent of all exterior lead-based paint  Non-metal components include
 such items as trim, window sills, doors, soffit, and fascia. The breakdown by material substrate shows
 that wood has the largest amount of lead-based paint with 40 million square feet, or 48 percent of all
 exterior lead-based paint on public housing.

 2.23 Levels of Lead in Paint, Dust and Soil in Public Housing

      Table 2-33:  Arithmetic Mean Paint Lead Loadings by Characteristics - For exterior painted
 surfaces, a clear trend is apparent in paint lead loadings (mg/cm2) from newer to older public housing
 units. Old lead-paint has more lead in it man newer lead-based paint  This  is consistent with the paint
 manufacturing trends, where the amount of lead added to paint has dropped since the  1940's. For more
 information, tables with geometric means  (which approximate the median) are given in the Appendix B
of tins document (Table B-4).   Although geometric means are the same for interior and  exterior
 surfaces, the arithmetic means indicate exterior surfaces have higher paint lead loadings.  This reflects
very high values measured on a few exterior surfaces which distort the arithmetic means (see Table 2-
 35).
                                            2-35

-------
                                           TABLE 2-31

              AMOUNTS OF LEAD-BASED PAINT (LBP) ON INTERIOR SURFACES
     BY ARCHITECTURAL COMPONENT AND SUBSTRATE FOR PUBLIC HOUSING UNITS
                                 (LBP Concentration  > = 1.0 mg/sq cm)



Component/Substrate

Components:
Walls/criling/flT
Metal component (2)
Non-metal component (3)
Shelves/other (4)
Totals
Substrates:
Wood
Metal (5)
Diywafl or plaster
Concrete
Undetermined
Totak

National Total Amount of LBP

(millions of
sqft)

193
22
32
4
252

35
23
136
51
7
252
(percent of an paint
on component/substrate)

10%
31%
19%
8%
12%

18%
29%
9%
21%
8%
12%
Amount LBP (1)
Per Housing
Unit With
Lead-Based Paint
(square feet)

282
32
47
6
367

51
34
198
74
10
367
(1) Base equals the estimated 685,000 units with lead-based paint on interior surfaces.
(2) Includes metal trim, window sills, molding, doors, air/heat vents, and radiators,
(3) Includes non-metal trim, window sDls, molding, doors, and ait/heat vents.
(4) Includes shelves, cabinets, fireplace, and closets, on any substrate.
(5) Metal substrate refers to any architectural component on metal substrate.

Note: Because of rounding, totals may not be exactly the same as the sums of the numbers.
                                        2-36

-------
                                                TABLE 2-32

                  AMOUNTS OF LEAD-BASED PAINT (LBP) ON EXTERIOR SURFACES
         BY ARCHITECTURAL COMPONENT AND SUBSTRATE FOR PUBLIC HOUSING UNITS
                                      (LBP Concentration  > = 1.0 mg/sq cm)



Component/Substrate

Components:
Walls
Metal component (2)
Non-metal component (3)
Other (4)
Totals
Substrates:
Wood
Metal (5)
DxywaU or plaster
Concrete
Undetermined
Totals

National Total Amount of LBP

(millions of
soft)

8
28
44
3
83

40
28
0
8
7
83
(percent of aD paint
on component/substrate)

1%
16%
15%
3%
7%

12%
14%
0%
3%
4%
7%
Amount LBP (1)
Per Housing
Unit With
Lead-Based Paint
(square feet)

13
45
71
4
133

64
45
0
13
11
133
(1) Base equals the estimated 622,860 units with lead-based paint on exterior surfaces.
(2) Includes only metal windows, doors, soffit and facia, columns, and railings.
(?) Includes non-metal windows, doors, soffit and facia, columns, and railings.
(4) Includes porches, balconies, stairs, etc., on any substrate.
(5) Metal substrate refers to any architectural component on a metal substrate
  including aluminum siding on exterior walls.

Note: Because of rounding, totals may not be exactly the same as the sums of the numbers.
                                               2-37

-------
      Table 2-34:   Arithmetic  Mean  Paint  Lead Loadings by Component/Substrate  and
Construction Year - This table further breaks  down Table  2-33's construction year category by
component/characteristic.  The data shows the same trend — an increase in paint lead loadings from
newest to oldest. For additional information on the geometric means of this data, refer to Appendix B,
Table B-5.

      Table 2-35:  Percentiles and Mean for MAP/XRF  Measurement Statistics by Sample
Location - Arithmetic means, standard deviations, and selected percentiles are provided for the actual
MAP/XRF measurements taken at public housing units.  The descriptive statistics are grouped by
sample location (interior of unit, exterior of unit, and all common areas).  Two important findings
outlined by the table are that a substantial number of samples had no detectable lead and the highest
measurements were recorded from exterior painted surfaces.

      Table 2-36:  Percentiles and Mean for Soil Lead  Measurement  Statistics by Sample
Location - Descriptive statistics,  including  arithmetic means,  standard  deviations and selected
percentiles, are presented for lead concentrations in soil samples collected during the National Survey.
Soil samples came from three locations on the property of each dwelling: a drip-line sample near an
exterior wall of the dwelling, potentially contaminated with deteriorated lead-based paint; an entrance
sample collected near the most commonly used entrance, to measure the potential for track-in lead; and
a remote sample, intended to measure background lead from sources other than lead-based paint  These
locations are analyzed separately in this table, and overall statistics are presented.

      Arithmetic mean concentrations are generally lower in the public housing  soil samples than in
soil samples collected at private housing sites (see Table 2-16). One evident cause  is  that the public
housing  distribution  is  tighter, without extremely  high  values.  In the private housing samples,
concentrations ranged from 1  to 22,000 ppm, and the data is very skewed to the right.  These large
values increase the private hanging arithmetic means. By examining the medians, however, the public
and private housing appear more similar.

      A major problem with the public housing soil lead data collected during the National Survey is
that most units (about 70 percent)  did not have exposed soil present to collect samples; most were
surrounded by pavement This is reflected in the  small number of samples collected at each location
(see Table 2-36). Therefore, the representativeness of public housing units with soil nearby to all public
housing units is unknown.  Since many public housing units are in inner cities, and soil in inner cities is
usually cited as having the highest average lead concentrations (although this was not observed in the
private housing data), it would be expected mat the soil samples collected from public housing should
be higher in lead than from private housing  This was not the case, however, and the small sample sizes
and/or the unknown representativeness of the data could be the reason.

      Another conclusion from the table  is that the three sampling locations for public housing are
much more similar to each other than the three locations for private housing.

      Figure 2-3:  Boxplots of Dust Lead Loadings by Location - See Figure 2-1 for examples of
boxplots. Because the data is approximately log normal (skewed to the right), it is plotted on a log
scale.  In doing so, the data approximates a normal distribution, reflected in the  symmetry of the
boxplots. From mis display of the data it is possible to visually compare lead loadings  from all of the
sample locations inside dwellings, simultaneously.  As with the private housing data, the highest lead
loadings are generally found in window wells.
                                            2-38

-------
                               TABLE 2-33
   ARITHMETIC MEAN PAINT LEAD LOADINGS IN PUBLIC HOUSING UNITS
           BUILT BEFORE 1980, BY SELECTED CHARACTERISTICS
Characteristic
Interior Surfaces
  (mg/sq. on.)
Exterior Surfaces
  (mg/sq. on.)
Total Public Hmisfr'g
 Units Buflt Before 1980
0.4 (  0.3   0.5 )
                                                         1.2 f  0.4
          1.5  )
Construction Yean
           1960-1979
           1950-1959
           Before 1950
0.4 (  0.2   0.5 )
0.4 (  0.2   0.5 )
0.5 (  0.3   0.8 )
03 (  0.1   0.5  )
1.1 (  0.1   2.1  )
2.3 (  0.5   4.1  )
                               TABLE 2-34
    ARITHMETIC MEAN PAINT LEAD LOADINGS BY PAINTED COMPONENT
          AND CONSTRUCTION YEAR FOR PUBLIC HOUSING UNITS
Chflrarteristv
Interior Surfaces Exterior Surfaces
(mg/sq. cm.) (mg/sq. cm.)
'WalklrfStrngelflnnr

Metal
Non-metal
Other
1960-1979
1950-1959
Before 1950
1960-1979
1950-1959
Before 1950
1960-1979
1950-1959
Before 1950
1960-1979
1950-1959
Before 1950 	
0.4 ( 0.2
OJ ( 0.1
0.5 ( 03
0.5 < 0.1
1.3 ( 0.5
1.0 ( 0.6
0.4 ( 0.2
0.4 ( 0.2
0.7 ( 0.2
0.2 ( 0.1
0.4 ( 0.0
03 ( 0.0
0.5 )
0.5 )
0.8 )
0.8 )
2.0 )
1-4 )
0.7 )
0.7 )
1-3 )
03 )
0.9 )
0.7 )
0.3 ( 0.1
0.9 ( -0.1
1.0 ( -1.1
0.3 ( 0.0
0.6 ( -0.1
2.9 ( -1.1
03 ( 0.0
2.8 ( -03
5.4 ( 1.6
1.1 ( -0.2
0.3 ( -0.1
1.4 { -0.7
0-5 )
1.3 )
6.9 )
0.6 )
13 )
6.9 )
0.6 )
5.8 )
9.3 )
2.4 )
0.6 )
3.5 )
                                   2-39

-------
                 TABLE 2-35
            AND MEAN FOR XRF MEASUREMENTS
FOR PUBLIC HOUSING UMTS BY SAMPLE LOCATION
                (UNWEIGHTED)
          (Paint Lead Concentrations in mg/sq cm)


Braiuft Blum
1%
5%
10%
25%
Median
75%
90%
95%
99%
mj* • •••

Mean
Std. Der.
No. of Samples
Location
Interior
0.00
0.00
0.00
0.00
0.05
0.21
0.68
1.74
2.64
4.78
12.76

0.58
2.81
1,731
Exterior
0.00
0.00
0.00
0.00
0.00
0.14
0.72
3.44
7.18
22.52
34.50

1.22
9.64
267
Common Areas
0.00
0.00
0.00
0.00
0.06
0.31
1.08
2.42
4.58
23.28
23.96

1.17
7.898
553
                  2-40

-------
                  TABLE 2-36

PERCENTELES AND MEAN FOR LEAD IN SOIL SAMPLES
FROM PUBLIC HOUSING UNITS BY SAMPLE LOCATION
           (Soil Lead Concentrations in ppm)

Mean
Minimum
1%
5%
10%
25%
Median
75%
90%
95%
99%
M-
No. of Samples
All Locations
92
5
5
9
12
20
39
126
206
424
753
753
• «/••
89
Drip Line
104
8
8
9
15
25
47
186
438
483
527
527
•/«# I
28
Entrance
112
11
11
14
15
23
49
167
265
499
872
872
26
Remote
79
6
6
9
11
25
49
101
219
243
615
615
29
                   2-41

-------
                                                       FIGURE 2-3

                                            nOXPLOTS OF DUST LEAD LOADINGS
                                        FOR PUIILIC HOUSING, INTERIOR LOCATIONS
K
1UU.UUU. -
10,000.
1,000.
100. -
LOADING
lug/spit, 10. •
* w * • *
log scale)
1. •
0.1

o.ot
nnni •









_






•
•
rh
1 H
J I
1
o
L O
0 0
	 1 	 1 	 1 	
O

— p- —p.
I J











1 B
T L
- I
*-JL«





O

••
_l 	 1 	 1 	 1 	 1
                                Dry Floor    Wat Floor
Entryway    Dry Sill     Wat 8111

        LOCATIONS
Dry Wefl     Wet Wed

-------
                3. SOURCES OF ERROR IN THE NATIONAL SURVEY DATA


      An evaluation of data quality is necessary in order to assess the utility of the survey data far lead
research and policy development  It is the errors in the data that generally determine data quality.  In this
context, "error" refers to deviations of obtained survey results from those that are true reflections of die
population. These errors  are a function of the processes that generated the data at the varies stages of the
survey: sample design, sample selection, field sampling and data collection, laboratory measurement, data
processing, and data analysis.

      This chapter describes the quality of the data collected from the national survey of lead-based paint
and the statistical techniques used to identify and measure error. Much of the chapter focuses on errors due
to the MAP/XRF equipment used to detect and quantify the amount of lead in painted surfaces. As will be
seen, it is possible to adjust the data to "correct11 for these errors.  Section 3.2 reports a detailed analysis of
nonresponse and other potential sampling biases of estimates for both private and public housing.  Section
3.3 examines measurement errors for MAP/XRF devices. Section 3.4 presents a bias and variable analysis
of the MAP/XRF measurement errors on classifying homes with lead-based paint  Section 3.5 explores
classification error due to incomplete sampling of painted surfaces within dwelling units. Finally^ Section
3.6 looks at the quality of the laboratory measurements and the effects of small dust sample weights on the
3.1   Statistical Concepts and Terminology

      An error is simply the difference between the sample estimate and the population parameter that we
wish to estimate.  We can talk about errors for a single measurement, or for an average based on many
observations. For example, suppose the national average of lead in paint on wet room window sills is 0.8
mg/sq cm. ffasingtemeasurernenttorawetroomwindowsiUis 1.5m
cm.  Similarly, if the (weighted) average of all measurements taken from window sills in wet rooms is 0.6
mg/sq cm, then the error is -0.2 mg/sq cm.

      There are two types of errors: bias and variable error.

      •     Bias - is a  constant  error because an possible surveys  using the same  design would
            overestimate  (or underestimate) the population parameter, on the average. Biases arise from
            a  number of sources,  including differences between  the sample  frame and the target
            population, differential  response rates from different census blocks (i.e,  segment) of the
            sampled population,  uncalibrated  or  mis-calibrated  field  or  laboratory measurement
            equipment, and some types of data-reduction procedures.

      •     Variable Error -  is an error caused by the random variation inherent in any sampling or
            measurement process. A variable error is specific to each measurement and generally cannot
            be estimated or statistically corrected and remain in the data even if all systematic errors have
            been eliminated. Variable errors result from both sampling and measurement processes, and
            are typically reported as a variance or standard deviation.

      •     Precision - refers to the size of the variable error. If the variable error is small, we say mat
            the estimate is precise.
                                              3-1

-------
      Throughout this chapter we will make use of the following term:

      •      Sample Weight - is the number of housing units in the target population that a sampled unit
             represents.  The sample weight can be calculated by taking the inverse of the probability of
             selection for that unit Thus, if the probability of selection is .01, the sample weight is 100.
             With multi-stage samples, the overall probability of selection is the product of the conditional
             probabilities of selection at each stage.
3.2   Response Rates and Potential for Non-Response Bias

      The Comprehensive and Workable Plan for the Abatement of Lead-Based Paint in Privately
Owned Housing: Report to Congress includes a brief discussion of response rates in its Appendix A. That
Appendix reports national response rates by construction year stratum for single and multi-family housing
units

      The objective of the analysis described in this section is to estimate the potential impact of national
survey nonresponse on the estimated prevalence of lead-based paint in housing.  To accomplish this, a
detailed analysis of response rates at each stage of the survey was conducted for each of the  150 census
blocks that were  surveyed (see  Appendix I for a description of the  survey methodology, including
definitions of terms).  The analysis looked at the relationship between response rates and factors such as
ethnicity, geographic location and economic measures of wealth (rent, home value, and income) that might
be related to response rates. In addition, the analysis studied the association between housing units in the
same or nearby census blocks with respect to the presence or absence of lead in paint, dust, and soil

3.2.1 Private Housing

      For private housing, no statistically significant relationship was observed between response rate and
ethnicity, income or age of the housing units.  In addition, there was a strong positive association between
inspected housing units in the same census block with respect to the presence or absence of lead in paint,
dust, and soil.  On the other hand, the lowest rent category homes (< $200/month) and the highest market
value homes  (<$ 150,000)  appear  to  be  slightly  under  represented, and  the  South  is somewhat
overrespresented  in the  sample.   On balance,  these findings suggest mat the  potential bias due to
nonresponse is likely to be small. Therefore, the estimates of lead-based paint prevalence were not adjusted
for non-response.

      The first step in the non-response analysis was the calculation of the rates  of being successfully
contacted, being eligible for inclusion in the survey, and being a respondent in the surveys for each census
block.  Then these rates were analyzed by region and economic variables using frequency distributions,
Pearson correlation coefficients, and tabulations of rates  versus the different variables. The results of the
analysis are presented in Tables 3-1 through 3-5.

      Table 3-1 shows national contact rates, response rates, and eligibility rates, where applicable. These
rates can be viewed  as conditional success rates  — each  were calculated using only those potential
respondents who reached mat particular stage.

      The meaning of the term ''complete" varies  with each stage  of data collection.   A case was
considered completed  at the screener stage if an interview was completed, regardless of eligibility- Thus,
the screener response rate measures the rate at which the  screening interview reached its logical conclusion
(eligible or ineligible), and did not terminate due to other causes (refusals, break ofis,  etc.).   At the
                                               3-2

-------
telephone interview stage, only eligible respondents who completed interviews and scheduled appointments
were considered "completes."  Finally, a case was considered "complete" at the inspection stage if an
inspection was completed

      A "refusal" occurred during the telephone interview when an eligible respondent refused to schedule
an appointment for an inspection. A "refusal" occurred during the inspection when an eligible respondent
refused to schedule an appointment for an inspection, or scheduled an appointment and then refused the
field team entry into the housing  unit  A "break off1  occurred when the field team was refused an
inspection after the start of the inspection interview.

                                           TABLE 3-1

     NATIONAL RESPONSE, CONTACT, AND ELIGIBILITY RATES AT EACH DATA
                                    COLLECTION STAGE
Data Collection Stage
Screener
Telephone Interview
Inspection
Contact Rate
77%
83%

Response Rate
63%
55%
90%
Eligibility Rate
89%


      National rates in the table were calculated as follows:
               ScronerCcnUctRjte- (Total ytoeniptj- Not alHoo^-Vacmty^

               Screener R«p«ge Rite «= (Complrt«y(ToUl Attempts - Vacant Hornet)
               Scraaer Eligibility Rale - (Eligibk*y(Cfluapleto)
               Tdqfe
               Tdepb
i Interview Contact Rate - (Completes+Language Problems + Itefiialsy(Tottl Attempts)
t Interview Response Rtte - 
-------
                                         TABLE 3-2

CENSUS BLOCKS LOST BY CENSUS REGION AND SURVEY DATA COLLECTION STAGE
Region
Northeast
Midwest
West
South
Total
Segments in
Regions
40
35
25
50
150
Survey Data CoDection Stage
Listing
2
1
1
1
5
Screening
2
0
2
1
5
Telephone
Interview
6
3
4
5
18
Inspection
1
2
1
3
7
Total
11
6
8
10
35
       One important measure of the respreseativeness of the National Survey is to examine how the
distributions of the housing characteristics and socioeconomic and ethnic factors in the National Survey
compare to national distributions. National distributions were obtained from the American Housing Survey
(AHS) for 1987 performed by the Bureau of the Census and the Department of Housing and Urban
Development The distributions of ethnicity, income, building age, region of the country, monthly rent and
market value from the National Survey were compared  to their respective  national distributions and
presented in Table 3-3.  Cbi-square tests were used to determine how the distributions in the National
Survey compared to those from the American  Housing  Survey.   In  most  cases, the  distribution  of
households in the National Survey were not significantly different from those in the American Housing
Survey.  No significant differences were observed in the distributions of ethnicity,  building age,  or
household income. The few cases where there were significant differences involve the following.  There
were fewer tenant-occupied housing units with lower rents (rent less than $200 a month) man expected
given national estimates; more expensive owner-occupied housing (market value greater then $150,000)
than expected given national estimates; and more dwelling  units located in the  South than expected given
national estimates.
                                            3-4

-------
                                        TABLE 3-3

CHI-SQUARE RESULTS FOR DEMOGRAPHIC AND SOCIOECONOMIC VARIABLES
       a.  Ethnicity
Ethnicity
Observed frequency in National
Survey
National distribution, from AHS
(000)
Expected frequency, from AHS*
Individual Chi-square values*
Black
29
9,261
33
0.508
Hispanic
24
4,977
18
2.169
Other1
230
64,943
232
0.019
"The chi-square statistic was calculated assuming a fixed total of 283 homes with data on race (3 cells and
2 degrees of freedom).

Total Chi-square statistic              2.812
P-value with 2 degrees of freedom      0.245
i
 •flrtigr nr* {tigtaiW tmn-Htrpamf mfait»^ faifnr ««xl V*riRf Tel«tu4»f* VJetmns «nH Amfr^rrmn Tmti«n» «nH nfarr TtanMtrle*
       b.  Building Age
Building Age
Observed frequency in National
Survey
National distribution, from AHS
(000)
Expected frequency, from AHS**
Individual Chi-square values**
pre-1940
77
21,215
76
0.011
1940 to 1959
87
21,001
75
1.810
1960 to 1979
120
36,965
133
1.194
**The chi-square statistic was calculated assuming a fixed total of 284 homes with data on building age (3
cells and 2 degrees of freedom).

Total Chi-square statistic              3.015
P-value with 2 degrees of freedom      0.221
                                            3-5

-------
       c. Region
Region of the Country
Observed frequency in National
Survey
National distribution, from AHS
(000)
Expected frequency, from AHS**
Individual Chi-square values**
Northeast
52
17,618
63
1.911
Midwest
69
20,344
73
0.190
South
116
25,589
91
6.585
West
46
15,628
56
1.740
*The cbi-square statistic was calculated assuming a fixed total of 283 homes with data on region (4 cells
and 3 degrees of freedom).

Total Chi-square statistic              10.423
P-value with 3 degrees of freedom       0.015
       d. Household Income
Household Income
Observed frequency in National
Survey
National distribution, from AHS
(000)
Expected frequency, from AHS**
Individual Chi-square values**
< $10,000
51
15,482
53
0.050
$10,000 to
$19,999
49
17,090
57
1.062
$20,000 to
$29,999
56
15,102
50
0.680
> $30,000
107
31,147
103
0.121
**The chi-square statistic was calculated assuming a fixed grand total of 263 homes (4 cells and 3
of freedom).

Total Chi-square statistic              1.913
P-value with 3 degrees of freedom       0.590
                                              3-6

-------
       e. Monthly Rent
Monthly Rent
Observed frequency in National
Survey
National distribution, from AHS
(000)
Expected frequency, from AHS**
Individual Chi-square values**
<$200
12
5,886
22
4.436
$200-$399
47
12^30
45
0.057
>$400
40
8,560
32
2.132
"The chi-square statistic was calculated assuming a fixed grand total of 105 homes (3 cells and 2 degrees
of freedom).

