November 28,  1995
               DRAFT REPORT

STATISTICAL EVALUATION OF THE RELATIONSHIP
 BETWEEN BLOOD-LEAD AND DUST-LEAD BASED ON
DATA FROM THE ROCHESTER LEAD-IN-DUST STUDY

                    for

                 Task 4-13
           Battalia Task Leader
              Warren Strauss
            Battelle Task Team
        Bruce Buxton, Steven Rust,
     Halsey Boyd, and Claire Matthews
                 BATTELLE
              505 King Avenue
           Columbus, Ohio  43201
          Contract No.  68-D2-0139
Janet Reamers, EPA Work Assignment Manager
     Jill Hacker, EPA Project Officer
        Technical  Programs Branch
       Chemical Management Division
Office  of  Pollution  Prevention and Toxics
   U.S.  Environmental Protection Agency
           Washington, DC  20460

-------
                  Battelle Disclaimer

    This is a report of research performed by Battelle
for the United States Government.   Because of the
uncertainties inherent in experimental or research
work,  the above parties assume no responsibility or
liability for any consequences of use, misuse,
inability to use, or reliance upon the information
contained herein, beyond any express obligations
embodied in the governing written agreement between
Battelle and the United States Government.
                           ii

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

2.0 DATA PREPARATION	   1

3.0 THE STATISTICAL MODEL   	   3

4.0 STATISTICAL MODELING RESULTS  	   4

5.0 PROTECTIVE DUST LEAD LEVELS   	   6

6.0 EXCEEDANCE PROPORTIONS  	   6

7.0 PARTIAL SIGNIFICANCE OF WINDOW WELL PB MEASUREMENTS   .  .  20


                           APPENDICES

APPENDIX A.  STATISTICAL MODELS 	   A-l

APPENDIX B.  ALTERNATIVE REGRESSION PARAMETER
             ESTIMATION IN THE PRESENCE OF
             MEASUREMENT ERROR  	 B-l

APPENDIX C.  TOLERANCE BOUNDS  AND  CONFIDENCE
             INTERVALS FOR PERCENTILES  AND
             EXCEEDANCE PROBABILITIES IN A
             REGRESSION SETTING 	 C-l

APPENDIX D.  ESTIMATION OF MEASUREMENT  ERROR
             VARIANCE COMPONENTS  	 D-l

APPENDIX E.  PARAMETER ESTIMATES FOR STATISTICAL
             MODELS	E-l

APPENDIX F.  PROTECTIVE DUST LEAD  LEVELS AND
             EXCEEDANCE PROBABILITIES FOR  ERRORS IN
             VARIABLES SOLUTION 	 F-l

APPENDIX G.  PROTECTIVE DUST LEAD  LEVELS AND
             EXCEEDANCE PROBABILITIES FOR  LEAST
             SQUARES SOLUTION   	 G-l
APPENDIX H.
PLOTS COMPARING THE ERRORS IN VARIABLES
MODEL RESULTS FOR BRM AND WIPE SAMPLING   .  .  .  . H-l
                               ill

-------
                        TABLE OF CONTENTS
                            (Continued)


                          LIST OF TABLES
Page
Table 1.  Results of Fitting the Statistical Model to
          the Rochester Lead-in-Dust Data Using the
          Errors in Variables Approach  	   5
Table 2.  Estimated Dust Lead Levels for Floors, Window
          Sills, and Window Wells at Which the 85th,
          90th, 95th, and 99th Percentiles of Childhood
          Blood Lead Concentrations Reach 10, 15, and
          20 /ig/dL (Based on the Errors in Variables
          Approach)	16
Table 3.  Estimated Proportion of Children with Blood-
          Lead Concentrations Greater than 10, 15, and
          20 /xg/dL as a Function of Floor, Window Sill,
          or Window Well Lead Loadings (Based on the
          Errors in Variables Approach) 	  18
Table 4.  Results of Fitting Floor, window Sill, and
          Window Well Lead Loadings or Concentrations
          Simultaneously on Blood-Lead Concentrations
          Based on the Least-Squares Approach	21
Table 5.  Estimated Pearson Correlation Coefficients Between
          Natural Log Transformed Lead Levels from Children's
          Blood and Floor, Window Sill and Window Well Dust  .  23


                         LIST OF FIGURES

Figure 1. Rochester Lead-in-Dust Study Floor-Lead
          Loadings From BRM Sampling -- Estimated
          Regression Curve and Tolerance Bounds 	 7
Figure 2. Rochester Lead-in-Dust Study Window Sill-Lead
          Loadings From BRM Sampling -- Estimated
          Regression Curve and Tolerance Bounds 	 8
Figure 3. Rochester Lead-in-Dust Study Window Well-Lead
          Loadings From BRM Sampling -- Estimated
          Regression Curve and Tolerance Bounds 	 9
Figure 4. Rochester Lead-in-Dust Study Floor-Lead
          Loadings From Wipe Sampling -- Estimated
          Regression Curve and Tolerance Bounds 	  10
Figure 5. Rochester Lead-in-Dust Study Window Sill-Lead
          Loadings From Wipe Sampling -- Estimated
          Regression Curve and Tolerance Bounds 	  11
Figure 6. Rochester Lead-in-Dust Study Window Well-Lead
          Loadings From Wipe Sampling -- Estimated
          Regression Curve and Tolerance Bounds 	  12
Figure 7. Rochester Lead-in-Dust Study Floor-Lead
          Concentrations From BRM Sampling -- Estimated
          Regression Curve and Tolerance Bounds 	  13
                                IV