Total Chi-square statistic              6.625
P-value with 2 degrees of freedom      0.036
       £ Current Market Value
Current Market Value
Observed frequency in
National Survey
National distribution,
from AHS (000)
Expected frequency, from
AHS**
Individual Chi-square
values**
<$40,000
39
11,885
41
0.051
$40,000-
$59,999
21
10,228
35
5.472
$60,000-
$79,999
25
9,173
31
1.235
$80,000-
$99,999
16
5,582
19
0.471
$100,000-
$150,000
29
6,281
21
2.724
>$150,000
42
7,405
25
11.211
"The chi-square statistic was calculated assuming a fixed grand total of 172 homes (6 cells and 5 degrees
of freedom).

Total Chi-square statistic              14.406
P-value with 5 degrees of freedom      0.013
                                               3-7

-------
      Table 3-4 shows the distribution of the overall Census block response rate by region. Census blocks
in the Northeast had the lowest overall response rate where only 35% of the census blocks had response
rates over 25 percent In contrast, census blocks in the Midwest and South had Ihe highest overall response
rates where 66% and 68% of the respective census blocks had response rates over 25 percent


                                          TABLE 3-4

  NUMBER AND PERCENT OF CENSUS BLOCKS BY OVERALL RESPONSE RATEi AND
                                      CENSUS REGION
Region
Northeast
Midwest
West
South
Total
Segments with Indicated Overall Response Rate
Overall Response Rate
Equal to 0%
11
(28%)
6
(17%)
8
(32%)
10
(20%)
35
(23%)
0% and <
25%
15
(37%)
6
(17%)
4
(16%)
6
(12%)
31
(21%)
> = 25% and
= <75%
13
(33%)
21
(60%)
12
(48%)
29
(58%)
75
(50%)
More than
75%
1
(2%)
2
(6%)
1
(4%)
5
(10%)
9
(6%)
Total
40
(100%)
35
(100%)
25
(100%)
50
(100%)
150
(100%)
  The Overall Response Rate = (Scnoier Iteqwose RjteXTde^tooe Response Ra^XInqiection Respoose Rale)
      Table 3-5 addresses the question: "If one housing unit in a segment has (or does not have) a
particular lead characteristic, then do all the other housing units in the same segment of similar age (year of
construction) have the same lead characteristic?" For purposes of mis table, census blocks with less than
two housing units inspected were excluded.  Fifty percent of the census blocks surveyed had two or more
inspected housing units.

      The first column in Table 3-5 displays the number of census blocks where all housing units in a
segment had the same lead characteristic.  The last two columns break down census blocks where mere
were differences into two categories: census blocks where there were differences within houses in the same
construction year category, and census blocks where differences could always be explained by construction
year category.  Table 3-5 shows that lead characteristics were the same for most census blocks for each of
the four types of lead characteristics analyzed, irrespective of the year of construction. For example, if we
look at the characteristic "lead-based paint" we find that all houses within a segment had the same lead
characteristic  for 64 percent  of the census blocks.   Of the remaining 36 percent where there were
differences, 8  percent is explained by construction year category.  Differences  in  lead-based paint
characteristic between house built in different year categories were found in only 28 percent of the census
blocks.
                                             3-8

-------
                                           TABLE 3-5

ASSOCIATION BETWEEN INSPECTED HOUSING UNITS IN THE SAME CENSUS BLOCKS
  WITH RESPECT TO THE PRESENCE/ABSENCE OF LEAD IN PAINT, DUST, AND SOIL
Presence or Absence
of Lead
Lead-based paint
Lead in dust above
guidance
Lead in soil above
PIHQfl^Hyfi
Damaged lead-based
paint
Segments1 Where All Housing
Units Have the Lead
Characteristics or AH Housing
Units Do Not Have the Lead
Characteristics
48 (64%)
61 (81%)
59 (79%)
51 (68%)
Segments1 Where Some Housing Units Have
and Some Do Not Have the Lead
Characteristics
All Housing Units in
the Same Construction
Year Category2 Have
the Same Lead
Characteristics
6 (8%)
3 (4%)
10 (13%)
9 (12%)
Some Housing Units in
the Same Construction
Year Category2 Have
Different Lead
Characteristics
21 (28%)
11(15%)
6 (8%)
15 (20%)
      Note: •    The resalfc are based on unai^ustedXIU7 readings of 1.0 rng/sq on or greito.
               Of the 150 segments tint were surveyed, 75 segments bad two or moee housing unite in the segment
                                                 nngtfc* f**1 """*••• nF*vffnr*lx mth tarn cr mrve durglltng
1   The percentage of segments in the three columns were calculi
    units as the base.
2   Housingumts were grouped into 3 categories based on cwstrochon year Rre 1940,1940-1959, and 1960-1979.
3   At least 5 square feet of damaged lead-based paint
3.2.2  Public Housing

      For the public housing component of the national survey of lead-based paint, a sample of 110
projects from a national frame of public housing projects12 was drawn according to the design described in
Chapters 2 and 3 of Appendix I of this report. The survey design specified a visit to one randomly selected
housing unit fiom each of these 110 projects.  The survey was completed in 97 of the 110 sampled projects.
Ofthe 13 nonrespondent projects, eight were excluded because they were found to be out of scope - they
either had no family units, did not exist, or were built since 1980.  The remaining five projects were not
completed because of problems encountered in scheduling interviews during the short period when the field
team was in town.

      The nation's public housing projects were well represented in the sample.  The response rate was 95
percent (97 completes among 102 eligible projects). Despite the fact that me sample design was unbiased
and there was no evidence of bias in the selection of the projects, vacant housing units appeared to be over-
represented in the sample.  Of 97 eligible units, 44 were vacant Such a large number were undoubtedly
12 A public bousing project is a complex of bousing nails developed and built at the same time, cri
together fir administrative process.
                                                             nits acquired fiom different locations bat grouped
                                               3-9

-------
vacant because field technicians were sometimes steered to vacant units selected by the public housing
project manager.

      Although a comparison of the prevalence of lead in paint between vacant and occupied housing units
showed no significant differences, the impact on other target parameters is unknown.  For example, it is
difficult to estimate the percentage of public housing units with both lead-based paint and children under
the age of seven because so few occupied apartments were sampled.  Furthermore, dust samples collected
in vacant apartments probably do not represent samples collected in occupied apartments.  Because the
public housing sample sizes were small and the occupied housing unit sizes were even smaller, definitive
conclusions resulting in estimates of dust lead levels based on the national survey data are not possible and
are not recommended.

      The sample was carefully designed to be representative. A sample is said to be representative if the
distributions of its characteristics are about the same for the sample as they are for the target population.
Because of sampling error, perfect agreement is not expected.  The design for the national survey  is
described in Appendix I, Chapter 3, Sample Design and Selection.  The design has the property of being
statistically unbiased. That is, every public housing project in the nation had a known, positive probability
of being selected into the sample.  The sample weights were calculated to properly reflect these varying
probabilities.

      We can assess representativeness by comparing the characteristics of the weighted sample to the
characteristics of the target population.  Tables 3-6, 3-7,  and 3-8 present tabulations of the public housing
frames and sample by construction year, census region,  and size of the public housing authority (PHA),
respectively. The first column of all three tables is the same and shows the steps 'mat were taken to develop
the sample from the target population:

      •    Full National Frame —  is the inventory of HDD's public housing units data file  that
            constitutes the target population.

      •    County Extract, Unedited — refers to the 30 primary sampling units (or PSUs) mat were
            selected from the full national frame.

      •    County Extract, Updated -  refers to a revision to the number of housing units in the
            sampled counties that was made  after contacting the PHAs.  Details  concerning these
            sampling procedures are contained in Appendix I, Chapter 3 of this report.

      •    Final Sample — refers to the sample drawn from the 30 country updated extract.

The projected  figures given for the county extracts and the final sample were obtained by applying the
sample weights to the data.

      We can assess the representativeness of the sample in Tables 3-6, 3-7, and 3-8 by comparing the
actual percentages for the full national frame to the projected percentages shown for the final sample.
When examined by construction year  category (Table 3-6), we see that the distribution of the sample
projections is similar to the distribution for the full national frame. When examined by geography (Table
3-7), the sample projections are relatively high in the Northeast (44% versus 31%) and low in the Midwest
(11%  versus  17%)  and the  South  (34% versus  41%),  compared  to  the  figure for the  full
                                              3-10

-------
                           TABLE 3-6

   DISTRIBUTION OF PUBLIC HOUSING FAMILY UNITS
                  BY CONSTRUCTION YEAR
FDe
Fidl National Frame (1)
Plumber of units
Percent
30 County Extract, Unedited (1)
Number of units
Percent
Projected to Nation (2)
Number of units
Percent
30 County Extract, updated (3)
Number of units (4)
Percent
Projected to Nation (2)
Number of units (5)
Final Sample
Number of units
Percent
Projected to Nation (6)
Number of units
Tfn,,, n ,,«
Construction Tear Stratum
ore-1950
161401
20%
37,060
21%
158,534
20%
44,700
22%
165,233
21%
30
31%
182,000
20%
1950-1959
246,680
31%
56,580
31%
244,610
31%
77,189
38%
281,529
35%
24
25%
273,000
30%
1960-1979
388,475
49%
86,355
48%
394,496
49%
80,073
40%
353,427
44%
43
44%
455,000
50%
Total
796,656
100%
179,995
100%
797,640
100%
201,962
100%
800,189
100%
97
100%
910,000
100%
NOTES:
 (1) Source: HUD's inventory of public housing units data file.
 (2) National Projection obtained by use of tbePSU weights.
 (3) Source: Revisions to HUD's inventory made after contacting the respective PHA's.
 (4) Excludes 2^19 units in 8 projects found to be out of scope during field work
 (5) Excludes 62,555 units projected from the 8 projects in note 4.
 (6) National Projection obtained by use of the final sampling weights, which were adjusted to
     conform to HUD's counts of the number of family units in each construction year stratum.
                            3-11

-------
                                         TABLE 3-7

                 DISTRIBUTION OF PUBLIC HOUSING FAMILY UNITS
                                    BY CENSUS REGION

FOe
Full National Frame a)
Number of units
Percent
30 County Extract, Unedited (1)
Number of units
Percent
Projected to Nation (2)
Number of units
Percent
30 County Extract, updated (3)
Number of units (4)
Percent
Projected to Nation (1)
Number of units (5)
Percent
Final Sample
Number of miits
Percent
Projected to Nation (6)
Number of units
Percent
Census Region
Northeast

271,924
31%

99,926
5496

328,091
40%

391,240
43%

374,844
47%

43
44%

391,236
43%
Midwest

151,661
17%

36,747
20%

107,932
13%

102,960
11%

84,550
11%

11
11%

102,962
11%
South

361,280
41%

35,339
19%

283,800
34%

305,660
34%

260,566
32%

32
33%

305,661
34%
West

90,154
10%

14,198
8%

103,463
13%

110,140
12%

83,710
10%

11
11%

110,140
12%

Total

875,019
100%

186,210
100%

823,286
100%

910,000
100%

803,670
100%

97
100%

910,000
100%
NOTES:
 (1)  Source:  HUD's inventory of public housing units data file, including post-1980 buildings.
 (2)  National Projection obtained by use of the PSU weights.
 (3)  Source:  Revisions to HUD's inventory made alter contacting the respective
 (4)  Excludes 2,519 units in 8 projects found to be out of scope during field work
 (5)  Excludes 62,555 units projected from the 8 projects in note 4.
 (6)  National Projection obtained by use of the final sampling weights, which were adjusted to
      conform to HUD's counts of the number of family units in each construction year stratum.
                                          3-12

-------
national fiame.  Finally, when examined by PHA size (Table 3-8), the sample projections are relatively low
for the smallest size category (29% versus 40%) and high for the largest size category (52% versus 32%).

      Thus, although the sample appears fairly representative with regard to age, some differences appear
for both geography and PHA size.  A second look at Tables 3-7 and 3-8 helps explain where these
differences occurred. For both tables, the distribution appear fairly consistent among the final sample and
30 county extracts. It  is at the first stage of sampling,  when the 30 counties were selected from the full
national frame, that the major differences appear to have occurred.   The sample of 30  counties was
designed to be representative of privately-owned housing and not public housing. The survey design is
described in Appendix I, Chapter 3, of mis report entitled: Sample Design and Selection.

      Techniques such as post-stratification  can be used to adjust the sample weights so the distribution
by characteristics is more similar to the distribution of units on the full national frame.  Representativeness
in itself is desirable because it eliminates a possible source of bias.  But post-stratification for this sample
would also increase the variance of the estimates. Consider, for example, representation with regard to size
(Table 3-8). If we post-stratify by PHA size (Table 3-8), then the weights for housing units in huge PHAs
would become smaller  and the weights in  small PHAs would become larger.  This  would increase the
heterogeneity in the weights, since the large PHAs would now tend to have the small weights  and the small
PHAs the huge weights.  Increased heterogeneity in the weights results in increased variance, or reduced
precision.

      There was a second reason why post-stratification was undesirable. Post-stratification would have
required the use of the  HUD public housing data for the entire nation.  We encountered numerous errors
when updating the 30-county extract,  which suggested that the error rate is high in the entire national file.
Of course it was not possible to remove the errors in the entire national file.  Thus,  there  is no way to
assess the accuracy of the distributions of characteristics  derived from the full national file.

      Thus, while post-stratification would eliminate one source of bias, it would tend to increase the
variance of estimates due to the increased heterogeneity of the sample weights, and it would introduce a
second source of bias and variance: the errors in the full  national frame.  On balance, the disadvantages of
post-stratification were felt to outweigh the advantages,  especially since there was no  need to present the
survey results by PHA size or census region.

      The public housing survey design called for randomly selecting one housing unit from each sampled
project.  This design was difficult to  implement in the field.  As mentioned above, 44 of the 97 housing
units surveyed were apparently vacant  The number of vacant units was so large because  some public
housing project managers apparently steered the field team to vacant housing units.  This is a potential
source of bias since the public housing project managers may have steered the field teams to housing units
that were not representative of the units in the housing  projects with respect to the survey variables. A
systematic difference between vacant and occupied housing units would be an indication of such a bias.
Since only one unit was selected from  each project, no estimates of bias could be constructed.

      Table 3-9 gives the number of housing units sampled broken down by occupancy status and
prevalence of selected  lead-related characteristics.  To examine the possibility of "steering  bias," the
percentages of occupied and vacant  housing units that had  each lead characteristic were  recompared.
                                              3-13

-------
                                               TABLE 3-8

                      DISTRIBUTION OF PUBLIC HOUSING FAMILY UNITS
                                              BYPHASIZE

Fife
FoD National Frame (1)
Number of units
Percent
30 County Extract, Unedited (I)
Number of units
Percent
Projected to Nation (2)
Number of units
Percent
30 County Extract, updated (3)
Number of units (4)
Percent
Projected to Nation (1)
Number of units (5)
Percent
Final Sample
Number of units
Percent
Projected to Nation (6)
Number of units
Percent
PHA Size (No. of Family Units)
<1,250

368,777
40%

18,737
10%

252,774
31%

13,885
7%

224,182
28%

22
23%

263,165
29%
1 .250-2350

93,396
10%

6,192
3%

60,437
7%

7,882
4%

49,688
6%

7
7%

64,548
7%
2351-10,000

169,807
18%

15,927
9%

113,355
14%

13,634
7%

89,694
11%

14
14%

109,793
12%
> 10.000

301,593
32%

145,354
78%

396,720
48%

167,645
83%

442,970
55%

54
56%

472,506
52%

Total

933473
100%

186,210
100%

823,286
100%

203,046
100%

806,534
100%

97
100%

910,000
100%
NOTES:
 (1)  Source: HUD's inventory of public housing units data file, including post-1980 buildings and buildings outside the 48
      contiguous states.
 (2)  National Projection obtained by use of the PSU weights.
 (3)  Source: Revisions to HUD's inventory made after contacting the respective PHA's.
 (4)  Excludes 2,519 units in 8 projects found to be out of scope during field work
 (5)  Excludes 62,555 units projected from the 8 projects in note 4.
 (6)  National Projection obtained by use of the final sampling weights, which were adjusted to
      conform to HUD's counts of the number of family units in each construction year stratum.
 (7)  Sums may not equal total due to rounding error.
                                                   3-14

-------
                                       TABLE 3-9

PAINT LEAD, PAINT DAMAGE AND DUST LEAD IN PUBLIC HOUSING, BY OCCUPANCY
                           UNWEIGHTED SAMPLE COUNTS




Characteristic
Interior
Lead-based paint (LBP)
Damaged LBP(l)
Damaged Paint (1)
ffigjb. lead dust (2)
Exterior:
Lead-based paint
Damaged LBP (1)
D-amattoA VMirnt f1\
amageupamt (ij
Priority hazard present (3)
Public Housing
Occupied
53 Units
TQ Sfljnplc
Number Percent

42
1
10
1

35
4
10
6

79%
2%
19%
2%

66%
8%
19%
11%
Vacant
44 Units
in Sample
Number Percent

35
3
15
3

32
2
8
6

80%
7%
34%
7%

73%
5%
18%
14%
         (1) Paint is considered to he 'damaged' if more man five square feet of it is peeling, chipped, or
            otherwise damaged.

         (2) Dust is considered to be Trigh lead dust" if the dust lead level exceeds the clearance levels
            in the HUD Guidelines, 200ug/sq ft on floors, 500 ug/sq ft on window sffls, and 800 ug/sq fton
            window wells.

         (3) Priority hazard is present if the dwelling unit has LBP, either inside or outside and rimer high
            lead dust or more than five square feet of damaged LBP (total, inside and outside).
                                               3-15

-------
Using interior damaged paint as an example, the test statistic, z,  was calculated to be 0.71 using the
standard formula for comparing two independent proportions.13 The critical value for the significance test
is 1.96. Since 1.2 is less than 1.96, we cannot conclude that there is any difference between occupied and
vacant housing units with regard to the prevalence of interior damaged paint.  Similar calculations for the
other lead characteristics and no significant differences were found between vacant and occupied housing
units.  Thus, there is no statistical evidence of any differences between vacant and occupied housing units
with respect to the selected lead-related characteristics.


3.3    Correcting for Measurement Bias

       This section describes the methods used to correct measurements for calibration bias and censoring
bias (to be defined below).

       Definitions of Four Types of Measurements

       Each measurement passed through a series of transformations from the time it was first calculated
internally by the MAP/XRF instrument, to when it was fully corrected for bias.  To avoid confusion,
different terminology will be used to refer to the measurement as  it passed through different stages of
processing.  Four types of measurements are now defined:

        •      An internal measurement is a value the MAP/XRF  instrument calculated internally
               before it displayed a number.  The technician cannot observe this number.  Although the
               true lead concentration cannot be negative, the internal measurement on surfaces with little
               or no lead paint can be negative either because of measurement bias or variable error.

        •      A field measurement is the number displayed by the MAP/XRF instrument and recorded
               in the field.  The field measurement is different from the internal measurement when the
               latter is negative, in which case the field measurement is zero. This is called censoring.

        •      A recalibrated measurement is a field measurement after being corrected for calibration
               bias.

        •      A corrected measurement is a field measurement after being fully corrected for  both
               calibration and censoring bias.   The national estimates for the prevalence of lead-based
               paint are based on the corrected measurements.

       Measurement bias is a common phenomenon in a field study that uses equipment such as MAP/XRF
instruments. According to our field procedures, a single measurement was taken on each surface measured
within a dwelling unit  Compared to the true lead concentration, the measurement tended to be larger
(upward bias) in some situations, and smaller  in others (downward bias).   There were two types of
measurement biasc

        •      Calibration bias  occurred when the internal measurement tended to be systematically
               different from the true lead concentration being measured; and
13
  Z=
-------
       •       Censoring bias occurred when a negative internal measurement was displayed as zero.
               All zero field measurements were said to be censored because the internal measurement
               could not be observed.   The only thing  known  about  tie  internal measurement
               corresponding to a censored field measurement is mat the internal measurement was less
               than or equal to zero.

      The rernainder of this section  describes in detail the methods used to correct the field measurement
first far calibration and then far censoring bias.

3 J.I  Adjusting Field Measurements for Calibration Bias

      The possibility of calibration bias was anticipated in advance of the  field period.  In order to
estimate and subsequently correct for bias, a provision was  added to the survey procedures  to collect
"validation11 measurements.

      Validation Measurements

      Validation measurements consisted of field  measurements taken on surfaces (called shims) with
known lead concentrations.  The shims were prepared by painting 3 by 4 inch heavyweight paper sheets
with lead-based paint  According to NET, the  shims had lead concentrations of 0.6 and 2.99 mg/cm2.
Technicians placed shims over substrates made from four different types of material: wood, steel, drywalL,
and concrete.  One set of shims and one set of substrate samples were kept with each of eight MAP/XRF
instruments.  The survey field technicians collected validation measurements on the shims during three
different time periods:

       •       Baseline validation measurements were taken before the field period. After the MAP/XRF
               instruments were received from the manufacturer but before they were used in the field for
               foe survey, technicians  collected  eight replicate measurements  on  each of the  four
               substrates using each of the two shims, for a total of 64  measurements.  In addition, the
                         took measurements using the substrate material alone with no shim (roughly
              equivalent to a shim with 0.0 mg lead/sq. cm).

       •      Daily validation mgas"T«nients were taken during the actual field period.  Field technicians
              took one measurement of each shim on each substrate at the beginning and end of each
              day, for a total of 16 measurements per day.  The Daily validation measurements were
              used to develop the calibration equations.

       •      Closeoot validation measurements were taken after the field period. These measurements
              were taken in a feshitm similar to the Baseline validation measurements, except they were
              taken after the field period but before the instruments were returned to the manufacturers .

      The  validation measurements were studied and found  to be  consistent with the following
assumptions:

       •      The MAP/XRF uistrument calculates  an internal value, referred to here as the internal
              measurement.
                                              3-17

-------
        •      The MAP/XRF instrument display; the field measurement which is equal to the
               of zero and the internal measurement.

        •      The relationship between the internal measurement and the true lead concentration in the
               shims is approximately linear.

        •      The distribution of the variable error for a measurement, or measurement variance, for the
               internal measurements is approximately normal.

        •      The standard  deviation of the measurement variance does not depend on the true lead
               concentration in any substantial way.

      It should be noted that corrected measurements analyzed by NIST14 are also consistent with these
distributional assumptions.

      Outliers in the Validation Data

      During the preliminary exploratory data analysis, and later while processing the data^ a few mmgnal
measurements and, patterns were identified. After examining the original data sheets and any notes written
by the field technicians, 24 measurements fless than 1% of the data) were classified as outliers because they
were either very unusual, or taken at the same time as several other measurements that were also imngnal
With these outliers  excluded, the data are consistent with the assumption that the measurement variance
had a normal distribution. It was then valid to perform the statistical tests available for linear regression
procedures which assume normality.

      Estimating the Mean of the Internal Measurements

      For purposes of constructing the calibration equations, it was necessary to estimate the theoretical
mean of the internal measurements, say Ufj, for each combination of the ith substrate and jth shim.  With
uncensored and normally distributed data, regression is the preferred method of statistical analysis. But
censoring complicates the estimation of model equations because it results in upward bias for the regression
estimate of u« and downward bias for the regression estimate of the sample variance. An alternative
estimator, the sample median^ is an unbiased estimator of u.
-------
The RMSE for the sample mean varies with the percentage of zeros in the data. The bias for the sample
mean is also graphed and can be seen increasing with the percentage of zeros in the data.  The RMSE for
the sample median is a constant, and the bias is zero, as long as the percentage of zeros is 50% or less.

      With less than 7% censoring, both the sample mean and the sample median had little bias.  But the
sample mean had  lower RMSE and  was therefore the preferred estimate of ujj.  With 37% zeros, the
sample mean and sample median had the same RMSE, but the sample mean was quite biased. Thus, in this
case, the sample median was the preferred estimator.  When the percentage of zeros is between 7% and
37%, the choice between estimators was a matter of judgment

      Factors that Affected the Means of the Internal Measurements

      The daily validation measurements were used in a regression analysis to determine the factors fra*
affect the means of the  internal measurements, to develop the model, and to calibrate equations.  It was
desirable to use as much of the data as possible and, at the same time, to minimise the effect of censoring.
Thus all the data were used from instrument-shim-substrate combinations where the percentage of zeros
was less than 20%. If there were fewer than 20% zeros in the data set, the mean was used; otherwise the
median was used.  A cutoff of 20% was chosen because it is midway between 7% and  37%. Since there
were few data sets  where the percentage of zeros was near 20%, the results of the analyses are not sensitive
to the choice of the cutoff

      As will be described below, the magnitude of the measurement variance differed across instruments,
substrates, and, to  some extent, shuns. Accordingly, weighted regression was used to identify factors mat
had a significant effect  on the calibration equations.  Each weight was calculated as  the inverse of the
estimated measurement variance within each instrurnent-substrate-shim combination.  A preliminary
weighted regression analysis  showed that the model depends on a number of factors:

        •       Period. The same set of shims was kept with each MAP/XRF instrument throughout the
               field period. However, no attempt was made to use these same shims during the Baseline
               and Closeout validation periods.  It is possible that different shims (with slightly different
               lead concentrations) may have been used for the different validation periods for some or all
               of the MAP/XRF instrument  Thus  it was decided to use only the Daily validation
               measurements to derive the model and calibration equations.

        •       Instrument, Substrate, and Shim.   It is  not possible  to estimate how much of the
               apparent instrument^to-instrument differences are  due to shim  differences, substrate
               differences, or other differences associated with the instruments.  This is because only one
               set of shim* and substrates were used for each instrument within a given period.  The
               shims had lead concentrations of 0.6±0.02 and 2.9ft±0.30 mg/cntf, where the error ranges
               are standard deviations.^
15 The olculatiaos ate bued on a nnple of 45 ooraally dtoflwted metwianenls (the range iwnibawithm^
staMlirddeMJ^onof0^5iii^iqem,fxwn
-------
                                FIGURE 3-1

   RMSE AND BIAS OF THE SAMPLE MEAN AND MEDIAN WHEN ESTIMATING THE
                       MEAN OF INTERNAL MEASUREMENTS
    0.1  T
   0.09 -h
          o *
   0.08 -r-5 S.
   0.07 H-
uj  0.06 -|-
co
   0.05 •+•
  0.04 H-
  0.03 -1-
  0.02 +
  0.01 -f-
                                  2 "°
                                  £ v
                                  w t
                                  c5
                                  ^ e
                                  T3 5.
                                  0 &
       0%
                           •'V
                          A
                         X    i
                            +
10%
20%       30%       40%

 Percent zeros in the  data
50%
60%
          Bias in the Mean
                 RMSE for Mean
                              RMSE for Median
  Calculations assume 45 normally distributed values with standard deviation  of 0.25
                     and rounding to tenths of a unit
                                  3-20

-------
       •       Age of Photon Emission Source.  The rate of photon emissions is known to decay over
               time. The rates of decay were not significantly different across instruments or substrates.
               Therefore, a smgle formula was used for all measurements.

      The effect of operator was investigated, but the results were not statistically significant.  Using the
Daily validation measurements,  a separate model equation,  and subsequently a  separate calibration
equation, was developed for each, combination of MAP/XRF instrument and substrate.

      Factors that Affected the Variances of the Internal Measurements

      The spectrum analyzer MAP/XRF instrument measures the lead content in the wall by exposing a
small portion of the wall material to radiation. In response to this exposure, the lead on both the surface
and the substrate emit photons at specific frequencies. The MAP/XRF instrument measures the energy and
number of photons emitted by the lead and substrate.  The internal measurement is based on the relative
number of the photons at different energies.

      For the Daily validation measurements, mere were  significant  differences in the  measurement
variance associated with each combination of shim, substrate,  and MAP/XRF instrument (p<.0001).
Consistent differences in variance were also found among substrates and shims across  all instruments.
Factors that affect variance include:

       •       Instrument characteristics. Characteristics such as the shim samples, substrate samples,
               calibration, temperature,  battery power, and how the technician holds and operates the
               instrument, can all induce variance.   Differences in the strength of the instrument's
               radioactive source or the  sensitivity of the electronics can result in differing measurement
               variance among MAP/XRF instruments

       •       Photon Emission.  The  number of photons emitted and detected during  the sampling
               period has inherent variability. Because the field measurement depends on the number of
               photons detected per unit time, there is a natural measurement variance even with all other
               factors  held constant   Using  longer exposures  reduces  this  variance.   All the
               measurements collected in the nytinnal  survey were based on a 60 second exposure.  Based
               on physical  principles,  the number  of photons  detected  should  follow a Poisson
               distribution.

       •       Surface Characteristics.  Variance in the homogeneity of the substrate material can result
               in differences in the measurement variance among substrates.

       •       Lead  concentration.    Considering  the various actors mat can affect  the field
               measurement, the variability of the measurements is expected to either remain constant or
               increase as lead concentration increases.  This relationship was investigated  using the
               Daily validation measurements. The hypotheses could not be  rejected that the variability
               of the measurements remains constant across different lead concentrations (p=.62).

      Table 3-10 and Figure 3-2 show the pooled measurement standard deviation across all instruments
for each combination of shim and substrate.  Also shown are the measurement standard deviations
determined by NIST for comparable 60-second sampling periods.  The measurement variance for me data
collected in the national survey is similar to that found by NIST.  Any differences in the measurement
                                             3-21

-------
variance between the NIST data and the Daily validation measurements may reflect instruments, instrument
programming, or the conditions under which the measurements were taken.

                                       TABLE 3-10

       STANDARD DEVIATION OF REPLICATE FIELD MEASUREMENTS BY
              SUBSTRATE, POOLED ACROSS MAP/XRF INSTRUMENTS
Data Source
Daily
Validation
Data
NIST
Substrate
Wood
Steel
Drywall
Concrete
Wood & Plaster
Std. Deviation of Replicate Measurements
0.6 shim 2.99 shim Pooled
0.17
0.16
0.24


0.31
0.25
0.26
0.44

0.25
021
0.25
0.49
0.3a
       a Applies to lead concentrations below 2.0 mg/cm2.

       Testing the Linearity of Measurements with Lead Concentration

The existence of a linear relationship between the validation measurements and shim concentration was
tested using available data for three lead concentrations: 0.0,0.6, and 2.99 mg/cm?.  Because there were
significant differences in the model between the Baseline, Daily, and Closeout periods, the results of this
test were only approximate. Furthermore, sets of measurements without shims and with less man 20%
zeros were available only for steel substrates with three instruments in the baseline period and four
instruments in the closeout period. A quadratic curve was fit to these data (see Figures D-l through D-7 in
Appendix D). Ahhrmgh gfimft quadratic tarmc \nrn-. statistically gignifirant tfa» dirarrinn nif the curvature
was not consistent across instruments, or between the baseline and closeout period for one instrument
Because of the limited data and inconsistent results, the assumption was made that the relationship between
the field measurement and the shim concentration was linear.

       A Formal Test of Factors

      The effects of factors on the means  of internal measurements were tested for  significance using
weighted regression on daily validation measurements for substrate-instrument-shim combinations  where
less than 20% of the measurement were censored. Factorial interaction terms were included for instrument,
substrate, and shim concentrations. These  terms were highly significant (p<.0001).   These differences
might have been  due to differences among  the substrate samples. Because some show quite different
patterns on some substrates,  it was assumed mat there were significant differences between instruments
beyond any variability due to  the shims and substrates. Accordingly, a different model equation was fit for
each combination of instrument and substrate.
                                            3-22

-------
                        FIGURE 3-2

   POOLED STANDARD DEVIATION OF 60 SECOND VALIDATION
              MEASUREMENTS BY SUBSTRATE
                                                   Wood
0.5       1       1.5      2      2.5

   Lead  concentration (mg/sq cm)
                                                    MIST measurements on
                                                    wood and plaster
                          3-23

-------
      The model also included a term for time.  The rate of photon emissions is known to decay over
time. With all instrument and substrates combined, the decay was statistically significant (p=.017), as
expected.  Furthermore, the rates of decay were not found to be significantly different across instruments or
substrates. Therefore,  a single formula  was used to adjust all measurements for this decay.  This was
accomplished by including a single trend term for all instruments that represented time.  The estimated
change in the expected field measurement for each additional day in the field was -0.00099 mg/cm2.

      The Calibration Model

      Based on the preliminary analyses described above, the calibration equation was assumed to have
the following form:

        Y = afj +  pt,  *X- 0.00099  * /,


where Y = the field measurement, X = the lead concentration, t = time (Le. the number of days since
2/1/90), a = the intercept, p = the slope, and the subscripts i and j represent instruments and substrates,
respectively.
      Because the Daily validation measurements haH measurements at only two shim concentrations, the
model equations were simple to calculate.  Since the regression line passes through the estimated means for
the two shims (after adjusting the measurement for the time trend), it was a simple matter to calculate the
estimates, a and b, of the model parameters a and p.  The calibration equation was then determined by
inverting the estimated model equation as follows:

            (7 -a, -+0.00099*r)
                   >
where x is an estimate of the lead concentration X. Since me lead concentration cannot be less than zero,
the Recalibrated measurement were calculated as

        Recalibrated measurement = Maximum (x, 0).

      Figure 3-3 illustrates the calibration equation for instrument #32 on steel.  (Figures D-8 through D-
38 in the Appendix D show the calibration equations for other combinations of instrument and substrate.)

      •     The solid line is the calibration equation.

      •     The dotted line is shown for reference purposes. This line has an intercept of 0 and a slope
            of 1 and shows the hypothetical case where no adjustment is necessary.  The more the  slopes
            of the solid and dashed lines differ, the more important it is to adjust for calibration.

      •     The open  circles represent the actual measurements.  They are plotted  as a  sideways
            histogram, rather man on top of each other, in order to reveal more detail in the plot.

      •     The filled circles represent outliers that were removed from the analysis.
                                              3-24

-------
                                    FIGURE 3-3

           VALIDATION MEASUREMENTS BY SHIM LEAD CONCENTRATION
                        FOR XRF INSTRUMENT #2 ON STEEL

Lead  concentration  (mg/sq cm) = 0.5793 + 0.6161*Measurement + 0.00061 "(Days since 2/1/90)

   5 T
                       X = Lead Concentration (mg/sq em)
                                   Measurement

                                   Outlier (not used)

                                — Calibration Equation

                                — X = 1, projected to Y
                       When X = 1, Y = approximately 0.65
                       Median  used for 0.6 shim
                                        3-25

-------
      •     The dashed line is also given for reference purposes.  This line illustrates the calibration
            adjustment for the special case where the true lead concentration is  1.0 mg/cm2.  This
            corresponds to a field measurement of about 0.68.

      With censored validation data, there is a tendency for the Recalibrated measurement to have upward
bias when the actual lead concentration is greater then 2.99 mg/cntf. This problem is particularly severe
for the calibration equation for instrument #34 on concrete.  Thus, for this particular instrument and
substrate combination, a different calibration procedure was used: the slope of the model equation was
estimated as the average  of the estimated  slopes for  the  other seven MAP/XRF instruments  when
measuring on concrete.

      Precision of the Recalibrated Measurements

      Direct calculation of the precision of the recalibrated measurements is quite difficult because of the
relatively complex procedure for defining the calibration equation.  Therefore, simulations were used to
estimate the precision of the recalibrated measurements.  The simulations incorporate the variance in the
nominal lead concentration of the shims, the internal measurements used to derive the calibration equation,
and the field measurements which are to be recalibrated.  Simulations were performed to answer the
question: what are the mean, standard deviation, and  percentiles of 1000 independent  recalibrated
measurements, each made on a surface with the same known true lead concentration.

      One thousand simulations were performed for each of the following lead concentrations:  0.0, 0.4,
0.75,1.0,1.25,2.0,3.0, and 4.0 mg/cmZ. Five steps were used for each simulation:

      1)    simulate the true lead concentration in the two shims;

      2)    simulate 45 Daily validation measurements for each shim;

      3)    calculate the calibration equation;

      4)    simulate the field measurement corresponding to the true lead concentration in the measured
            surface; and

      5)    calculate the  Recalibrated measurement for the field measurement from step  4 using the
            calibration equation from step 3.

      The simulations required  making  assumptions about the  true relationship  between  the  lead
concentration and the  field measurements, which is the true model equation.   A different true  model
equation was assumed for  each substrate. As with the model equations, the true model equations were
assumed to be linear.  The true model equations were  determined by pooling the measurements  for each
substrate across all MAP/XRF instruments and using the procedure defined above to derive the model
equations.  The measurement variance for measurements on each substrate are  based on the variances
shown in Table 3-10.

       Figures D-39 through D-43 (see Appendix D) are sets of boxplots that  show percentiles of the
simulated recalibrated measurements. Each box covers the central 50% of the values. For some substrates
and low concentrations, the lower percentiles of the adjusted lead measurements all fall at 0.6, causing the
box to collapse halfway or entirely. The bar through the center of the box is plotted at the sample median.
The simulated recalibrated measurements  are  clustered close  to the dashed  line (where the  actual
                                              3-26

-------
measurement = simulated measurement).  This is convincing evidence of the effectiveness and validity of
the recalibration procedures.

       Table 3-11 presents the mean and standard deviations of the Recalibrated measurements calculated
using the simulations.  For concrete, unlike other substrates, the assumed true relationship between the field
measurements and the lead concentrations is likely to be incorrect  This is because almost all the data for
the 0.6 shim is censored (see the column under assumption 1 for concrete in Table 3-11). Therefore, the
alternate assumption (assumption 2) also was made that the true model equation has a slope of 1.0 and
passes through the median value for the 2.99 shim.  The difference in the statistics in Table 3-11 for these
two sets of assumptions demonstrates the uncertainty in the Recalibrated measurements for concrete, due to
the censoring of the data.

                                       TABLE 3-11

       MEAN AND STANDARD DEVIATION OF SIMULATED RECALIBRATED
                                    MEASUREMENTS
True
Lead
Cone.
(rug/am)
0
.4
.75
1
125
2
3
4
Substrate

Wood
Mean
0.60
0.61
0.78
1.01
1.28
2.04
3.07
4.11
SD
0.00
0.04
0.16
0.20
022
0.27
0.39
0.51

Steel
Mean
0.12
0.41
0.74
1.02
1.26
2.02
3.01
4.02
SD
0.16
0.27
0.29
0.30
0.31
0.34
0.41
0.53

Drywall
Mean
0.60
0.62
0.77
1.00
127
2.02
3.07
4.08
SD
0.00
0.05
0.17
020
0.24
0.28
039
0.51
Concrete
(Assum
Mean
0.70
0.81
1.02
1.16
1.33
2.06
3.09
4.16
ptionl)
SD
0.26
0.40
0.54
0.61
0.68
0.82
0.87
0.95
Concrete
(Assum
Mean
0.60
0.60
0.61
0.64
0.69
1.46
3.13
4.92
ption2)
SD
0.00
0.00
0.09
0.17
0.26
0.77
1.01
1.32
Note: SD = Standard Deviation

332  Adjusting Recalibrated Measurements for Censoring Bias

      Field measurements of zero are said to be censored because the corresponding original internal
measurement could not be observed. In order to simplify die discussion mat follows, distinctions are made
between two types of Recalibrated measurements:

      •     Censored Recalibrated measurements are  recalibrated measurements that correspond to
            field measurements that were zero.  Most censored recalibrated measurements had values
            near 0.6 mg/ cm2.

      •     Uncensored Recalibrated measurements are recalibrated measurements that corresponded to
            field measurements that were not zero.

      Figiire3-4showsalrctogiamoftherecauT>ratedmeas^                         Thelabelson
the x-axis show the midpoint of every fourth interval. The gray bars show frequencies of the uncensored
                                            3-27

-------
recalibrated measurements, and the stacked black bars show frequencies for the censored recalibrated
measurements.  The distribution of the uncensored recalibrated measurements above .63 is fairly regular in
appearance.  The irregular appearance of the lower portion of the histogram is due to the stacking of
censored recalibrated measurements at specific values.  Recall that a different calibration equation was
developed for each combination of instrument and substrate. For each such combination, all of the zero
(censored) field measurements were mapped into approximately the same value when recalibrated.  Thus
censored recalibrated measurements  appear  stacked  at  a  limited number  of values, each  value
corresponding to a combination of instrument and substrate.

      Most censored recalibrated measurements corresponded to negative internal measurements, although
certainly there were some mat corresponded to zero internal measurements. Since the censoring process
itself tended to increase the measurement, and never decreased  it, it is easy to  see how censoring induced
upward bias in the field measurements.  This upward bias carried over to the recalibrated measurements.

      In the previous section, it was described how field measurements were corrected for calibration bias,
applying the same calibration formula to  zeros and non-zeros alike.   In  this section, the  censored
recalibrated measurements were  singled out and statistically  corrected  for censoring bias.  This was
accomplished by replacing the censored recalibrated measurements with expected values of corresponding
uncensored  recalibrated  measurements.   The union of these expected  values with the uncensored
recalibrated measurements will be referred to as the corrected measurements.

      Figure 3-5 shows a histogram of the corrected measurements.  The black bars have been spread to
the left and the histogram has a fairly regular appearance throughout its  entire range. The gray bars are
unrViangftH because the original uncensored recalibrated measurements needed no correction.

      Determining Approximate Expected Values

      As noted above, the censored recalibrated measurements were statistically corrected for censoring
bias by replacing them with  approximate expected values  of corresponding uncensored recalibrated
measurements.   The distribution of the  uncensored recalibrated measurements have  an  approximate
lognonnal distribution. If the parameters of this distribution can be estimated,  then approximate expected
values of all the ordered observations can be approximated as percentiles of the estimated distribution. The
available data were censored.  There are some well-established approaches to estimating parameters with
censored data that include maximum likelihood estimation, probability plotting, and regression analysis.

      A plot of the  logarithms of the ordered uncensored recalibrated measurements  versus  expected
standard normal order statistics will tend to be a straight line.  Weighted regression  was used to estimate
the intercept and slope of the regression line, the  weights being the inverses of the variances of the
corresponding standard normal order  statistics.  The regressions were performed separately  for each
combination of instrument and  substrate.  Because our interest was in the distribution of the uncensored
corrected measurements  at low lead  concentrations, and the  lognonnal distribution may not fit these
measurements well at high lead concentrations, the regression was run using only uncensored recalibrated
measurements below 4.0 mg/cm2.  This cutoff value was used because the relationship appeared reasonably
linear in tfa*t range.

      With the distribution of the uncensored measurements fully  estimated, it was then possible to
calculate expected values for all uncensored ordered measurements in the full sample.  The  censored
recalibrated measurements were then replaced with these expected values which will be referred to as
corrected recalibrated measurements.
                                               3-28

-------
                                  FIGURE 3-4


                HISTOGRAM OF RECALIBRATED MEASUREMENTS IN
                     PUBLIC AND PRIVATE DWELLING UNITS
  800  -r
  700  -
  600  -
I 500
o

0»

I

I 400  -
  300  -
  200  -
   100  -
                                  D Censored

                                  ^ Uncensored
       O  CO  in
       «-  «-  CM
       O  O  O
        •   •   •
           O  O
q
6
en
(O
q
6
i-  oo
6  £
   6
i-  00  «-
If)  0>  CO
es  co  CD
666
                           10  CM
                           CO  T-
                                                   «-  «-  O
a>  o>  «-

CO  ^  5
 •   •   •
lO  IO  O)
«-  CM  CO
                             Lead Concentration (mg/sq cm)
                                    3-29

-------
                                  FIGURE 3-5



            HISTOGRAM OF CORRECTED MEASUREMENTS IN PUBLIC

                        AND PRIVATE DWELLING UNITS
  800  -n
  700
  600  -•
1 500
  400  --
B>
a
a>


•^
e

a
= 300
  200  -|
   100
    0  -•
         o

         o

         n
         ho
         CO
         01
                                                          Censored


                                                          Uncensored
       «-«-eNOeo.~:
       ooOQO°

          do     d
                           '-eMp)«D
                           dodo"
if)  en
                                                            c
             \n
             r-  CM
«-;  oq
U>O>
                            Lead Concentration (mg/sq cm)
                                   3-30

-------
      Assigning Expected Values to Dwelling Units

      A remaining task was to assign corrected measurements for the censored recalibrated measurements
to specific  dwelling units.   While the  corrected measurements  have a  natural order,  the censored
recalibrated measurements do not since, for a particular combination of instrument and substrate, all of the
censored recalibrated measurements had about the same value.

      One result of the  simulation study, to be discussed below, was that corrected measurements on
similar surfaces were more similar within dwelling units than between dwelling units.  Therefore, the
censored recalibrated measurements were ordered using the average estimated concentration of lead-based
paint within a dwelling unit on surfaces in the same room with similar architectural type.  la a few cases,
the corrected measurement was undefined because the substrate was unknown, the MAP/XRF instrument
serial number was missing, or there was not enough data to perform the regression analysis. In these cases,
the corrected measurements were set equal to the recalibrated measurement


3.4   Correcting for Bias in the National Estimate of Lead-Based Paint Prevalence

      One of the primary objectives of the survey was to estimate the national prevalence of dwelling units
with lead-based paint This was done by computing the weighted proportion of sampled dwelling units that
were classified as having lead-based paint Each dwelling unit was clarified as having lead-based paint on
the basis of the maximum corrected measurement for that dwelling unit  But this sample maximum is a
biased estimator of the maximum true lead concentration in the dwelling unit  This section describes the
causes of this bias and the procedures used to correct for it in  order to construct an unbiased estimate of
lead-based paint prevalence.

3.4.1 Three Types of Estimates at the Dwelling Unit Level

      National estimates based on the survey data are all weighted (inversely to the probability of
selection) meang of estimates made at the dwelling unit level.  It is important to distinguish among three
different types of estimates made at the dwelling unit level:

      •      Averages of measurements.   Examples are  the  average  (i.e.,  weighted  mean)  lead
             concentration in paint and die average surface area of lead-based paint in a dwelling unit

      •      Maxhnums of measurements.   One  crucial example is the national estimate  of the
             prevalence of lead-based paint  The maximum  of multiple measurements is used to define a
             lead-based paint indicator for a dwelling unit  The reason for using the maximum steins from
            the following definition- a dwelling unit is said to have lead-based paint if any one surface in
            the dwelling unit has an average lead concentration greater man or equal to  1.0 mg/cmZ.

      •      Classification Estimators. The lead-based paint  indicator defined above:fbr a dwelling unit
             is a classification estimator since it classifies each sampled dwelling unit as having (or not
             having) lead-based paint.

      The bias of an average of measurements is equal to the average of the biases of the measurements.
Section  3.3 describes  how calibration bias and  censoring  bias  were  removed from the individual
measurements.  Since the corrected measurements are unbiased (or have a very small bias), estimates of
average  lead concentration and surface area in a dwelling unit are unbiased.   However,  both
maxhnums  of  measurements and classification estimators  can be  biased,  even if the underlying
measurements are unbiased..
                                              3-31

-------
      Bias of the Sample Maximum

      The bias of a sample maximum (for estimating the true maximum) depends on two additional
factors, even if the corrected measurements are unbiased:

      1)    Upward bias due to multiple miss-classification opportunities.  This kind of bias results
            from having measurement variance in conjunction with multiple measurement opportunities.
            To see how mis bias arises, assume that a painted surface has a true lead concentration of 1.0
            mg/cm2- A single unbiased measurement with a symmetric measurement error distribution is
            equally likely to be above or below the true lead concentration, 1.0 mg/cm2- However, the
            maximum of two independent measurements from this distribution has a 75 percent change of
            exceeding the true lead concentration ( 1 - .S2 = .75).  The maximum is therefore a biased
            estimator.  With 12 independent measurement^ the probability •that the maximum exceeds the
            true lead concentration rises to 1 - .512 or 99.98 percent.

      2)    Downward bias due to  incomplete sampling of rooms and surfaces.  By design, neither
            every room in a dwelling unit nor every  surface in a room was  sampled.  This results in
            downward bias.  To fix ideas, consider the extreme case when  there is no measurement
            variance. Suppose a dwelling unit has 50 surfaces and the highest concentration of lead is on
            5  of these surfaces (which could happen if they were all painted at the same time). If only
            one surface is randomly sampled, the probability of selecting one of these five surfaces is
            only 10%. If 12 surfaces are sampled, the probability rises to 69%. The probability reaches
            100% only if 46 surfaces are sampled (since at least one must have the highest).17

      Bias Due to Classification

      Classification bias can occur when measurements are used to classify a unit For example, a painted
surface  is classified as having lead-based paint if the measured lead content  of the paint  exceeds  1.0
mg/cm2. Estimates of percentages derived from these classifications can be biased, even if the underlying
measurements are unbiased.  Classification bias depends on the density function of true lead concentrations
in the vicinity of 1.0 mg/cm.2.  If the density is increasing, then there tend to be more lead concentrations
just above the threshold than just below.  Because of measurement error, mere is a tendency to make more
Type n errors man Type I errors.  This results in downward bias in our classification estimator.  The
reverse is true if the density is decreasing.
      The Corrected Maximum Measurement

      Although the three sources of bias discussed above will offset each other somewhat, it is virtually
impossible that they will cancel each other out at the dwelling unit level  Furthermore, mere was the
possibility that the net bias in the national estimates could be substantial.  Therefore it was important to
develop a method to adjust the maximum corrected measurement for bias.  Accordingly, the following
definition is made:

      •      Corrected Maximum measurement is the maximum corrected measurement at the dwelling
             unit level  after it has been further corrected for upward bias due  to multiple  miss-
             classification  opportunities, downward bias due to  incomplete  sampling of  rooms and
             surfaces, and classification bias when estimating lead-based paint prevalence.
17 The test models the data with the hyper geometric distribution and uses Fisher's Exact Test Set Introduction to Statistical Analysis, by WiiSnd
Dixoo and Frank Mascey, Jr. (1969), 3rd edition, page 243.
                                              3-32

-------
      Purposive Samples

      In an attempt to compensate for the downward bias due to incomplete sampling of rooms and
surfaces, Ihe sample design specified that purposive samples be taken.  The first purposive sample was
taken on the surface inside the dwelling unit which the technician believed to have the highest lead content.
If that field measurement was zero, a second purposive measurement was taken inside the dwelling unit
One or two purposive samples also were taken on the exterior of the building, Mowing the same sequential
procedure.  If the technicians were skilled at finding paint with the highest lead concentration, purposive
measurements could have offset the downward bias due to incomplete sampling of rooms and surfaces. On
the other hand, if technicians were unable to distinguish surfaces with higher lead content, the purposive
measurements would be not much different than adding one or two more randomly selected surfaces.

      In order to assess the effectiveness of the purposive measurements, purposive measurements were
compared to non-purposive measurements in the same dwelling units, in rooms with the same wet-dry
status, and from components with the same architectural type. The average difference was small and not
statistically significant  This indicated mat, within the precision which can be obtained with the survey
data, once an architectural component and room type was selected by the technician, his ability to find a
surface with the highest lead concentration was not much better man chance.  (This analysis has  not been
performed.)

      Regardless of the skill of the technician in finding lead-based paint, including the additional samples
at the dwelling unit level reduces downward bias due to incomplete sampling of rooms and surfaces. But
         samples also increases upward bias due to multiple miss-classification opportunities.  The net
effect of purposive samples on the bias of prevalence estimates cannot be determined without further
analysis. (This analysis has not been performed.)

3.4.2 A Simulation Approach

      In order to correct for the bias in the national estimates  of lead-based paint prevalence, it is
necessary to estimate the biases  caused  by incomplete  sampling of rooms  and surfaces and  by
classifications based on measurements.  This, in turn, requires knowledge of the distribution of lead
loadings on all painted surfaces in a dwelling mitt, including the means, variances, and correlations between
different rooms and surfaces. Unfortunately, no applicable and adequate data set exists.  While there are
data sets with the needed data, e.g., the data from HUD's abatement demonstration project, none of these
data are from general population samples of housing units that used the same methodology as the national
survey to measure lead in paint  The only available alternative was to develop a simulation model and use
it to correct for the bias in the national estimates  of lead-based paint prevalence.  The simulation also
allowed assessment the effectiveness of the purposive samples.  The model was developed using the
following three steps:

      (1)    Characteristics of non-sampled rooms and surf aces were simulated. Room characteristics
             were known for sampled  rooms, but  not for non-sampled rooms.  A hot deck imputation
             procedure was used to simulate characteristics in these non-sampled rooms. This completed
             the information on the population of surfaces and rooms  from which would be sampled in the
             simulation process.

      (2)    A simulation model was developed that contained terms for all the components  that were
             believed to determine measurement location and variance.

      (3)    The model was calibrated to be consistent with aggregate characteristics  of the national
             survey data.
                                              3-33

-------
      Hot Deck Simulation of Surfaces and Rooms

      Imputation is a procedure by which missing data is replaced with data believed to be similar. The
analytical process is usually simplified when there are no missing values.  In this study, imputation made
the simulations possible.  There are many approaches to imputation.  The simplest is to replace each
missing value with the mean of comparable known values.  One disadvantage of this approach is that it
creates too much regularity in the data and variance estimates are often biased downward.

      The imputation procedure chosen was  the hot deck method.  Hot  deck imputation was chosen
because it preserves both means and variances. With mis method, missing data were replaced with actual
data from the survey that came from a similar  room in a similar dwelling unit. When choosing a similar
room, the degree of similarity depended on: 1) the total number of rooms in the dwelling nnit^ 2) public or
private status, 3) age category, 4) wet or dry status, 5) number of surfaces  in the room, and 6) the most
common substrates within architectural components. When multiple rooms were equally similar, one was
chosen at random.  A  similar imputation was performed for common rooms  where type of common room
(laundry room, office,  etc.) was also a consideration.  For purposes of the simulations, data for purposive
samples were retained and treated like other non-purposive samples.

      Correlation Between Rooms

      The question arose as to how to model the correlation of measurements taken in different rooms of
the same dwelling unit Such a correlation can  result from using the same paint in multiple rooms over the
painting history of the dwelling unit. Analysis of differences between the purposive and non-purposive
samples for similar architectural components suggested that the magnitude of the variance component for
differences between rooms was near zero. The  same conclusion was reached when the differences between
wet and  dry  rooms was assumed to be random rather man fixed.  Data from the HUD  Demonstration
Project did not help answer the question because the measurement error  was too large to permit the
estimation between-room correlation.

      While the paint on some surfaces in different rooms was the same, the choice of paint for the other
surfaces was  often made independently.  The correlation of measurements taken in different rooms depends
on the proportion of surfaces that  were painted independently.  Therefore,  the model included an extra
parameter18 that represented tins proportion. Because no data were available to support an estimate of this
parameter, the parameter was set equal to 50% for the final simulations.  However, the parameter was
varied from 0% (paint  chosen independently  ) to 100% (same paint used) in order to determine the
sensitivity of the estimate of lead-based paint prevalence to the changes in the parameter. Results of these
simulations indicated that the estimate was not very sensitive to the value of this parameter, except when it
was close to zero.  These simulations also provide a rough estimate of the component of variance in the
estimate of lead-based paint due to the uncertainty in the value of the parameter.
l^Thf rT"n*t'"' ™" tncncpnrated into the tjirailrtinpc hy jwi-fr"'n"g • B^^'flj ftH frf f"* mrftrf  Due to tfae MtUfC of bow it W»S illuaiKTltfti.
it does not appear explicity in the simulation model below.
                                             3-34

-------
      The Simulation Model

      Based en an accumulation of empirical evidence, the distribution of corrected measurements for
lead-based paint is skewed to the right and can be fit reasonably well with a lognonnal distribution. The
simulation model had the following exponential form:
                                                          Ip+e)
        Corrected Measurement = e                                  + v

where:

        Uijki =      mean of the log transformed lead concentration on surfaces of the same type.

        Xim=      random dwelling unit component,  normally distributed with variance a2- which
                   depends on the age category i.

        SijHmh =    random surface component, normally distributed with variance a2^.

        Ip =       adjustment to account for the purposive samples which have a slightly different mean
                   than the random samples.

        e =         random within surface component, normally distributed19 with variance 02.

        v =         measurement variance associated with the MAP/XRF instrument^ normally distributed
                   with standard deviation20 equal to (ag+bg*lead concentration).

        h —         index for an individual surface within a dwelling unit.

        i =         index for the vintage of the dwelling unit (Before 1920, 1920-1939, 1940-1949, 1950-
                   1959, 1960-1969, 1970-1979).

       j=         index for the area type (interior, exterior, common rooms, playground).

        k=         index for the room type (wet, dry).

        1 =         index for the architectural component (walls-ceiling-floor, metal substrates, nomnetal
                   substrates, other).

        m =        index for dwelling unit within age category.
19TheestimiteofthBptnuiieterv«obUii^fiom«iiiil^                          lawsuits ofUris analysis were sensitive to
theh«xffii«ofiinnien«soutlieBMd«1hu.iioti««e. fc the modd.tbesUndiid deviation of the *^»rfi«
judgedthittiiB(Btim«tewMwilhm»fictoroftwoofthe«ct>i«lestinMle. Tteefcce, amnUJions in which the witbmsuifece wuncc was dunged by
 ' lie itmlrt deviations show m TiUe 3-10 were wed to develop * la^
                                               3-35

-------
       Most of the terms in the simulation model were selected after exploratory analysis of the survey
 data. Through this analysis, factors were identified that had a significant affect on the mean and variance
 of the  corrected  measurements,  particularly in the  vicinity of 1.0  mg/cm2.   The SAS procedure
 VARCOMP  was  used to  estimate the proportion  of the between dwelling unit variance which was
 associated with differences between surfaces.  The estimates of variance components were obtained by
 subtraction.

       As a check on the accuracy of the model, the actual survey data were compared to the simulated
 data and significant differences were found for certain characteristics.   Calibration procedures were then
 performed that involved perturbing parameters until there were no longer any significant measurable
 differences.  The main effect of this effort was to increase the between-surface variance relative to tie
 between-dwelling-unit variance.

       The simulations were repeated ten times for each dwelling nnit  The simulation process itself
 contributed random error to the estimates. Simulation error can be reduced by increasing the number of
 simulations.   But mis requires additional  computer  time and  resources.   After 10 simulations, the
 simulation variance was approximated and it was determined mat the benefits of additional simulations
 were negligible.

 3.4.3  Adjusting the Maximum Corrected  Measurement to Remove Bias in the Estimate of Lead-
       Based Paint Prevalence

       The last step in the analysis was to  determine the corrected maximum measurement.   A linear
 adjustment was made so that the lead-based paint indicator based on the corrected maximum measurement
 was essentially unbiased for estimating the prevalence of lead-based paint

       A Linear Adjustment to the Maximum Corrected Measurement

       As defined above, the corrected maximum measurement is the maximum corrected measurement at
 the dwelling unit level after it has been further corrected for upward bias due to multiple miss-classification
 opportunities, downward bias due to incomplete sampling of rooms and surfaces, and classification bias.
 The corrected maximum measurement allowed the prediction of which particular dwelling units had lead-
 based paint, and to cross^tabulate these predictions with other characteristics of dwelling units that were
 not incorporated into the simulations, such as the number of children in the dwelling unit.

       The calibration of the model assured that there were no gignifi«mt measurable differences between
 the actual survey and the grmniat^ data at the aggregate level. Thus survey data characteristics drove the
 simulation model only through the fbrmulation of the model and the determination of parameter values.
 When gmnilating measurements for a dwelling unit, no additional use was made of the actual survey
 measurements for that dwelling unit In this sense, the simulated data were not paired to the actual survey
 data

       The maximum corrected measurement was computed  at the aggregate level by transforming its
 percentiles to equal the percentiles of the maximum true lead concentrations that were developed from the
 simulations.  A linear adjustment was deemed appropriate because the relationship between the percentiles
• of the distributions of the simulated and actual maximums was approximately linear in the range of 1.0 to
 2.0mg/cm2.  A different linear formula was determined for each vintage on dwelling unit

       After making this final correction, it  was noted mat the corrected maximum measurement was
 greater than the maximum corrected measurement   Recall that corrected maximum measurement is the
                                              3-36

-------
maximum corrected measurement at the dwelling unit level after it has been further corrected for upward
bias due to multiple miss-classification opportunities; and downward bias due to incomplete sampling of
rooms and surfaces.  Upward bias is corrected by decreasing the  maximum corrected measurement.
Downward bias is corrected by increasing the maximum corrected measurement Since the net effect of the
adjustment was an increase, this implies that the downward bias due to incomplete sampling of rooms and
surfaces was greater than the upward bias due to multiple miss-classification opportunities.

      For a few dwelling units, no corrected measurements were recorded.  In the past these dwelling units
have been treated as if they have no lead-based paint  The procedures used to  calculate the  predicted
maximum average lead concentration make the same assumption.  As a result, the  adjusted estimates may
slightly underestimate the proportion of dwelling units nationally which have lead-based paint

        Variance of the Unbiased Estimate of Lead-Based Faint Prevalence

        The prevalence of lead-based paint is the proportion of dwelling units with lead-based paint The
variance of a sample proportion depends on the proportion itself, p. The arcsine transformation was used
to stabilize the variance and make it independent of p.  The variance  of the transformed proportion is
approximately  l/4n, where n is the sample size.  This approximation was then multiplied by the design
effect of 1.45. Finally, a constant was added to take into account the uncertainty of the model, simulation
variance,  and variance due to imprecision  in  our parameter estimates, particularly the within surface
variance component and the effective proportion of surfaces with independent lead concentrations.  This
constant was  estimated to be 0.00224 after computing the  variance of results from multiple sets of
simulations  This procedure yielded the following estimate of the variance:

              1 45
        a\  = — + 0.00224


        In terms of our lead-based paint prevalence estimate, 95% confidence limits are given by:
        sin * (p ± 1.96 *      + 0.00224 ).
                           4n
                                               3-37

-------
       Table 3-12 presents confidence intervals for selected values of p and n.

                                       TABLE 3-12

  NINETY-FIVE PERCENT CONFIDENCE INTERVALS FOR LEAD-BASED PAINT
          PREVALENCE BY PREVALENCE ESTIMATE AND SAMPLE SIZE
Prevalence
Estimate
2%
5%
10%
20%
30%
40%
50%
60%
70%
80%
90%
95%
98%
Sample Size
284
0%to7%
l%toll%
4% to 18%
12% to 30%
20% to 41%
29% to 52%
38% to 62%
48% to 71%
59% to 80%
70% to 88%
82% to 96%
89% to 99%
93% to 100%
100
0%to8%
I%tol3%
3% to 21%
10% to 33%
17% to 44%
26% to 55%
35% to 65%
45% to 74%
56% to 83%
67% to 90%
79% to 97%
87% to 99%
92% to 100%
50
0%toll%
0%tol6%
2% to 24%
7% to 37%
14% to 49%
22% to 59%
31% to 69%
41% to 78%
51% to 86%
63% to 93%
76% to 98%
84% to 100%
89% to 100%
3.5   Laboratory Measurement Error and the Effects of Small Dust Sample Weights on the findings

      The objective of this section is to estimate the extent of laboratory measurement error associated
with a number of sources, including dust sample weight, location of the sample, and analytical protocol
The data files constructed for this analysis consist of entries for samples analyzed by laboratories.  The
entries contain identification codes, site codes, laboratory codes, weight of the sample, kind of sample (dust
or soil), location within the dwelling unit, micrograms of lead detected, weight of dust sampled, area
vacuumed, and lead loading (micrograms of lead per square foot for the dust samples only). The analysis
began with a review of the data file to determine if any data values should be trimmed from the file before
proceeding.

3.5.1  Trimming the Dust Analysis file

      Examination of the data showed that there were some unusually large values for bom weight of the
dust samples collected and for micrograms of lead measured in the dust Some of the large weights are
samples in which the dust was collected by a wet wipe (see discussion below). Some of the large readings
may represent errors in  data entry  or recording,  and some may represent actual dust samples with
atypically heavy concentrations of lead. Also, when the laboratory results (micrograms of lead detected
and weight of the sample analyzed) were converted to parts per million (ppm or micrograms of lead/grams
of sample), a number of very large readings appeared.  Even though some of the large readings could have
                                           3-38

-------
 been correct, it was judged that the review of dust samples having extremely large ppm of lead would
 contribute little to a judgment concerning the accuracy of the chemical analyses. Therefore, it was decided
 to leave all dust samples with lead values greater than 100,000 ppm out of the analyses and to consider
 other possible dust lead cutoffs based on lead loadings and sample weights as described below.

       A few dust samples had very high lead loading readings  (micrograms of lead per square foot of
 surface). This is an important statistic since it is probably the best measure of lead exposure to children
 under seven years old.21    High values however, have little relevance in terms of the ability of the
 laboratories to detect levels  of lead near threshold exposure levels (very low loading levels).  For this
 reason, samples with lead loadings exceeding 2,000 micrograms per square foot vacuumed were dropped
 from the QC analysis of laboratory data.

       Other  dust  samples eliminated from  analysis were unauthorized wipe  samples not collected
 according to the dust sampling protocol developed during the design phase of the survey.  The protocol
 specified dust samples collected in homes with a vacuuming technique. A few however, were collected in
 homes with wet wipes  when vacuuming was impossible (e.g., because no electricity was accessible) in
 order to avoid not collecting any dust samples at all.  It was necessary however, to eliminate the wipe
 samples from analysis because their treatment during laboratory analysis resulted in substantial errors in
 the "total dust weight collected" determination. The cause of the error was that the weight of the wet wipes
 and the weight of the dust were reported as one number.   Thus,  dust lead concentration  (ppm)
 determinations were impossible to calculate since the denominator (grains of dust + grams of baby wipe)
 was artificially large.  This resulted in very small (usually fractional) ppm values.  For this reason, and
 because of possible background contamination from the untested baby wipe commercial brands, the wet
 wipe s?'ninfftg were ehTninated_sdi6nevgr they could be identified.  Wet wipe samples also were eliminated
 from the dust analysis reported elsewhere in this report  The laboratory report showed that there were 81
 such samples (out of 2,178 dust samples-see page 17 of the MRI Report). Of the 81 samples, only 36 are
 identified in the detailed laboratory data and another 17 were identified in internal analyses conducted for
 mis report.  One of the latter could not be located in the file.

       After elimination of the identified wet wipe samples, samples greater than 100,000 ppm and lead
 loadings greater th«" 2,000 micrograms per square foot (see above), the remaining samples were classified
 by location within the homes, floors, window suls, and window wells.  The average weights and 95th and
 98th percentiles of weights were examined, as well as other possible cutoffs near the top of the scale.
 Based on the analysis,  and because many of the wet wipe samples could not be positively identified, a
Decision was made to exclude all dust {samples that wejefrfd. grea'ff1" ^^ 2Q ff3™5 collected from floors. 5
 gams collected from windgwjflfcjmdJLgCTrc*ftotn window wells.

       Table 3-10 shows the number of cases dropped from the analysis file for the various  reasons
 specified above. Many of the 111 samples with missing data appear to have had sample weights too small
 to measure or at least too small to analyze. The actual samples are listed in Appendix C. Atotal of 1,974
 dust samples remained in the file after the eliminations.
 21 Davies, DJJL; Thoratoo, L, Witt, J.K1; et al: "Rdati«Hhip between blood lead and intota in two year oU urban children in the UK,
 TotalEnv. 90:13 (1990).
                                               3-39

-------
                                         TABLE 3-13

      NUMBER OF DUST SAMPLES EXCLUDED FROM THE FILE BY REASON FOR
                                        EXCLUSION

Reason	Number of Samples	

Single reason

Missing data (for ppm or for lead loading)                           111
Dust collected by wet wipe                                         21
Excessively large ppm                                              5
Excessively large lead loading                                      22
Dust sample weight too large (may include                            16
   unknown wipes)

Two reasons

Dust sample weight too large and dust                                31
   collected by wet wipe
Dust sample weight too large and excessively                           1
   large lead loading
Dust sample weight too large and
   data (for ppm or for lead loading)

   Total	208	


3.5.2 Trimming the Sofl Analysis File

      There  was no  need to truncate the weights of the soil samples  since the analysis required
approximately one gram of soil and generally, much more man mis was collected in the field. There was a
problem however, with large soil readings; some were as large as 43,000 ppm. Although mere is no reason
to believe that the large  readings are  not factual, such readings contribute little to an analysis of the
precision of the laboratory work, as noted in the discussion above.  Therefore,  a decision was made to
truncate the soil readings at 2,600 ppm because a natural break occurred in the data between samples with
moderate and relatively high values.  This cutoff only  dropped  13  soil samples out of 869  in the file,
leaving 856 for analytical purposes.  The dropped samples are listed in Appendix Table C.

3.5.3 Laboratory Comparisons by She for Dust Analysis

      The laboratory  analyses  of the dust samples were performed by  three laboratories: Midwest
Research  Institute in  Kansas City, Missouri;  Core  Laboratories in  Casper, Wyoming;  and  Core
Laboratories in Aurora, Colorado. This section investigates whether each of the three laboratories, on the
average, determined lead concentrations equally.  If the  samples were randomized in the field  and sent in
batches to the three laboratories equally, it could be assumed lead concentrations in the dust were equally
distributed across the laboratories and a simple analysis comparing the results from all three laboratories
would be possible.  Therefore, an analysis across the 30 counties was conducted to determine how the dust
samples were distributed among the three laboratories.
                                             3-40

-------
      For the purpose of this analysis, the dust sample data OTS trimmed fiirther to include only those
samples with less than 5,000 ppm of lead.  The reason for the exclusion of the larger values was that they
distorted the comparison between the laboratories; a single large value could cause a laboratory average for
a given county to vary by a factor of two.  Even with this constraint, the variation between laboratories
within a site was large because the distribution of samples from counties to the laboratories was uneven
and the proportion analyzed by each laboratory in a single county was not equal Table 3-14 shows that
there was wide variation in the allocation of the dust samples sent to the laboratories (the samples were not
randomized between laboratories).  Even if the proportions had been equal, however, the large variation in
dust lead content within each county would mask any possible  distinction among the laboratories.  The
conclusion made from the analysis is that a determination of significant differences among the laboratories
can not be made.

3.5.4  Laboratory Comparisons by She for Soil Analysis

      Soil samples were only analyzed by two of the laboratories mentioned above-Core Laboratories in
Casper and Midwest Laboratories  in Kansas City.  Table 3-15 shows the total number of soil  samples
included in the analysis, the average amount of lead (in ppm)  and the standard deviations for  the two
laboratories conducting the analysis. Table 3-12 shows that samples were not randomly distributed among
the laboratories, therefore, no conclusions about the accuracy of the laboratory work can be drawn from
this analysis for the same reasons as mentioned above.

3.5.5  Evaluation of the Laboratory Dust Analysis Method

      Graphite furnace atomic  absorption (GFAA) spectrometry was chosen for the  analysis  of dust
samples over  other methods, specifically over the less  expensive inductively coupled argon plasma
spectroscopy (TCP),  because GFAA is  more sensitive — it can detect and quantify lower levels  of lead.
This decision,  made in the design phase of the survey, was based on pretest information that suggested
small  dust sample sizes could be expected and mat possible low dust lead concentrations in the  samples
should necessitate the most sensitive analytical method available (GFAA).  Because GFAA is considerably
more  expensive than ICP analysis, a question arises  as to what data would have been lost if  the less
expensive method had  been used.   By looking  back at the data, and by using pretest information on
        . sample weights required for ICP analysis, an evaluation of the quality of data is possible.  The
minimum
number of samples from the trimmed file of 1,974 samples mat would not have met the mmtnmm dust
weignt requirement for analysis by the ICP is 210, or about 11 percent How different are these samples in
terms of ppm and lead loading?

      The samples too small for analysis by the ICP method had smaller average concentrations (678
ppm) than the total dust sample file (1571 ppm), although this particular measure is  subject to high
variation  for small samples as discussed in the next subsection.  The samples mat could not have been
analyzed by the ICP method represent cases in which the lead loading (the measure probably most closely
related to exposure) is quite small  These samples had a mean loading of 0.2 ug/sq & versus an overall
mean of 41.6 ug/sq ft The maximum loading of these samples is about 1.3 nricrograms of lead per square
foot vacuumed. Compared to HUD dust lead guidelines, these small loading values are trivial and most
likely do not represent a hazard.  Since loading is reflective of the amount of dust collected, the tautology
(Le., a clean house has little dust and hence less dust lead exposure), is repeated here.  In any case, it
appears that little would have been lost in the identification of hazardous levels of lead in housing if the ICP
method had been used in place of the GFAA method, but national estimates of the prevalence of dust lead
levels would not have been as accurate.
                                             3-41

-------
                                     Table 3-14

       NUMBER OF DUST SAMPLES, AVERAGE PPM OF LEAD AND STANDARD DEVIATION BY
           LABORATORY AND SITE, FOR SAMPLES WITH LESS THAN 5,000 PPM OF LEAD
           Number of Samples
Standard Deviations
   NOTE:

C=CORE LAB, CASPER, WY
A=CORE LAB, AURORA, CO
M=MRI, KANSAS CITY, MO
                                      3-42

-------
                                Table 3-15

   NUMBER OF SOIL SAMPLES, AVERAGE PPM OF LEAD, AND STANDARD DEVIATION BY
      LABORATORY AND SITE, FOR SAMPLES WITH LESS THAN 2,600 PPM OF LEAD

Site

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30

All Sites
Number of Samples
C

28
30
38
0
34
23
14
17
35
0
29
18
36
25
21
2
14
3
0
12
30
36
12
23
24
32
21
27
24
20

628
M

0
0
27
48
0
0
0
0
0
53
0
0
0
0
39
0
14
1
2
0
0
0
0
0
30
0
0
14
0
0

228
Avera
C

28.97
371.98
226.03

105.52
67.15
391.61
97.37
261.39

119.17
479.30
108.88
104.35
30.57
452.88
112.07
66624

68.83
155.87
38.36
139.62
285.09
99.33
36.02
137.55
38.44
193.44
384.09

160.2
geppm
M



154.21
196.63





341.16




242.19

160.17
974.44
852.08





85.24


147.64



97.12
Standard Deviation of ppm
C

32.03
530.95
348.73

166.58
52.84
592.56
225.10
637.93

148.30
440.15
210.71
296.89
43.82
74.50
195.71
418.44

99.54
320.64
91.34
92.05
453.38
151.21
23.65
370.50
25.37
332.93
583.36

320.09
M



191.85
403.79





551.01




222.76

89.86

115.13





72.49


160.97



112.29
NOTE:

C=CORE LAB, CASPER, WY
M=MRI
                                   3-43

-------
3.5.6       The Problem of Small Dust Weights

      The Midwest Research Institute laboratory report states on page B-15, "A minimum of 10 mg of
dust will be needed to achieve method detection limits suitable  for the data quality objectives of this
Survey."  Many of the dust samples however weighed less than this minimum and were analyzed  Of the
1,974 dust samples in the analysis file, 669 or 35 percent weighed less than 10 mg and 448 or 23 percent
weighed less than 5 mg.  These fractions were consistent among the three laboratories. This subsection
examines the ppm and lead loading for the smallest samples, ie.,  those under 5 mg.  The weights of the
dust samples were "tap weights" obtained by tapping out the dust from the collection container.  This
method does not yield reliable weight data, especially for small dust samples.

      Table 3-16 provides a summary by weight of the 448 samples under 5 mg. Note that the 51 smallest
sample weights were, presumably, rounded to 0.1 mg.  Only one of these samples with 0.1 mg of weight
was deleted from the analysis file, though (reading number 116 in Appendix Table C-l), and that sample
was dropped because of an excessively high ppm reading value.

      The average ppm in the total analysis file equaled 1,571. Table 3-16 shows that most of the ppm
averages for samples under 5 mg exceed that average. This is not surprising in view of the smallness of the
denominator in the ppm computation and the uncertainty with which that denominator is measured.

      The lead loadings per square foot are more critical to the objectives of the study, however, and they
tend to be quite low. Only five measurements among the samples weighing less  man 5 mg exceeded 50
rmcrograms per square foot of surface. The five readings are 57,58,71,86, and 152.  Between the sample
weights of 5 mg and 10 mg the samples representing more than 50  micrograms per square foot are 57, 75,
87, 88, 89,287, and 529.  Adherence to the 10 mg cutoff would have ftlJmjnated several sample cases mat
appear to be significant  The real and unknown issue however, is  how accurate the readings are for such
small sample weights.  This issue could be explored further by comparing the post digestion duplicate
results for small dust-weight samples  Yet, this comparison could not be  made because the post-digestion
        for the duplicate procedures were not identified by weight of the dust sample in the analysis file.
                                             3-44

-------
                                  TABLE 3-16

  DISTRIBUTION OF NUMBER OF DUST SAMPLES, AND AVERAGE AND STANDARD
 DEVIATIONS OF PPM AND LEAD LOADING PER SQUARE FOOT Qig/ft2) FOR SAMPLES
                      WEIGHING LESS THAN 5 MILLIGRAMS
Dust weight
0.1
0.2
0.3
0.4
0.5
0.6-1.0
1.1-1.5
1.6-2.0
2.1-2.5
2.6-3.0
3.1-3.5
3.6-4.0
4.1-4.5
4.fr4.9

No. of samples
51
13
18
14
13
51
50
42
42
38
35
35
19
27
ppm
Average
7,185
2,397
6,231
1,378
5,295
4,116
1,687
3,674
1,319
986
3,874
886
1,228
986
Lead Loading per sq. ft.
StiDev.
10,192
2,795
11,843
1,130
11,218
11,957
2,753
13,813
3,046
1,903
16,543
1,226
2,956
1,452
Average
0.59
029
1.16
0.66
1.82
1.86
1.88
6.37
3.36
1.27
4.71
4.07
5.38
3.61
Std-Dev
0.97
0.39
2.10
0.87
5.17
4.19
3.65
23.93
11.09
2.68
15.23
10.80
14.32
6.29
AUwts.
448
3,026
9,033
                                                             2.77
                                                  10.37
3.5.7      Quality of Laboratory Data-Conclusion

     The laboratory work was carefully done, both in terms oTthe designof the accoin^flity, ihe
quality control procedures, and the execution. The resulting high data quahly aUows for meamngfiu
statistical analysis to predict national estimates of residential lead m dust and sofl.
                                      3-45

-------
                                     4. CONCLUSIONS
4.1   Major Conclusions
      This section presents a brief discussion of the major conclusions to be derived from the
experiencejjftfae national survey.

4.1.1  Study Findings

      Lead-based paint is widespread in housing.  An estimated 64 million homes, 83 percent of the
privately owned housing units built before 1980, have lead-based paint somewhere in  the building.
Twelve million of these homes are occupied by families with children under the age of seven years old.
An estimated 49 million privately owned homes have lead-based paint in their interiors.  There are no
statistically significant differences in the prevalence of lead-based paint by type of housing, market
value  of the home, amount of rent payment, household income, geographic region or degree of
urbanization.

      Thirteen million homes - 17 percent of the pre-1980 stock - have dust lead levels in excess of the
federal guidelines, regardless of whether or not they have lead-based paint  However, excessive dust
lead levels are associated with the presence of damaged lead-based paint Fourteen million homes, 19
percent of the pre-1980 housing stock, have more than five square feet of damaged lead-based paint
Nearly half of mem (47 percent) have excessive dust lead levels.

      While a large majority of pre-1980 homes have lead-based paint, most of them have relatively
small amounts of it The average privately-owned housing unit with lead-based paint has an estimated
601 square feet of it on interior surfaces and 869 square feet on exterior surfaces.  Over half of the
leaded paint is on walls, ceilings, and floors.  The amounts of lead-based paint per housing unit vary
with the age of the dwelling unit  Pre-1940 units have, on average, about three times as much lead-
based paint as units built between 1960 and 1979.

      Lead paint is even more widespread in public housing; 86 percent of all pre-1980 public housing
family units have lead-based paint somewhere in the building. While most public housing units have
some lead-based paint, most of them have small ammmts of it  The average public housing unit with
lead-based paint has an estimated 367 square  feet on interior surfaces and 133 square feet on exterior
surfaces. Most of the interior lead-based paint is on walls, while very  little of the exterior walls are
painted.

4.1.2  Impact of Measurement Error and Lead Concentration Variations on die Data Analysis

      The data analyses  and reports  of findings should incorporate instrument and laboratory
      measurement error.
                   analyzer MAP/3TRF instrument produces readings with measurement errors. They
are systematically different from the actual lead concentrations in me painted surfaces and have random
variation.   Similarly, the laboratory protocols used to measure the lead in dust and soil have
measurement errors.  These measurement errors can induce systematic encro m the estimated extent of
the lead hazard from paint, dust and sofl, and can also result in underestimates of the uncertainty in the
estimated hazard.

      Therefore, the field procedures must provide for the collection of the QA data necessary to
estimate the measurement errors and their iinpact rathe study findings.  Further, the data analyses must
explicitly estimate and correct for the impact of the measurement errors.
                                             4-1

-------
      The data analyses and reports of findings should take account of the inherent variation in
      the painted surfaces.

      In the national survey, it was not possible to test for the lead content of every painted surface in
every sampled housing unit.  Consequently, surfaces wore sampled, as described in Appendix I.  This
sampling protocol was designed to control the project costs and respondent burden by controlling the
amount of time required for an inspection. However, it did not adequately provide for the estimation of
variation in lead  concentrations within  surfaces, between surfaces in the  same room, and  between
rooms. These sources of variation need to be addressed to produce accurate estimates of the uncertainty
in the national estimates.

      It  is therefore recommended that multiple MAP/XRF readings bo taken at randomly selected
locations on a subset of the selected surfaces; at two or more components of the same type in the same
room; and in two wet rooms and two dry rooms. Although it may not bo feasible to do this in all homes,
a subset should be selected for these additional readings.

4.1.3 Use of the Spectrum Analyzer MAP/XRF

      In contrast to the  HUD Interim Guidelines, substrata correction is a necessary step in the
      accurate determination of the presence  and amount of lead-based paint on surfaces, when
      using the MAP/XRF.

      In the  national survey, the  MAP/XRF generally produced readings that were systematically
different  from the amount of lead  in the paint being tested.  The direction and magnitude of the
systematic differences were related to the substrate material, the  lead loading in the paint, and, to a
lesser extant, the age of the Co97 source.  The exact nature of the relationships varied significantly from
one individual MAP/XRF to another.   Furthermore, the precision of the readings depended on the
substrate.  Therefore, substrate corrections are needed to obtain  accurate measurements  of lead
loadings. There are two possible ways to do this:

      1.     Take frequent validation readings, as described in Appendix I, and analytically correct the
            readings using methods as described in Chapter 3. Readings need to be done on three or
            more different shims however, not just two shims as in the national survey.  With only two
            shims, only a linear model  can bo used to correct the MAP/XRF readings.  It is  possible
            however, that a non-linear model better describes the  relationship  between the readings
            and the actual lead concentrations in the painted surfaces.

      2.     Perform substrate corrections in the field. The HUD Interim Guidelines describe substrate
            correction procedures appropriate for direct reading MAP/XRFs, not  the spectrum
            analyzer MAP/XRF.  At present, HUD is developing field substrate correction procedures
            for the spectrum analyzer MAP/XRF.


4.2   Additional Conclusions and Recommendations

      The main purpose of this section is to identify lessons learned during the conduct of the National
Survey and to develop recommendations for future field operations.

      Objective of Recommendations

      The objective of many of the following recommendations is to improve the representativeness and
statistical validity of the  data, e.g., develop methods to enhance respondent participation.   Other
recommendations concern the logistics of moving inspection teams around the country in an  efficient
and cost-effective fashion. Recommendations for improved in-home protocols are aimed at ensuring


                                            4-2

-------
that inspections are conducted correctly and completely,  In an effort to improve lead-based paint
totting, a Aill section ii devoted to recommendation! concerning the uso of the MAP/XRF.

      Applicability of Recommendations

      The National Survey operated under constraints that are not typical of other lead-based paint
studies.   Therefore, some of the following recommendations would not bo appropriate under other
research designs.  For example, most lead studies are conducted in a single location, alleviating the
problems associated with moving teams and equipment around the nation under a tight schedule.  The
units in the National Survey were randomly selected from the national population of dwelling units;
most other studies  limit themselves to a  smaller,  targeted population of units  which have  some
underlying rationale for  being included, e.g., they were FHA-mortgaged properties or they wore in a
targeted area of an inner  city. A rationale for each recommendation if provided and the reader must use
his or her judgment in determining if the recommendation is appropriate for another design.

      The recommendations  presented  below stem  from  lessons learned  in the  pretests  and in
implementing the final design.  They are grouped and discussed under the following headings:


      1.     Background of Field Operations Procedures

      2.     Pretest

      3.     Sample Frame Development for Private, Multi-family, and Public Housing

      4.     Field Activities

      5.     Dwelling Unit Visit and Inspection Protocol

      6.     In-Field Environmental Sampling

      7.     Use of the Spectrum Analyzer MAP/XRF

      Background of Field Operations Procedures

      Plans for field  operations were designed and pretested as part of a separate contract effort in
advance of the National Survey.  Based on the results of the pretest, a final survey design and field
recommendations were issued.  The contractor for the full National Survey evaluated the design and
recast certain  portions  to accommodate changes in  research objectives.  In  addition, the survey
contractor developed the schedule,  budget, and detailed field procedures that were not present in the
original design.  The  revised plan  was pretested and  further modifications  were made based on the
pretest results. The most significant and far-reaching change was the decision not to sample dwelling
unit rooms and architectural  features (components) in the  field.  Sampling was done based on the
information gathered in a telephone interview. This allowed the inspection team to complete work in
one visit per dwelling unit as opposed to  the two visits (i.e., one for statistical sampling, one for
inspecting and environmental sampling) called for in the original design.
      Pretest
            Pretests should be performed that test all study design features and technologies
            employed  in a  study  and  pretest  subjects  should  represent the  diversity of
            respondents and situations expected in the full study.
                                             4-3

-------
      For field studies, pretests done in diverse settings are essential to evaluating the study design and
technologies employed.  Although the initial pretest of the original design rendered many valuable
lessons, a number of aspects were not thoroughly pretested prior to making the recommendations for the
original design. This led to a number of unanticipated problems, specifically:

      •     The initial pretest included a large percentage of vacant units. This tended to present the
            field problems involved with resident contact.

      •     The initial pretest sample lacked diversity of circumstances, e.g., working m bad weather
            or at night.

      •     The initial pretest created lists of dwelling unit owners/managers of rental properties by
            reviewing telephone books and working with owner/managers to reach rentals. That was
            an impractical solution given the time and cost constraints of the full study. The sampling
            approach used in the National Survey (contacting dwelling unit resident first) was never
            pretested.

      Sample Frame Development for Private (Single and Multi-family) and Public Housing

      •     AD listing and screening should be completed before beginning unit inspections

      Completion of listing and screening before beginning unit inspections means that the full sample
frame can be developed before sampling is begun.  Because of the tight field schedule of the National
Survey, it was necessary to  begin inspections before listing and screening were completed in some
counties. Although it was possible to take samples from completed counties based on estimates of the
final national distribution of eligible housing, it was necessary after screening was totally completed to
adjust the final sample to the actual distribution. There were two ways to accomplish this: retroactively
adjust the sample hi the counties which had been previously sampled; or sample the remaining counties
to balance the sample to the national distribution. Although the former approach yields a better sample
statistically, it would have created costs that were deemed to outweigh the gain m statistical power. The
National Survey adopted the  second course,  adjusting the sample in the counties mat had not yet been
sampled.

      If schedule and budget permit,  it is desirable  to  complete all  listing  and screening before
beginning the sampling process, to produce a superior statistical sample.

      •     When screening vacant and high security buildings, additional methods  (other  than
            knocking on die door) of reaching dwelling unit residents should be in place.

      A number of buildings sampled for screening  were  inaccessible  because of high security
measures, uncooperative doormen, and unknown/inaccessible management companies. The field staff
did not have time to develop means of accessing these buildings (e.g., find owners and convince diem to
participate). Inaccessibility resulted in having to substitute dwelling units. The negative consequences
could have been lessened if clerical staff had been in place at the  listing/screening stage  to lend
assistance to field interviewers  in  contacting owners  and  management  companies.   "Crisscross"
directories Unking addresses to phone numbers could be referenced, for example. If a phone number is
available for a dwelling unit, the resident can be contacted and screened over the phone.

      •     During the initial listing and screening period, increased time should be allowed for
            listing and screening of dwelling units in rural areas.

      The sampling plan required the listing of all homes in the selected segments. The number of
dwelling units listed in a county varied from a low of 220 in Cascade, Montana (a small rural county),
to a high of 3,239 in Fairfax, 'Virginia (a large urban county). The time required to list all homes varied


                                            4-4

-------
immensely, not in relation to the number of dwelling units, but in relation to the area of the county in
squaremiles.  Small rural counties took about 50 percent longer to list than the more densely populated
urban counties.  Similar increases in "time to complete" were found for the screening phase, as well.

      •     Increased time should be allowed for listing and screening of dwelling units if weather
            is inclement

      Listing and screening of dwelling units requires that the interviewers spend a large amount of
time outdoors.  Their progress wffl be  substantially delayed if there is serious inclement weather.
During bad weather, the Survey had a higher rate of illness and attrition rate among interviewers. A
significant number of new interview staff had to be recruited, and in January, six weeks after the start of
the field period, a second 3-day training session was held.

      Attempts to screen during holidays were also unproductive. On the days near Thanksgiving and
Christmas, there was a notably higher rate of incomplete screener interviews due to people not at home
and refusals to take the time to answer screener questions.  Interviewers were less willing to work over
the holidays, and  the request mat they do so  contributed to the loss of some of the original field
interviewers.

      The net result of working during holidays and inclement weather was a higher non-participation
rate among dwelling units and waste of field staff effort

      •     Homes with two construction dates should carry the oldest date in the records.

      Some homes had a section built many years after the original construction date. In some cases a
newly added room may have been added after the cut-off date for inclusion in the study (1979). The
home was entered into me data base using the earliest date.

      •     For public housing  dwelling  unit sampling, several  additional  weeks  should  be
            allowed for dwelling unit sample selection in large PHAs.
                                                                     The reason varied, e.g.,
the contact was on vacation, there was difficulty finding the contact, the contact needed authorization,
etc. Without exception, selection of PHA dwelling units in the five largest PHAs in the sample took
several weeks longer man it did for me 25 other PHAs.

      Field Activities

      •     When dealing with PHAs,  interviewers  should closely coordinate with  the  PHA
            representative.

      Coordination with the PHA management staff is imperative to effective sampling and inspection
of public housing. To ensure that all responsible and affected parties are involved in the process, it is
advisable to  establish a liaison with both a PHA headquarters representative and the manager of the
specific housing project

      Often teams met the PHA representative at me PHA office and then went tothesite.  ThePHA
staff "smoothed" the way many times. In neighborhoods where the inspection teams felt unsafe, PHA
staff provided a vital service as escorts .

      Because, the PHA staff are busy, every effort must be made to accommodate their involvement.
For example PHA offices often are not located close to the dwelling unit  Inspection  teams should
schedule their time to allow mem to meet the PHA staff at the PHA office and go from mere to _the
dwelling unit The PHA representatives typically tried to inspect as many units on each trip as possible.


                                           4-5

-------
This consideration, combined with frequent unexpected delays, suggests that the inspection team should
not schedule other inspections too soon after the PHA work, to avoid schedule conflicts.

      Upon meeting a PHA representative in the field, the inspection team should not assume that he or
she has been fully briefed on the study or the team's mission. The interviewer needs to go over the
objectives and procedures of the survey and the importance of the PHA. dwelling unit inspections.

      •     field staff should be trained to respond to  anyone's questions concerning what they
            are doing and what the study is about. Do not attempt to use local police or similar
            agencies, or the regional HUD office, to substantiate or verify presence or activities.

      In the National Survey each field person carried an ID badge, copies of the license that allowed
him to handle the MAP/XRF and its radioactive source, endorsement/introductory letters from survey
and agency principals, the 800 number for the survey coordinator, and a number for the Washington
HUD representative.  Field staff were  questioned on several occasions by police officers,  building
mangers, or nearby residents for walking around a house after dark to take the MAP/XRF readings and
soil samples (the MAP/XRF looks like a gun, especially at night).  The documentation materials listed
above explained and substantiated the activities of the inspection tea™ and proved invaluable.

      Although  it seems like a good idea,  contacting local police in  advance and men referring
questioners to the police can lead to a bigger problem; Le., the police contacts inadvertently denying
knowledge of the survey.  If the person who gets the call at the police station has  not been apprised of
the study, he or she will deny knowing anything about it, even though the police  department has been
informed of the activities. There is also a problem with overlapping jurisdictions.  The study might
notify the comity sheriff, although the inquiry comes into the state police.

      •     The inspection team should  have  special  training and support  in  working  in
            potentially unsafe neighborhoods. Two sub-recommendations follow.  One concerns
            the safety of the inspection team, the second the safety of the equipment

      It was reported in both the initial pretest of the original design and in the National Survey mat
inspection teams were reluctant to enter certain public housing. The issue was not public housing, per
se, but one of being a stranger in a potentially unsafe neighborhood, carrying expensive equipment and
driving a rental car. Having a PHA official escort the inspecting team significantly helped to alleviate
these concerns.  Official  escorts  also  meant that the visit took place during working hours,  Monday
through Friday.  Hence, lessened anxiety when escorted may have been a function of the time of day
much as the escort itself.  Visits in the evenings and on the weekends were regarded quite differently by
the field teams.

      •     Safety of Inspection Team:  Potentially unsafe neighborhoods should be identified at
            the screening stage.  Trips should be scheduled to occur during daylight hours on
            weekdays;  operational and budgetary provision should  be made for a paid escort
            (e.g., off-duty police officer) to accompany the inspection team.

      Operating in potentially unsafe neighborhoods proved very problematic  for  the two person
inspection teams.  Listing and screening were more problematic because the interviewer, usually a
woman,  was working alone.    Screening and inspection  completion  rates were  low  in unsafe
neighborhoods.  Accommodations requested by the field staff included: extended time in the county so
appointments could be scheduled  exclusively during weekday, daylight hours; and paid escort, e.g., an
off-duty police officer.  Addressing the perceived and real dangers to the field staff is essential if they
are to produce complete results in certain neighborhoods.

      •     Safety of Equipment:  The wfthm-unit procedures need to be seqnenced so all tests
            with one set of equipment are completed before the other set is needed.


                                             4-6

-------
      In^ regard to the safety of the equipment, there are two issues:  the safety of the
        ^
(MAP/XKF, vacuum, supplies, etc.) from being stolen or damaged, and the safety of me general public
if unauthorized persons tamper with unattended equipment.

      The equipment in use was considered adequately supervised. Equipment in a car trunk also was
generally secure. Therefore, the difficulty lay inassuring that only the set of exnupment currently being
used (MAP/XRF or vacuum) was out of the car.

      The easy part of the solution was to remove only one set from the car trunk, perform all sampling
requiring that equipment, return it permanently to the car,  and retrieve the next set. Unnecessary trips
to the car wasted time, disconcerted the residents, and brought undue attention to the activity.  The
described approach minimized trips back and forth to the car. However, difficulties with this approach
arose when;

      •     There were several common areas to be inspected, e.g., playgrounds, laundry room.

      •     The inspector had to park the car some distance away, forcing him to spend the time
            walking and putting himself at risk carrying equipment around unsafe areas.

      •     The equipment was in jeopardy of being abused or stolen while in the car.

      •     The equipment  was at risk  of being damaged while in the trunk (e.g., excessive heat or
            water leakage).

      •     There was a danger the car would be stolen wim equipment in it.

      Dwelling Unit Visit and Inspection Protocol

      •     A member of the inspection team, preferably die team member with responsibility for
            operating  testing  and  sampling equipment,  should have  general  training hi
            engineering and architectural terms and project-specific training in terms used for
            the data collection.

      The initial pretest and the survey itself both encountered difficulties in uniformly and consistently
categorizing the conditions  of walls  and paint,  identifying  substrate  materials,  and  categorizing
architectural components into study-specific  categories.   Specific architectural, construction,  and
engineering expertise qhoiild he brought together in advance wim survey staff to work out the exact type
of background needed for the inspector  and the study-specific categories for recording architectural
features  and  conditions.    The  effort should go  as  far as  specifying training  in  pertinent
architectural/engineering areas and in the use of the study-specific categories.

      The underlying issue is the uniformity and accuracy with which architectural components were
named (categorized) and evaluated.

      •     Sampling of rooms and components should be conducted in advance of the inspection
            visit to the home,  based on home inventories conducted over  the telephone (or a
            previous visit) and data collection forms  should be customized in advance  for the
            sampled rooms and components.

      The final pre-field test of the original design found that in-field sampling of dwelling unit rooms
and components was  likely to create a source of error.  Additionally, he pre-field test had the potential
to waste the residents' time and convey a sense of disorganization on the part of the field team and the
entire survey.  In-field selection lengthened the inspection by 30  to 50 niinutes, taken up with what
                                             4-7

-------
appeared to the residents as confused paper shuffling  Frequently, just as the team was finally prepared
to begin the actual inspection, the resident was ready for them to leave.

      In an effort to minimise in-field selection, the final in-field sampling protocol divided the room
and component sampling task into three parts: randomly numbering the four exterior walls of the house
or apartment building; selecting a wet and a dry room; and selecting architectural components.

      When the inspection team arrived at a dwelling unit, the interviewer noted  which wall of the
house faced the street (as named in the address of the house).  That became wall 1.  Going  clockwise,
the remaining three walls were numbered 2, 3, and 4.  That scheme was applied to interior and exterior
sampling. This was the only sampling thai the inspection team

      With regard  to wet or dry room selection, residents of units initially sampled from the  field
screening provided an inventory of the home on the phone.  This allowed the wet or dry room sampling
to be done at the field headquarters, prior to the field effort, and printed on the Interior Observation
Form customized for each unit. Using the Interior Observation Form,  the  field  team could easily
designate which components should be selected and tested, based  on a random  sampling  priority
specifically generated for that unifs components.

      The use of pre-printed, customized forms helped alleviate the problems of wasted time, errors,
and omissions by the field teams while performing the inspection and testing in the unit

      •     As long as the inspection can be completed in one trip, dwelling unit sketches are not
            necessary.

      Because the  original design  called for two trips to the dwelling unit,  one to gather sampling
information and one to inspect, dwelling unit sketches were needed under mis plan. Under the single
visit design, mere was no need for sketches.

      •     Allow additional time when inspecting public housing and private multi-family units
            for common areas and coordination with escorts.

      Inspections that include the additional common areas cannot be accomplished in the same amount
of time as  single-family units.  Longer time periods should be automatically blocked  out on the  field
team's scheduling calendar whenever the inspection is to include common areas.

      •     XRF scores (measurements) should not be repeated out loud in the dwelling unit

      The original  design called for the inspector to read the XRF reading out loud and for the
interviewer to  copy it down.  This procedure raised the resident's curiosity and led  to a barrage of
questions.  Although it was important to be open with the residents concerning all aspects of the study,
reciting technical data in the midst of the inspection needlessly alarmed them and often had the opposite
effect

      In-Field Environmental Sampling

      •     The purposive XRF  paint lead  reading should be collected either from a  pre-
            established  location in each dwelling unit or only in dwelling unit areas previously
            entered as part of other testing.

      The study design called for the field technician to perform an XRF reading at a spot he thought
had a high chance of having leaded paint It was not reasonable, though, to expect the dwelling unit
resident to  allow the inspector to walk through the entire  house looking around. Further, such a foray
took a lot of time. One of the following alternatives seems preferable:


                                             4-8

-------
      •      Specify on a study-wide basis that a reading will be taken ina standard location that has a
             high probability of having lead-based paint

      •      Employ the final protocol established for the National Survey, viz., limit the search to
             dwelling unit areas entered as part of the rest of the inspection, including the two rooms
             sampled and areas walked through getting to those rooms.

      •      The analytic model concerning dust analysis and pathways needs to be fully specified
             before an adequate dust sampling protocol can be designed.

      Dust sampling and test result analysis are the subject of continuing investigations. Insufficient
dust was a recurring problem in the field. Outside of insufficient dust, mere were few field problems
encountered with dust sampling.

      A traveling team operates under two major constraints — it must carry its own equipment and it
must limit its time in the dwelling unit to a reasonable length.  Getting more dust by using a bigger,
more powerful vacuum or vacuuming a larger area were limited options.  The vacuum carried by the
survey team weighed ten pounds, plus tubing, nozzles, and extension cords.  The National Survey found
that it took four minutes to completely collect a dust sample (vacuuming and re-vacuuming a 4-square-
foot area as specified in the dust sampling protocol).  A bigger vacuum or more time spent vacuuming
were not reasonable in the context of the National Survey.

      At one point in the procedure design stage it was suggested mat in-field technicians evaluate the
quantity of dust collected in a cassette.  They would collect more dust in the cassette if they determined
there was insufficient dust  This suggestion proved to be irnpractical   First, inspectors  could  not
reliably determine by visual inspection if enough dust had been collected.  Second, movement of the
cassette while attached to die vacuum could cause dust to 611 out of the cassette. Third, allowing the
technician individual discretion to collect samples would lead to inconsistencies in the procedure and
findings  Last, the amount of time spent vacuuming had to be limited to keep the visit to a reasonable
length. Therefore, this suggestion was not implemented in the National Survey.

      A final suggested  approach to increasing dust sample yield was  "wet  wipe" testing, used in
conjunction with or following vacuuming, to pick up leftover dust  This technique would not have
improved effective yield, because  wet wipe test results  could not be compared or added in any
meaningful way to vacuumed dust sy^p^c results.

      Static electricity posed a problem to the inspection team. Dust would cling to the vacuum nozzle,
the edges of the template, etc. Efforts were made to build the template out of a material that did  not
accumulate static electricity but the phenomenon stifl occurred.

      •      XRFs can be transported safely by air (as luggage) or by Federal Express or other
             carrier in plain wrapping.  Interior wrapping must warn of the radioactive source.
             Authorizations for transport should be packed inside of the exterior wrapping but
             still accessible without opening the interior XRF case.

      There was much discussion about transporting the XRF, given its radioactive source. Because
the National Survey equipment was never detained in transit, the above packaging approach appears to
be an effective one  One member of the team must be authorized and licensed to transport and use the
XRF.

      Use of the Spectrum Analyzer MAP/XRF

             Make sure licenses for MAP/XRF use are amended for the intensity of source used.
                                            4-9

-------
      A member of the inspection team must have a state-issued license that allows him to transport
and use the MAP/XRF. Licenses are, typically, tied to the intensity of the radiation source and must be
valid for the intensity being used.  In most cases, the authorization for the higher millicurie level cost
more, sometimes by several hundred dollars.  The licenses should be applied for at least two months
before the scheduled beginning of the field period.

      All but one of the states in the National Survey required detailed information about where the
MAP/XRF would be, when and now long it would be in the state, and who would be responsible for it
Several states required the addresses of the sampled homes and dates of inspections. The study staff
needs to be prepared  to supply this information to those responsible for the MAP/XRFs and their
licensing.

      •      Use a "full intensity" radiation source in the XRF.

      The XRF should have a full intensity radiation source to get the best readings in the minimum
amount of time.  Full intensity radiation presented no additional risk to residents, technicians^ or the
general population. The National Survey used a 40 millicurie source and 1-minute-long readings.

      •      The XRF reading can be taken at any convenient place on a sampled architectural
             component; valid sampling does not require readings of a randomly selected spot on
             the component  This eliminates readings being taken in hard-to-reach locations, such
             as corners.

      The original design called for sampling a random location on the surface  of a component for
testing.  Though this is subject to further verification, it appears that components with common paint
history produced similar XRF readings.  Another stage  of samplmg would significantly burden the
inspection process.  Correct assessment of "common paint history" is more pivotal here.

      •      The MAP/XRF should not be used to  scan or take an  "average" reading on a
             component.

      The original MAP/XRF design called for "scanning" components with the MAP/XRF by running
the "eye" across all parts of a component The initial pretest results did not support this practice.  The
final survey protocol eliminated scanning. Experience with the MAP/XRF leads to the belief that the
readings produced tins way could be subject to unpredictable error because of the inability to evenly and
smoothly scan a component's surface.

      •     All MAP/XRF readings  should be recorded on paper by the inspector.

      The MAP/XRF used in the National Survey was programmed to store the spectrum results in
memory.  Serious problems were encountered, though, when the memory was in use.  The  memory
would fill up rapidly and the MAP/XRF would stop operating. The equipment had to be turned off, and
then back on, in order to  restore operation.  All memory was lost in the process.  The need to link
readings to a  location necessitated recording of a certain amount of data on paper in any event, so mat
recording MAP/XRF readings on paper involved little extra effort.

      The equipment problem aside, using the MAP/XRFs on-board electronic  storage necessitates
some procedures for downloading memory contents during the field period. In practice it did not prove
reasonable  or practical to return the MAP/XRF to the survey operations office periodically  for
downloading. Teaching field staff how to download the memory and transmit it to the field operations
office would have meant providing them with a properly-configured PC and modem in the field and
providing PC training  Adding responsibilities and equipment to the technicians' load was deemed
inadvisable.
                                           4-10

-------
      •     MAP/XRF should be equipped with  new batteries frequently so they are always
            running at full power.  Batteries can freeze in transit (e.g., in the cargo hold of a
            plane) and lose their charge.

      Uneven power supply from batteries seemed to be associated with quirkiness in the operating of
the MAP/XRF. The reliable response seemed to be simply installing new batteries (purchased locally)
upon arrival in each county and again after several dwelling unit inspections. As noted above, if new
batteries do not clear up the quirkiness in MAP/XRF performance, the console needs to be returned to
the manufacturer for adjustment  There are no in-field adjustments possible.
                                             4-11

-------
      APPENDIX A

Additional Data Tables for
     Private Housing

-------
                        TABLE A-l

        PREVALENCE OF LEAD-BASED PAINT (LBP) BY
               LOCATION IN THE BUILDING -
       PRIVATELY OWNED OCCUPIED HOUSING UNITS
Location of LBP
Unit Interior
Interior Common Area
Building Exterior
Playground
Somewhere in Building
Occupied Housing Units
With Lead-Based Paint
Number (000)
48,986
3,596
56,495
525
64,443
Percent (1)
63%
5%
73%
1%
83%
(1) Base equals all 77,177,000 housing units built before 1980.
(2) Numbers based on small sample sizes should be interpreted with caution.
                           A-l

-------
                                        TABLE A-2

        ESTIMATED NUMBER OF PRIVATELY OWNED OCCUPIED HOUSING UNITS
     BUILT BEFORE 1980 WITH LEAD-BASED PAINT BY SELECTED CHARACTERISTICS
                          (Paint Lead Concentration >= 0.7 mg/sq cm)
Characteristic
Total Housing Units Built Before 1980
One or More Children Under Age 7
Construction Year
1960-1979
1940-1959
Before 1940
Housing Type
Single Family
Multifamily
Total
Housing
Units (000) (1)
77,177
13,912
35,681
20,476
21,018
66,418
10,759
Housing Units
With Lead-Based Paint
Somewhere in Building
Percent
87%
(5%)
92%
(6%)
82%
(8%)
94%
(5%)
88%
(11%)
87%
(5%)
83%
(11%)
Number (000)
66,831
(3,670)
12,783
(863)
29,195
(2.708)
19,210
(1.099)
18.426
0254)
57,926
(3,462)
8,905
(1,160)
Number of
Housing Units
in Sample
284
90
120
87
77
227
57
(1) Total units data are from the 1987 American Housing Survey.

Note: Numbers in parentheses are approximate half-widths of 95% confidence intervals for the estimated
percents and numbers. For example, the approximate 95% confidence interval for the percent of housing units
with some lead-based paint is 87% +/- 5% or 82% to 92%.
                                        A-2

-------
                                        TABLE A-3

         ESTIMATED NUMBER OF PRIVATELY OWNED OCCUPIED HOUSING UNITS
     BUILT BEFORE 1980 WITH LEAD-BASED PAINT BY SELECTED CHARACTERISTICS
                          (Paint Lead Concentration >= 12 mg/sq cm)
Characteristic
Total Housing Units Built Before 1980
One or More Children Under Age 7
Construction Year
1960-1979
1940-1959
Before 1940
Housing Type
Single Family
Multifamily
Total
Housing
Units (000) (1)
77,177
13,912

35,681
20,476
21,018

66,418
10,759
Housing Units
With Lead-Based Paint
Somewhere in Building
Percent
80%
16%)
85%
(8%)

69%
(9%)
89%
(7%)
88%
(11%)

80%
(6%)
75%
(12%)
Number (000)
61.475
(4.336)
11,873
0.118)

24,769
(3.236)
18,281
(1.411)
18,426
(Z2S4)

53.423
(4.113)
8.052
(U33)
Number of
Housing Units
284
90

120
87
77

227
57
(1) Total units data are from the 1987 American Housing Survey.

Note: Numbers in parentheses are approximate half-widths of 95% confidence intervals for the estimated
percents and numbers. For example, the approximate 95% confidence interval for the percent of housing units
with some lead-based paint is 80% +f- 6% or 74% to 86%.
                                       A-3

-------
                                       TABLE A-4

        ESTIMATED NUMBER OF PRIVATELY OWNED OCCUPIED HOUSING UNITS
     BUILT BEFORE 1980 WITH LEAD-BASED PAINT BY SELECTED CHARACTERISTICS
                         (Paint Lead Concentration >= 2.0 mg/sq cm)
Characteristic
Total Housing Units Built Before 1980
One or More Children Under Age 7
Construction Year
1960-1979
1940-1959
Before 1940
Housing Type
Single Family
Multifamily
Total
Housing
Units (000) (1)
77,177
13,912
35,681
20,476
21,018
66,418
10,759
Housing Units
With Lead-Based Paint
Somewhere in Building
Percent
68%
(6%)
73%
(10%)
48%
(10%)
83%
(8%)
88%
(11%)
69%
(7%)
66%
(14%)
Number (000)
52,690
(5.013)
10,128
(1.407)
17,219
(3.509)
17,045
(1.703)
18,426
(2,254)
45,602
(4.810)
7,088
(1.457)
Number of
Housing Units
in Sample
284
90
120
87
77
227
57
(1) Total units data are from the 1987 American Housing Survey.

Note: Numbers in parentheses are approximate half-widths of 95% confidence intervals for the estimated
percents and numbers. For example, die approximate 95% confidence interval for the percent of housing units
with some lead-based paint is 68% -f/- 6% or 62% to 74%.
                                      A-4

-------
                                                                                            TABLE A-5

                                                                     ASSOCIATION BETWEEN LEAD IN SOIL AND EXTERIOR
                                                           LEAD-BASED PAINT CONDITION FOR PRIVATELY OWNED HOUSING UNITS
                                                                           (Numbers Represent Thousands of Housing Units)

'resence and
Condition of Exterior
Lead-Based Paint

Exterior Walls
NoLBP
LBP Present, Intact
LBP Present. Non-Intact
Playgrounds
NoLBP
LBP Present. Intact
Total
Lead In Soil
Entrance
Within
Guideline (1)
Number

18.266
38.782
5.099

61.623
522
62,147
Percent

98%
8S%
59%

87%
100%
87%
Exceeding
Guideline (1)
Number

441
5.448
3.537

9.426
--(2)
9,426
Percent

2%
12%
41%

13%
--
13%
Drip line
Within
Guideline (1)
Number

17.506
37.066
3.976

58.026
522
58.547
Percent

92%
86%
48%

83%
100%
83%
Exceeding
Guideline (1)
Number

1,447
5,989
4.310

11.746
•-
11.746
Percent

8%
14%
52%

17%
--
17%
Remote
Within
Guideline (1)
Number

18.725
40.748
6,741

65.692
522
66,214
Percent

100%
92%
82%

93%
100%
93%
Exceeding
Guideline (1)
Number

--
3.387
1.450

4,837

4,837
Percent

--
8%
18%

7%
-•
7%
Any Location
Within
Guideline (I)
Number

17.719
35.914
3,815

56.926
522
57,448
Percent

91%
80%
44%

78%
100%
79%
Exceeding
Guideline (U
Number

1.660
9.215
4.821

15.695
--
15.695
Percent

9%
20%
56%

22%
--
21%
(I) Although there la no federal standard for residential soil lead contamination, many experts agree that 500 ppm Is a feasible threshold to designate "high" soil lead
   contamination In residential environments. BPA's Interim guidance on soil lead cleanup levels at Superfund sites sets the cleanup levels at 500 to 1000 ppm
   IDS BPA(Sepl. 7,1989). Interim Guidance on Establishing Soil Lead Cleanup Levels at Superfund Sites (OSWBR Directive #9355.4-02)).

(2) All the paint was within the Guidelines.
(3) There were no playgrounds In the sample with non-Intact lead-based paint present.

-------
                                      TABLE A-6

         AMOUNTS OF LEAD-BASED PAINT (LBP) ON INTERIOR SURFACES BY
         PAINTED COMPONENTS AND YEAR CONSTRUCTED FOR PRIVATELY
                          OWNED OCCUPIED HOUSING UNITS
                           (LBP Concentration >= 1.0 mg/sq cm)

Components:
Walls/ceiling/floor
1960-1979
1940-1959
Built before 1940
Metal component (1)
1960-1979
1940-1959
Built before 1940
Non-metal component (2)
1960-1979
1940-1959
Built before 1940
Shelves/other (3)
1960-1979
1940-1959
Built before 1940
National Total Amount of LBP
(minions of
soft)


4,920
7,121
6,106

24
45
37

328
873
5,971

6
208
3.798
(percent of all
paint)


5%
15%
11%

2%
6%
3%

4%
9%
47%

0%
7%
68%
Amount LBP
Per Housing
Unit With LBP
(square feet)


281
505
351

1
3
2

19
62
343

0
15
218
(1) Includes metal trim, window sills, molding, doors, air/beat vents, and radiators.
(2) Includes non- aetal trim, window sills, molding, doors, and air/heat vents.
(3) Includes shelves, cabinets, fireplace, and closets, on any substrate.
Note: Because of rounding, totals may not be exactly the same as the sum of the numbers.
                                        A-6

-------
                                    TABLE A-7

       AMOUNTS OF LEAD-BASED PAINT (LBP) ON EXTERIOR SURFACES BY
        PAINTED COMPONENT AND YEAR CONSTRUCTED FOR PRIVATELY
                       OWNED OCCUPIED HOUSING UNITS
                         (LBP Concentration  >= 1.0 mg/sq cm)

Components:
Walls/ceiling/floor
1960-1979
1940-1959
Built before 1940
Metal component (1)
1960-1979
1940-1959
Built before 1940
Non-metal component (2)
1960-1979
1940-1959
Built before 1940
Porches/other (3)
1960-1979
1940-1959
Built before 1940
National Total Amount of LBP
(millions of
soft)


8,825
10,423
19,199

76
146
180

1,575
1,857
6.098

26
208
492
(percent of all
paint)


28%
45%
80%

4%
8%
13%

15%
39%
78%

2%
19%
13%
Amount LBP
Per Housing
Unit With LBP
(square feet)


405
625
1,066

3
9
10

72
111
338

1
12
27
(1) Includes only metal windows, doom, soffit and facia, columns, and railings.
(2) Includes non-metal windows, doors, soffit and facia, columns, and railings.
(3) Includes porches, balconies, stairs, etc., on any substrate.
Note: Because of rounding, totals may not be exactly the same as the sum of the numbers.
                                       A-7

-------
                               TABLE A-8
  GEOMETRIC MEAN PAINT LEAD LOADINGS IN PRIVATELY OWNED OCCUPIED
     HOUSING UNITS BUILT BEFORE 1980, BY SELECTED CHARACTERISTICS
                   (Paint Lead Concentration >= 1.0 mg/sq cm)
Characteristic
Total Occupied Housing
Units Built Before 1980
Construction Year:
1960-1979
1940-1959
Before 1940
Housing Type
Single Family
Multifamily
One or More Children
Under Aee7
Census Region
Northeast
Midwest
South
West
Interior Surfaces
(mg/sq. cm.)
0.1

0.1
0.2
0.3

0.2
0.1
0.1

0.3
0.1
02
0.1
( 0.1 .

( 0.0 ,
( 0.1 ,
( 0.2 ,

( 0.1 ,
( 0.1 .
( 0.1 .

( 02 ,
( 0.0 ,
( 0.1 ,
( 0.0 ,
02 )

0.1 )
03 )
0.6 )

02 )
0.2 )
02 )

0.6 )
02 )
02 )
02 )
Exterior Surfaces
(mg/sq. cm.)
0.3

0.1
0.4
1.6

0.4
02
0.3

0.8
0.6
02
02
(. 02 ,

( 0.1 ,
( 0.2 ,
( 0.4 ,

( 02 ,
( 0.1 .
( 02 .

( 0.3 ,
( 03 ,
( 0.1 ,
( 0.0 .
0.5 )

02 )
0.8 )
1.1 )

0.6 )
0.6 )
0.5 1

2.1 )
1.1 )
03 )
0.6 )
(1) Numbers in parentheses are 95% confidence intervals for the respective geometric means.
                                  A-8

-------
                        ^f              TABLE A-9

GEOMETRIC MEAN LEAD LOADINGS IN PRIVATELY OWNED OCCUPIED HOUSING UNITS
  BUILT BEFORE 1980, BY ARCHITECTURAL COMPONENT AND CONSTRUCTION YEAR
                          (Paint Lead Concentration >= 1.0 mg/sq cm)
Component/Construction year
Walls/ceilings/floor



Metal (1)



Non-metal (2)



Other (3)




1960-1979
1940-1959
Before 1940

1960-1979
1940-1959
Before 1940

1960-1979
1940-1959
Before 1940

1960-1979
1940-1959
Before 1940
Interior Surfaces
(mg/sq. on.)

0.05
0.1
02

0.05
0.05
0.05

0.1
02
0.9

0.01
0.03
020

( 0.0 ,
( 0.1 ,
( 0.1 ,

( 0.0 ,
( 0.0 ,
( 0.0 ,

( 0.1 ,
( 0.1 ,
( 0.5 ,

( 0.0 ,
( 0.0 ,
( 0.1 .

0.1 )
0.2 )
0.3 )

0.2 )
03 )
0.5 )

0.2 )
0.4 )
1.6 )

0.0 )
0.1 )
0.4 )
Exterior Surfaces
(me/so, cm.)

0.1
03
25

0.003
0.1
0.1

02
0.6
1.9

0.01
0.06
03

( 0.0 ,
( 0.1 ,
( 13 ,

( 0.0 ,
( 0.0 ,
( 0.0 ,

( 0.1 ,
( 03 ,
( 1-1 ,

( 0.0 ,
( 0.0 ,
( 0.1 .

0.2 )
0.7 )
4.9 )

0.1 )
1.1 )
1.8 )

0.4 )
1.1 )
3.4 )

0.0 )
03 )
1.0 )
      Note: Numbers in parentheses are 95% confidence intervals for the respective arithmethc means.
      Interior
       (1) Includes metal trim, window sills, molding, doors, an/heat vents, and radiators.
       (2) Includes non-metal trim, window sills, molding, doors, and air/heat veu«s.
       (3) Includes shelves, cabinets, fireplace, and closets, on any substrate.

      Exterior.
       (1) Includes only metal windows, doors, soffit and facia, columns, and railings.
       (2) Includes non-metal windows, doors, soffit and facia, columns, and railings.
       (3) Includes porches, balconies, stairs, etc., on any substrate.
                                                 A-9

-------
                                 TABLE A-10




DESCRIPITIVE STATISTICS FOR THE DUST LEAD CONCENTRATION MEASUREMENTS (WEIGHTED)

Set of Data
Number of measurements
summarized
Arithmetric mean (ppm)
Percentiles (ppm)
maximum
upper quartile
median
lower quartile
minimum
Geometric mean (ppm)
Mean of the log transformed
measurements
Dry room
floor

270
631

11,287
378
188
102
3
224

5.41
Entry way
floor

269
813

18,563
922
380
201
21
423

6.05
Wet room
floor

265
440

8,376
483
198
83
6
204

5.32
Dry room
window sill

207
5,264

96,492
2,466
735
259
1
925

6.83
Dry room
window well

78
10,186

457,178
6,326
1,962
536
5
1,792

7.49
Wet room
window sill

131
5,083

104,368
2,583
826
289
1
1,011

6.92
Wet room
window well

71
13,132

83,633
7,450
2,432
575
22
3,236

8.08

-------
                              TABLE A-ll
DESCRIPTIVE STATISTICS FOR THE DUST LEAD LOADING MEASUREMENTS (WEIGHTED)

Set of Data
Number of measurements
summarized
Arithmetric mean (ug\sq. ft.)
Percentiles (ug\sq. ft.)
maximum
upper quartile
median
lower quartile
minimum
Geometric mean (ug\sq. ft.)
Mean of the log transformed
measurements
Dry room
floor

273
6.92

205
3.43
0.96
0.31
0.00
1.12

0.11
Entry way
floor

274
12.68

380
8.0S
2.59
0.71
0.03
2.44

0.89
Wet room
floor

275
4.14

233
2.51
0.68
0.21
0.00
0.74

-0.30
Dry room
window sill

233
91.20

2,638
24.70
5.04
0.95
0.00
5.17

1.64
Dry room
window well

84
841.40

40,455
475.40
85.90
15.40
0.04
95.10

4.55
Wet room
window sill

158
96.80

11,899
11.98
2.15
0.44
0.00
2.50

0.93
Wet room
window well

74
790.62

7,139
528.10
90.26
18.83
0.19
121.98

4.80

-------
                                                TABLE A-12




                    DESCRIPTIVE STATISTICS FOR THE DUST LOADING MEASUREMENTS(WEIGHTED)

Set of Data
Number of measurements
summarized
Arithmetric mean (ug\sq. ft.)
Percentiles (ug\sq. ft.)
maximum
upper quartite
median
lower quartile
minimum
Geometric mean (ug\sq. ft.)
Mean of the log transformed
measurements
Dry room
floor

269
912,523

40,000,000
579,832
180,995
61,876
1,895
2,246

7.71
Entry way
floor

74
2,190,863

23,143,421
1,734,428
387,136
90,841
4,473
22

3.08
Wet room
floor

203
18,099,923

1,658,775,736
5,716,381
1,269.646
298,705
45
100

4.60
Dry room
window sill

222
2,374,765

153,676,471
427,842
82,701
12,133
59
49

3.88
Dry room
window well

48
316,816

8,018,868
37,445
11,870
3,478
48
4

1.49
Wet room
window sill

150
1,388,937

16,239,316
466,473
91,792
18,182
148
26

3.28
Wet room
window well

71
558,829

16,666,667
64,103
16,906
4,975
423
10

2.32
ro

-------
      APPENDIX B








ADDITIONAL DATA TABLES





  FOR PUBLIC HOUSING

-------
                                    TABLE B-l

           ESTIMATED NUMBER AND PERCENT OF PUBLIC HOUSING UNITS
    BUILT BEFORE 1980 WITH LEAD-BASED PAINT, BY SELECTED CHARACTERISTICS
                           (Paint Lead Concentration >= 0.7 mg/sq on)

Total Public Housing Units Built
Before 1980
Construction Yean 1960-1979
1950-1959
Before 1950
Total
Public Housing
Units (000)
910
455
273
182
Housing Units
With Lead-Based Paint
Somewhere in Buildine
Percent
90%
( 82% - 98% )
88%
( 76% - 99% )
90%
( 77% - 100% )
97%
( 88% - 100% )
Number (000)
821
( 748 - 895 )
399
(347 - 450 )
246
(209 - 273 )
177
(160 - 182 )
Number of
Housing Units
in Sample
97
43
24
30
Note: Numbers in parentheses are 95% confidence intervals for the estimated percents and numbers.
                                      B-l

-------
                                     TABLE B-2

            ESTIMATED NUMBER AND PERCENT OF PUBLIC HOUSING UNITS
     BUILT BEFORE 1980 WITH LEAD-BASED PAINT, BY SELECTED CHARACTERISTICS
                            (Paint Lead Concentration >= 1.2 mg/sq on)
Characteristic
Total Public Bousing Units Built
Before 1980
Construction Yean 1960-1979
1950-1959
Before 1950
Total
Public Housing
Units (000)
910
455
273
182
Housing Units
With Lead-Based Paint
Somewhere in Building
Percent
85%
( 77% - 93%)
77%
( 64% - 91% )
90%
( 77% - 100% )
97%
{ 88% - 100% )
Number (000)
774
(697 - 850 )
351
( 290 - 412 )
246
( 209 - 273 )
177
( 160 - 182 )
Number of
Housing Units
in Sample
97
43
24
30
Note: Numbers in parentheses are 95% confidence intervals for the estimated percents and numbers.
                                      B-2

-------
                                     TABLE B-3

            ESTIMATED NUMBER AND PERCENT OF PUBLIC HOUSING UNITS
     BUILT BEFORE 1980 WITH LEAD-BASED PAINT, BY SELECTED CHARACTERISTICS
                            (Paint Lead Concentration >= 2.0 mg/sq on)
Characteristic
Total Public Housing Units Built
Before 1980
Construction Year: 1960-1979
1950-1959
Before 1950
Total
Public Housing
Units (000)
910
455
273
182
Bousing Units
With Lead-Based Paint
Somewhere in Building
Percent
77%
( 67% - 86% )
65%
( 50% - 81%)
82%
( 65% - 98% )
97%
( 88% - 100%)
Number (000)
697
( 614 - 780 )
297
( 228 - 367 )
223
( 177 - 268 )
177
(160 - 182 )
Number of
Housing Units
in Sample
97
43
24
30
Note: Numbers in parentheses are 95% confidence intervals for the «aimat»H patents and numbers.
                                        B-3

-------
                           TABLE B-4

GEOMETRIC MEAN PAINT LEAD LOADINGS IN PUBLIC HOUSING UNITS
       BUILT BEFORE 1980, BY SELECTED CHARACTERISTICS
Characteristic
Total Occupied Housing
Units Built Before 1980
Construction Yean



1960-1979
1950-1959
Before 1950
Census Region




Northeast
Midwest
South
West
Interior Surfaces
(mg/sq. cm.)
02

02
02
03

02
02
02
02
( 02

( 0.1
( 0.1
( 02

( 0.1
( 0.1
( 0.1
( 0.0
, 03 )

. 03 )
, 0.4 )
, 0.5 )

, 0.4 )
, 0.8 )
, 03 )
. 0.8 )
Exterior Surfaces
(me/so, cm.)
0.2

0.1
0.3
12

02
0.0
02
03
{ 0.1

( 0.0
( 0.0
( 0.6

( 0.0
( 0.0
( 0.1
( 0.1
, 0.4 )

. 02 )
, 23 )
, 2.4 )

, 13 )
, 0.8 )
, 0.4 )
0.8 )
 Note: Numbers in parentheses are 95% confidence intervals for the respective geometric means.
                               B-4

-------
                                         TABLE B-5

        GEOMETRIC MEAN PAINT LEAD LOADINGS IN PUBLIC HOUSING UNITS
 BUILT BEFORE 1980, BY ARCHITECTURAL COMPONENT AND CONSTRUCTION YEAR
Component/Construction year
WaHs/ceilings/floor



Metal a)



Non-metal (2)



Other (3)




1960-1979
1950-1959
Before 1950

1960-1979
1950-1959
Before 1950

1960-1979
1950-1959
Before 1950

1960-1979
1950-1959
Before 1950
Interior Surfaces
(mg/sq. cm.)

0.1
0.1
0.2

0.03
0.8
0.4

0.2
0.2
0.3

0.04
0.04
0.1

( 0.1
( 0.1
( 0.1

( 0.0
( 0.4
( 0.1

( 0.1
( 0.1
( 0.2

( 0.0
( 0.0
( 0.0

0.3 )
0.3 )
0.5 )

0.3 )
1.8 )
1.8 )

0.4 )
0.5 )
0.6 )

0.1 )
0.2 )
0.2 )
Exterior Surfaces
(mg/sq. cm.)

0.1
0.9
0.6

0.001
0.2
0.9

0.1
0.9
3.3

0.1
0.3
0.6

( 0.1
( 0.9
( 0.2

( 0.0
( 0.0
( 0.4

( 0.1
( 0.2
( 1.5

( 0.0
( 0.1
( 0.1

0.5 )
0.9 )
1.4 )

0.0 )
6.3 )
2.3 )

0.2 )
5.7 )
7.2 )

0.8 )
1.0 )
3.2 )
Note: Numbers in parentheses are 95% confidence intervals for the respective arithmetric means.

Interior:
 (1) Includes metal trim, window sills, molding, doors, air/heat vents, and radiators.
 (2) Includes non-metal trim, window sills, mokfig, doors, and air/heat vents.
 (3) Includes shelves, cabinets, fireplace, and closets, on any substrate.

Exterior:
 (1) Includes only metal windows, doors, soffit and facia, columns, and railings.
 (2) Includes non-metal windows, doors, soffit and facia, columns, and railings.
 (3) Includes porches, balconies, stairs, etc., on any substrate.
                                             B-5

-------
      APPENDIX C

Dust and Soil Samples Excluded
  From Data Quality Analysis

-------

-------
              TABLE C-l

DUST SAMPLES EXCLUDED FROM ANALYSIS
              (Continued)
               C-2

-------
                                   TABLECM

                    DUST SAMPLES EXCLUDED FROM ANALYSIS
                                   (Continued)
NT
                                                          til
 1 WET.WK..1 !•»••
    o.o0f.no* i • t
 i «or_no .
                                      C-3

-------
                   TABLE C=2

SOIL SAMPLES WITH GREATER THAN 2,6000 PPM OF LEAD,
          DROPPED FROM ANALYSIS FILE
LABCODE
M
C
C
M
M
M
C
C
M
C
M
M
M
LBP ID
1010909
1831106
1831304
1820802
1820802
1011709
950402
2441509
1010503
3011905
1011501
1011501
1010503
PB AOJ
1630
3530
3530
1890
2580
2500
5900
6006
4040
8610
6260
10900
22000
SAMP WGT
0.5465
1.0007
1.0007
0.5054
0.5370
0.5026
0.9965
1.0006
0.5650
1.0397
0.5079
0.5131
0.5102
SD ID
81
82
82
82
81
81
81
83
81
82
81
82
82
f 'T^n
2.983
3.528
3.528
3.740
4.804
4.975
5.921
6.002
7,150
8,281
12.325
21.243
43.120
                     C-4

-------
      APPENDIX D

XRF Validation Data and the
    Distribution of the
  Adjusted XRF Readings

-------
                                       FIGURE D-l

             BASELINE VALIDATION MEASUREMENT BY SHIM LEAD CONCENTRATION
                       FOR XRF INSTRUMENT #34 ON STEEL SUBSTRATE
3.5  T
                                The curvature  is not statistically significant
                                H	1	1	1	1
             0.5        1        1.5       2        2.5

                     Shim lead concentration (mg/sq cm)
3.5
                                                                             o

                                                                             •
              Measurement

              Outlier

              Regression line

-------
                                    FIGURE D-2

           BASELINE VALIDATION MEASUREMENT BY SHIM LEAD CONCENTRATION
                     FOR XRF INSTRUMENT #35 ON STEEL SUBSTRATE
3 T
           0.5       1        1.5       2       2.5

                   Shim lead concentration (mg/sq cm)
                                                                        °    Measurement
                                                                        •    Outlier
                                                                             Regression line
3.5

-------
                                   FIGURE D-3

            BASELINE VALIDATION MEASUREMENT BY SHIM LEAD CONCENTRATION
                     FOR XRF INSTRUMENT #37 ON STEEL SUBSTRATE
3.5 T
 3 +
2.5 +
l.t
1.5 +
0.5 +
              CD
                             •f
-f
                                               -f-
            0.5       1       1.5       2       2.5

                   Shim lead concentration (mg/sq cm)
                                                     o
                         3.5
                                                                      O


                                                                      •
                                     Measurement

                                     Outlier

                                     Regression line

-------
                                      FIGURE D-4

            CLOSEOUT VALIDATION MEASUREMENT BY SHIM LEAD CONCENTRATION
                      FOR XRF INSTRUMENT #34 ON STEEL SUBSTRATE
 4 T
0.5
                                                                            o
              Measurement

              Outlier

              Regression line
                                The  curvature Is  not statistically significant
            0.5        1        1.5        2       2.5

                     Shim lead  concentration (mg/sq cm)
3.5

-------
                                                 FIGURE D-5
                       CLOSEOUT VALIDATION MEASUREMENT BY SHIM LEAD CONCENTRATION
                                 FOR XRF INSTRUMENT #35 ON STEEL SUBSTRATE
             3 T
S
                        0.5       1        1.5       2       2.5

                                Shim lead concentration (mg/sq cm)
        °    Measurement

        •    Outlier
                                                                                          Regression line
3.5

-------
                                      FIGURE D-6

           CLOSEOUT VALIDATION MEASUREMENT BY SHIM LEAD CONCENTRATION
                      FOR XRF INSTRUMENT #37 ON STEEL SUBSTRATE
3.5  T
                                                                             °    Measurement

                                                                             •    Outlier

                                                                            •——— Regression line
                                The  curvature Is  not statistically significant
                                H	1	1	1	
            0.5       1        1.5        2        2.5

                     Shim lead  concentration  (mgtaq cm)
3.5

-------
                                                FIGURE D-7

                       CLOSEOUT VALIDATION MEASUREMENT BY SHIM LEAD CONCENTRATION
                                 FOR XRF INSTRUMENT #38 ON STEEL SUBSTRATE
           3.5 T
g
                       0.5       1       1.5        2        2.5

                                Shim lead concentration  (mg/eq cm)
3.5
        °    Measurement

        •    Outlier

        ——— Regression line

-------
                                          FIGURE D 8
                   VALIDATION MEASUREMENTS BY SHIM LEAD CONCENTRATION
                              FOR XRF INSTRUMENT #34 ON STEEL
Lead concentration (mg/sq cm) = -0.9472 + 1.0748'Measurement + 0.00106*(Days since 2/1/90)
   4.5 T
                                                      (D
i
                       1       1.5       2       2.5      3
                       X a Lead Concentration (mg/aq cm)
                                                                              °
Measurement
Outlier (not  used)
Calibration Equation
X = 1, projected to V
Y = X
                                                                        When X = 1, Y = approximately 1.79

-------
                                             FIGURE D-9
                      VALIDATION MEASUREMENTS BY SHIM LEAD CONCENTRATION
                                  FOR XRF INSTRUMENT #35 ON STEEL

Lead concentration (mg/sq cm) = -2.3895 + 1.9829'Measurement + 0.00196*(Days  since 2/1/90)

   6 T                                               •
 H
 >•
                       1       1.5      2       2.5      3

                       X a Lead Concentration (mg/sq cm)
                                                                               °    Measurement

                                                                               •    Outlier (not used)

                                                                              	Calibration Equation

                                                                              	X = 1, projected to V
                                                                                    Y = X
                                                                         When X = 1, Y = approximately  1.1

-------
9
k-l
o
                                                     FIGURE D-10

                              VALIDATION MEASUREMENTS BY SHIM LEAD CONCENTRATION

                                          FOR XRF INSTRUMENT #36 ON STEEL


        Lead concentration (mg/sq cm)  = -1.0265 + 1.0580*Measurement +  0.00105'(Days since 2/1/90)


           4.5  T
                       0.5      1       1.5      2      2.5      3


                                X s Lead Concentration (mg/sq cm)
3.5
                 °     Measurement



                 •     Outlier (not  used)



                 	Calibration Equation



                 	X = 1, projected to V



                       Y = X
                                                                                 When X = 1, Y = approximately 1.90

-------
                                       FIGURE D-1I
                VALIDATION MEASUREMENTS BY SHIM LEAD CONCENTRATION
                            FOR XRF INSTRUMENT #37 ON STEEL

Lead concentration (mg/sq cm)  =  -0.9274 + 1.2351'Measurement  + 0.00122*(Days  since 2/1/90)

    4 -r
              0.5
1       1.5      2      2.5       3

X  = Lead  Concentration (mg/sq cm)
3.5
                                                                               o

                                                                               •
                                                              Measurement

                                                              Outlier (not used)

                                                              Calibration  Equation

                                                              X = 1, projected to Y

                                                              Y = X
                                                                         When X = 1, Y = approximately 1.54

-------
                                                    FIGURE D-12
9
»-»
ro
                             VALIDATION MEASUREMENTS BY SHIM LEAD CONCENTRATION
                                         FOR XRF INSTRUMENT #38 ON STEEL

        Lead concentration (mg/sq cm) = -2.8854  + 2.0049'Measurement + 0.00198*(Days since 2/1/90)
            4 T
                      0.5      1       1.5       2       2.5      3

                               X = Lead Concentration (mg/sq cm)
3.5
                 °     Measurement

                 •     Outlier (not used)

              	Calibration  Equation

              	X = 1, projected to Y

                       Y = X


           When X = 1, Y = approximately 1.92

-------
                                            FIGURE 0-13

                     VALIDATION MEASUREMENTS BY SHIM LEAD CONCENTRATION
                                 FOR XRF INSTRUMENT #39 ON STEEL

Lead concentration (mg/sq cm) = -0.9832  +  1.1274'Measurement  + 0.00112*(Days S|nce 2/1/90)

   4.5 T

                                                      (D
              0.5      1      1.5      2      2.5       3

                        X = Lead Concentration (mg/sq cm)
3.5
                                                                               0     Measurement

                                                                               •     Outlier (not used)

                                                                                     Calibration  Equation

                                                                               	X = 1, projected to Y

                                                                                     Y = X
                                                                         When X = 1, Y = approximately  1.73

-------
                                             FIGURE D-14
                      VALIDATION MEASUREMENTS BY SHIM LEAD CONCENTRATION
                                 FOR XRF INSTRUMENT #41 ON STEEL

Lead concentration (mg/sq cm) = -0.3167  + 1.3345'Measurement  +  0.00132*(Days since  2/1/90)

   3.5 T
    3 ••
   2.5
 it
>•
    rt
   1.5   -
    1   r
   0.5
              0.5
   1.5      2      2.5      3

Lead Concentration (mg/sq cm)
3.5
                                                    °    Measurement

                                                    •    Outlier  (not used)

                                                   	Calibration Equation

                                                    	X = 1, projected to V

                                                         Y = X
                                                                         When X = 1, Y = approximately  0.95

-------
                                             FIGURE D-15
                      VALIDATION MEASUREMENTS BY SHIM LEAD CONCENTRATION
                                  FOR XRF INSTRUMENT #32 ON WOOD

Lead concentration (mg/sq cm) = 0.5773 + 0.6748*Measurement + 0.00067'(Days since 2/1/90)
    5 T
                                                                                     Measurement

                                                                                     Outlier (not  used)

                                                                                     Calibration Equation
                                                                                                      i
                                                                                     X = 1, projected to Y

                                                                                     Y = X
                                                                         When X = 1, Y = approximately 0.59
                                                                         Median used lor 0.6 shim
                        X s Lead Concentration (mg/sq cm)

-------
o\
                                                      FIGURE D-16

                                VALIDATION MEASUREMENT BY SHIM LEAD CONCENTRATION
                                           FOR XRF INSTRUMENT #34 ON WOOD

         Lead  concentration  (mg/sq cm) = 0.5108 +  0.7767*Measurement  + 0.00077*(Days  since 2/1/90)


                                                               8
                                                               o
                                                                                        o
                                                      Measurement


                                                      Outlier (not  used)


                                                      Calibration Equation


                                                      X = 1, projected to Y

                                                      Y = X
                                 I	   .        .
                                 pgggD  . I  . ... |  .
                        I  i i  t i i  i i  i i  I  When X = 1, Y = approximately 0.61
1.5
2.5
3.5
                                                                               4 Median used for 0.6 shim
                                 X s Lead Concentration (mgfoq cm)

-------
                                             FIGURE D-I7
                      VALIDATION MEASUREMENTS BY SHIM LEAD CONCENTRATION
                                  FOR XRF INSTRUMENT #35 ON WOOD

Lead concentration (mg/sq cm) = 0.5776 + 0.6842'Measurement  + 0.00068'(Days since  2/1/90)
    5 T
    4 -
 ii
>
                                                                                     Measurement

                                                                                     Outlier (not used)

                                                                                     Calibration  Equation

                                                                                     X = 1, projected to Y
                                                                         When X = 1, Y = approximately  0.59

                                                                         Median used lor 0.6 shim
   . t  •*-
                        X n Lead Concentration (mg/sq cm)

-------
                                                     FIGURE D-18
9
ss
                              VALIDATION MEASUREMENTS BY SHIM LEAD CONCENTRATION

                                          FOR XRF INSTRUMENT #36 ON WOOD


        Lead concentration (mg/sq cm)  = 0.5779 + 0.6977'Measurement + 0.00069*(Days since 2/1/90)
            5 T
            4 •
 °    Measurement


 •    Outlier  (not used)


	Calibration  Equation


 	X s 1, projected to Y


      Y = X
                                                                                  When X = 1, Y = approximately 0.59

                                                                                  Median used (or 0.6 shim
             1  -1-
                                X a Lead Concentration (mg/sq cm)

-------
                                            FIGURE D-19
                      VALIDATION MEASUREMENTS BY SHIM LEAD CONCENTRATION
                                 FOR XRF INSTRUMENT #37 ON WOOD
Lead concentration (mg/sq cm) = 0.2390 + 0.8879'Measurement + 0.0008B*(Days  since 2/1/90)
                                                                              °    Measurement
                                                                              •    Outlier (not  used)
                                                                             	Calibration Equation
                                                                             	X = 1, projected to Y
                                                                                   Y = X
                                                                        When X = 1, Y = approximately 0.83
                        X • Lead Concentration (mg/sq cm)

-------
                                            FIGURE D-20
                      VALIDATION MEASUREMENTS BY SHIM LEAD CONCENTRATION
                                 FOR XRF INSTRUMENT #38 ON WOOD

Lead concentration (mg/sq cm) = 0.5649 + 0.708rMeasurement + 0.00070*(Days  since 2/1/90)

    5 T
1
 °    Measurement

 •    Outlier (not  used)

	Calibration Equation
                       i
	X = 1, projected to Y

      Y = X
                                                                        When X = 1, Y = approximately  0.56
                                                                        Median used lor 0.6 shim
                       X a Lead Concentration (mg/sq cm)

-------
                                             FIGURE D-21
                      VALIDATION MEASUREMENTS BY SHIM LEAD CONCENTRATION
                                  FOR XRF INSTRUMENT #39 ON WOOD

Lead concentration (mg/sq cm)  =  0.4214 + 0.7530'Measurement + 0.00075*(Days since 2/1/90)

   4.5 T
                                                                               o

                                                                               •
Measurement

Outlier (not  used)

Calibration Equation
 •               ;
X = 1, projected to Y

Y = X
                                                         , ,  , .  )  t .  , ,  j  Wnen x = 1i Y = approximately 0.74
                                                                         Median used lor 0.6 shim
                        X n Lead Concentration (mg/sq cm)

-------
                                            FIGURE D-22
                      VALIDATION MEASUREMENTS BY SHIM LEAD CONCENTRATION
                                 FOR XRF INSTRUMENT #41 ON WOOD

Lead concentration (mg/sq cm) = 0.0933 + 1.4646'Measurement  + 0.00145*(Days since 2/1/90)

   3.5 T
                                                                               o    Measurement

                                                                               •    Outlier  (not used)

                                                                              	Calibration Equation
                                                                                                     i
                                                                              	X = 1, projected to Y
                                                                                    Y = X
                                                                         When X = 1, Y = approximately 0.58
   -0.5 -1-
                        X a Lead Concentration (mg/sq cm)

-------
                                             FIGURE D-23
                      VALIDATION MEASUREMENTS BY SHIM LEAD CONCENTRATION
                                FOR XRF INSTRUMENT #32 ON CONCRETE
Lead concentration (mg/sq cm) = 0.5440 4-  1.6648'Measurement + 0.00165*(Days since 2/1/90)
   3.5 T
     1  -
    0.5 •-
                                                                              o
                                                                              •
Measurement
Outlier (not  used)
Calibration Equation
X = 1, projected to Y
Y = X
                                                                        When X = 1, Y = approximately 0.24
                                                                        Median used lor 0.6 shim
   -0.5 -1-
                        X B Lead Concentration (mg/sq cm)

-------
                                            FIGURE D-24
                      VALIDATION MEASUREMENTS BY SHIM LEAD CONCENTRATION
                               FOR XRF INSTRUMENT #34 ON CONCRETE
Lead concentration (mg/sq cm) = 0.5529 + 1.4138'XRF  reading + 0.00205*(Days since 2/1/90)
   3.5  T
    3  --
   2.5  - •
   1.5 -•
   0.5 --
 °    Measurement
 •    Outlier (not  used)
	Calibration Equation
      Y = X
                                                                        Calibration equation based on readings
                                                                        on concrete from other XRF Instruments
  -0.5 -1-
                       X = Lead Concentration (mg/sq cm)

-------
                                            FIGURE D-25
                      VALIDATION MEASUREMENTS BY SHIM LEAD CONCENTRATION
                               FOR XRF INSTRUMENT #35 ON CONCRETE

Lead concentration (mg/sq cm) = 0.5435  +  1.7305*Measuremenl + 0.00171*(Days since 2/1/90)

   4.5 T
                                                                              °    Measurement

                                                                              •    Outlier (not used)

                                                                             	Calibration  Equation
                                                                                                     i
                                                                             	X = 1, projected to Y
                                                                        When X = 1, Y = approximately  0.24
                                                                        Median used lor 0.6 shim
   -0.5 -1-
                        X a Lead Concentration (mg/sq cm)

-------
                                             FIGURE D-26
                      VALIDATION MEASUREMENTS BY SHIM LEAD CONCENTRATION
                                FOR XRF INSTRUMENT #36 ON CONCRETE

Lead concentration (mg/sq cm) = 0.5692 +  1.5546'Measurement + 0.00154*(Days since 2/1/90)

   3.5 T
                                              2.5
3.5
                                                                              °     Measurement

                                                                              •     Outlier (not used)

                                                                             	Calibration  Equation
                                                                                    X =  1, projected to Y

                                                                                    Y = X
   When X = 1, Y = approximately  0.28
j  Median used for 0.6 shim

4
                        X » Lead Concentration (mg/sq cm)

-------
                                              FIGURE D-27
                       VALIDATION MEASUREMENTS BY SHIM LEAD CONCENTRATION
                                FOR XRF INSTRUMENT #31 ON CONCRETE

Lead concentration (mg/sq cm) = 0.5475 + 2.2547*Measurement  + 0.00223*(Days since 2/1/90)

   3.5 T
     3 •-
    2.5

i    «
    1.5  -
     1  -
    0.5   -
                                              2.5
                                                      (SD
3.5
                 °     Measurement

                 •     Outlier (not used)

                	Calibration  Equation
                                        i
                	X =  1, projected to Y
   When X = 1, Y = approximately 0.18
i   Median used lor 0.6 shim

4
                        X a Lead Concentration (mg/aq cm)

-------
                                             FIGURE D-28
                       VALIDATION MEASUREMENTS BY SHIM LEAD CONCENTRATION
                                FOR XRF INSTRUMENT #38 ON CONCRETE

Lead concentration (mg/sq cm) = 0.5753  +  1.1586*Measurement •«• 0.00115'(Days since 2/1/90)

   3.5 -r                                             o
                                                                              °    Measurement

                                                                              •    Outlier (not used)

                                                                              	Calibration Equation
                                                                                                     i
                                                                              	X = 1, projected to Y

                                                                                    Y = X
                                                                        When X = 1, Y = approximately 0.34
                                                                        Median used for 0.6 shim
     1 -1-
                        X » Lead Concentration (mg/aq cm)

-------
                                             FIGURE D-29
g
                       VALIDATION MEASUREMENTS BY SHIM LEAD CONCENTRATION
                                FOR XRF INSTRUMENT #39 ON CONCRETE

Lead concentration (mg/sq cm) = 0.5515 + 1.4860'Measurement + 0.00147*(Days since 2/1/90)

   3.5 -r
                                                                                    Measurement

                                                                                    Outlier (not used)

                                                                                    Calibration  Equation

                                                                                    X = 1, projected to Y
                                                                        When X = 1, Y = approximately  0.27

                                                                        Median used for 0.6 shim
   •0.5 -1-
                       X a Lead Concentration (mg/sq cm)

-------
                                             FIGURE D-30
                       VALIDATION MEASUREMENTS BY SHIM LEAD CONCENTRATION
                                FOR XRF INSTRUMENT #41 ON CONCRETE
Lead concentrallon (mg/sq cm) = 0.5835  +  0.8673'Measurement  + 0.00086*(Days since 2/1/90)
n
>•
                                                                              o
                                                                              •
Measurement
Outlier (not used)
Calibration Equation
X = 1, projected to Y
Y = X
                                                                        When X = 1, Y = approximately  0.47
                                                                        Median used for 0.6 shim
    -1 J-
                        X a Lead Concentration (mg/sq cm)

-------
                                                      FIGURE D-31
S
                               VALIDATION MEASUREMENTS BY SHIM LEAD CONCENTRATION
                                         FOR XRF INSTRUMENT #32 ON DRYWALL

        Lead concentration (mg/sq cm) « 0.56B4 + 0.93B9*Measurement + 0.00093*(Days since 2/1/90)

            3.5 T
             3  --
            2.5  ••
        2
         n
 °    Measurement

 •    Outlier (not  used)

	Calibration Equation

	X = 1, projected to Y

      Y = X
                                                                                 When X = 1, Y = approximately 0.43

                                                                      3-5      4  Median used lor 0.6 shim
                                X = Lead Concentration (mg/sq cm)

-------
                                             FIGURE D-32
o
K>
                      VALIDATION MEASUREMENTS BY SHIM LEAD CONCENTRATION
                                FOR XRF INSTRUMENT #34 ON DRYWALL

Lead concentration (mg/sq cm) = 0.5715 4- 0.8213'Measurement + 0.00081'(Days since  2/1/90)

     4 T
                               1.5
                                                                                       °
                                                                                    Measurement

                                                                                    Outlier (not used)

                                                                                    Calibration  Equation

                                                                                    X = 1 , projected to Y
                                                                                    Y = X
                                               I  i i  i i  I i  i i  i I i  i i  i I   When X = 1, Y = approximately 0.50
                                                      2.5
3.5
4 Median used lor 0.6 shim
                        X = Lead Concentration (mg/sq cm)

-------
                                              FIGURE 0-33
w
                       VALIDATION MEASUREMENTS BY SHIM LEAD CONCENTRATION

                                 FOR XRF INSTRUMENT #35 ON DRYWALL


Lead concentration (mg/sq cm) = 0.5730 -i- 0.6014*Measurement + 0.00079*(Days  since 2/1/90)


   4.5 T
                                              2.5
                                                                               0    Measurement



                                                                               •    Outlier (not used)



                                                                              	Calibration  Equation

                                                                                                      g

                                                                              	X = 1, projected to Y



                                                                                    Y = X
                                                                i ,  ,.,  , i  , , ,  , i   When X = 1, Y = approximately 0.51

                                                                                 Median used for 0.6 shim
                                                                3      3.5      4
                        X = Lead Concentration (mg/sq cm)

-------
                                             FIGURE D-34
                      VALIDATION MEASUREMENTS BY SHIM LEAD CONCENTRATION
                                FOR XRF INSTRUMENT #36 ON DRYWALL

Lead concentration  (mg/sq cm) = 0.1694 + 0.8220'Measurement + 0.00081'(Days since  2/1/90)

   4.5 T
                                                                              o    Measurement

                                                                              •    Outlier (not  used)

                                                                             	Calibration Equation

                                                                             	X = 1, projected to Y

                                                                                   Y = X
                                                                        When X = 1, Y = approximately 0.
   •0.5 -1-
                        X e Lead Concentration (mg/sq cm)

-------
                                           FIGURE D-35
g
                     VALIDATION MEASUREMENTS BY SHIM LEAD CONCENTRATION
                               FOR XRF INSTRUMENT #37 ON URYWALL

Lead concentration (mg/sq cm) = 0.2311 + 0.9186'Measuremenl 4- 0.00091'(0ays since  2/1/90)
     2 -•
   1.5  -
/ F
<\
\
>i i i i i 1 i i i r 1 i i-i-i l-ri n-l-r-r-n-l -r i i r 1
1 1.5 2 2.5 3 3.5 4
^ Measurement
• Outlier (not used)
Calibration Equation
i
= 1, projected to Y
hen X = 1, Y = approximately 0.81
                       X e Lead Concentration (mg/sq cm)

-------
                                                      FIGURE D-36
O
k
o\
                               VALIDATION MEASUREMENTS BY SHIM LEAD CONCENTRATION
                                         FOR XRF INSTRUMENT #38 ON DRYWALL

         Lead concentration (mg/sq cm) = 0.5839  + 0.7548'Measurement + 0.00075'(Days since 2/1/90)

             4  T
            3.5  -
        i
         n
        >•
                                                                                       o
                       Measurement

                       Outlier (not used)

                       Calibration Equation
                                        i
                       X a 1, projected to Y

                       Y = X
                                                                       I  i i  i i I  When X = 1, Y = approximately 0.53
                                                       2.5
3.5
4 Median used lor 0.6 shim
                                X  a Lead Concentration (mg/sq cm)

-------
                                             FIGURE D-37
                       VALIDATION MEASUREMENTS BY SHIM LEAD CONCENTRATION
                                 FOR XRF INSTRUMENT #39 ON DRYWALL
Lead concentration (mg/sq cm) = 0.3389 + 0.7943'Measurement + 0.00079*(Days since 2/1/90)
     4 -r
    3.5 +
1
    2.5 +
"    Measurement
•    Outlier (not  used)
	Calibration Equation
	X = 1, projected to Y
      Y = X
                                                                        When X = 1, Y = approximately 0.80
                                                                        Median used for 0.6 shim
                        X = Lead Concentration (mg/sq cm)

-------
                                                     FIGURE D-38
2
00
                               VALIDATION MEASUREMENTS BY SHIM LEAD CONCENTRATION
                                         FOR XRF INSTRUMENT #41 ON DRYWALL

        Lead concentration (mg/sq cm) = -0.3153 + 1.4959'Measurement  +  0.00148'(Days since 2/1/90)

           3.5  T
             *1 _ .
           2.5 -•
           1.5 ••
           0.5 ••
                       0.5
1       1.5       2       2.5      3

X  s Lead Concentration (mg/sq cm)
3.5
                                                       °    Measurement

                                                       •    Outlier (not  used)

                                                     	Calibration Equation

                                                     	X = 1, projected to Y

                                                             Y = X


                                                 When X  = 1, Y = approximately 0.84

-------
                            FIGURE D-39



              DISTRIBUTION OF SIMULATED RE CALD3RATED

                MEASUREMENTS FOR WOOD SUBSTRATES
   8.0 -r
  7.0 --
?6.0 +


O
o-
0
  5.0 --
c
o


£4.0 --
3
•o 3.0 •+-
o
3

E


5 2.0 -f
  1.0 --
  0.0
                                           B-"
                                B"
            •¥
H	h
     -0.5  0.0   0.5   1.0    1.5   2.0   2.5   3.0   3.5   4.0   4.5   5.0


                     Actual lead concentration  (mg/sq cm)
          Percentiles, 1%. 5%. 10%. 25%.

          50%, 75%, 90%. 95%. 99%
                                            Actual = Simulated
                                D-39

-------
                              FIGURE D-40



               DISTRIBUTION OF SIMULATED RECALIBRATED

                 MEASUREMENTS FOR STEEL SUBSTRATES
   8.0  -r
   7.0  --
f 6.0
B


f 5.0
c
e>

E

£  4.0
s
•>
a
o>
•o  3.0
0)
   2.0 --
   1.0 --
   0.0
      •0.5   0.0   0.5   1.0   1.5    2.0   2.5   3.0   3.5   4.0


                      Actual lead  concentration (mg/sq cm)
                 4.5   5.0
          Percentiles. 1%. 5%. 10%. 25%.

          50%, 75%. 90%. 95%. 99%
Actual = Simulated
                                 D-40

-------
                             FIGURE D-41

              DISTRIBUTION OF SIMULATED RECALIBRATED
               MEASUREMENTS FOR DRYWALL SUBSTRATES
   8.0 -r
   7.0 --

-------
                               FIGURE D-42


                DISTRIBUTION OF SIMULATED RECALIBRATED
                MEASUREMENTS FOR CONCRETE SUBSTRATES
                      (Assumptions based on censored data)
   8.0  -r
   7.0  --
   6.0  -'
o>
E 5.0  --
o
E
£4.0  --
a
   3.0 --
E
35  2.0 H-
  0.0
      -0.5   0.0    0.5   1.0   1.5   2.0    2.5   3.0   3.5   4.0    4.5   5.0

                      Actual lead concentration (mg/sq cm)
          Percentiles. 1%. 5%. 10%. 25%,

          50%. 75%. 90%. 95%. 99%
Actual = Simulated
                                 D-42

-------
                              FIGURE D-43


               DISTRIBUTION OF SIMULATED RECALIBRATED
               MEASUREMENTS FOR CONCRETE SUBSTRATES
                (Assumptions ignore data from 0.6 shim, slope = 1)
   B.o  -r
   7.0  --
   6.0 --
tr


I 5.0 -f-
E
£4.0 --
   3.0 --
3


35  2.0 -h
   1.0 --
   0.0
f-
H	h
      -0.5   0.0   0.5   1.0   1.5    2.0   2.5   3.0   3.5    4.0   4.5

                      Actual lead  concentration (mg/sq cm)
                                                   5.0
          Percentiles, 1%. 5%. 10%. 25%.
          50%. 75%. 90%. 95%. 99%
                            Actual = Simulated
                                  D-43

-------