-------
                        TABLE OF CONTENTS
                           (Continued)                       Page

Figure 8. Rochester Lead-in-Dust Study Window Sill-Lead
          Concentrations From BRM Sampling -- Estimated
          Regression Curve and Tolerance Bounds 	  14

Figure 9. Rochester Lead-in-Dust Study Window Well-Lead
          Concentrations From BRM Sampling -- Estimated
          Regression Curve and Tolerance Bounds 	  15

-------
        STATISTICAL  EVALUATION OF THE RELATIONSHIP BETWEEN
              BLOOD-LEAD AND DUST-LEAD BASED ON DATA
              FROM THE ROCHESTER LEAD-IN-DUST STUDY
1.0 INTRODUCTION
    The following statistical analysis investigated the
relationship between children's blood-lead concentrations and
levels of lead in interior household dust from data collected in
the Rochester Lead-in-Dust Study.  This research was conducted to
support regulatory decisions for Section 403 of Title X being
made by EPA's Office of Pollution Prevention and Toxics.
Specifically, this research was designed to give information on
the levels of interior dust lead found on floors, window sills or
window wells that would result in 85%, 90%, 95% and 99% of the
distribution of childhood blood-lead concentrations being below
10, 15, and 20 fig/dL.  Additionally this analysis investigated
the importance of levels of lead in'window well dust as a
predictor of blood-lead concentrations after taking into account
the levels of lead in floor and window sill dust.
    This analysis follows a similar approach to  that taken for
an analysis of data from the Repair and Maintenance (R&M) Study,
and the results are presented in the same format so that
comparisons between the two studies can be made.

2.0 DATA PREPARATION
    The sample consisted of 205 homes with one child per home.
The response variable in the statistical analysis was the natural
logarithm of child blood-lead concentration measured in units of
ln(/ig/dL) .  Predictor variables included lead loading and
concentration results observed in dust from floors, window sills
and window wells, and the date of blood sampling which was used
to account for time trends in blood-lead concentrations.  In 204
homes, blood samples and dust samples were all collected within a
three-week window of time.  For one home the dust sampling date
was not available.

-------
     Individual  samples  of  dust  lead were  collected from multiple
locations of a  given component  type  using both the  Baltimore  R&M
 (BRM) vacuum sampling method and wipe sampling.  The samples  were
often collected from locations  with  different surface areas.  For
example, homes  with uncarpeted  floors had between  1 and  5 floor
BRM dust samples with total area ranging from 1 to  5 ft2 sampled
per house.  Each house  also had 1 to 3 window sill  BRM samples
with total area ranging from 0.06 to 1.7 ft2 sampled per house.
In addition, each house had between  1 and 3 window  well  BRM
samples with total area ranging from 0.05 to 1.2 ft2 sampled per
house.  Since the area  and mass of each individual  sample varied
within a house, area weighted average lead loading  and mass
weighted average lead concentration  results were calculated for
floors, window  sills and window wells.  Thus, if two dust samples
were collected  from floor locations  within a house with  sample
areas of 1 ft2 and 3 ft2, the  lead  loading results  from  the  3  ft2
sample were weighted by a factor of  3 when calculating the area
weighted averages.  The natural log  of these weighted average
lead loading and concentration  results were used as predictor
variables in the statistical analyses, and therefore the
estimated relationships between blood lead and dust lead
correspond to dust lead averages and not individual 'hot spots'.
    Floor samples in the Rochester Lead-in-Dust Study were
collected from both carpeted and uncarpeted surfaces,  while floor
samples in the R&M Study were collected from only uncarpeted
surfaces.  Therefore,  to maintain comparability analysis of the
Rochester data was restricted to the data for uncarpeted
surfaces.  From the original sample of 205 Rochester homes,  there
were a total of 193 homes with BRM floor-dust samples collected
from uncarpeted floor surfaces.
    The median  blood-lead level from all 205 homes was 6.1 /xg/dL
(range 1.4 to 31.7 pig/dL) .   The median BRM floor dust-lead
loading from the 193 homes with uncarpeted floors was 13.2 jig/ft2
(range 0.1 to 74,100 jig/ft2) .  The  median  window sill  BRM dust-
lead loading from 197 homes was 265 jig/ft2 (range 0.7  to 118,000

-------
/xg/ft2) .   The median window well BRM dust-lead loading from 189
homes was 48,000 /xg/ft2 (range 7 to 3,000,000 /xg/ft2) .

3.0 THE STATISTICAL MODEL
    A log-linear statistical model was used which expresses
blood-lead concentrations as a function of environmental lead
levels.  Specifically, the model contains a single intercept, a
single slope relating In(blood lead) to In(dust lead) and a time
effect which adjusts for temporal trends in childhood blood-lead
concentrations.  Since the dates of environmental sampling ranged
from August 31, 1993 to November 20, 1993 in the Rochester Lead-
in-Dust Study, there was not a large enough range of time points
to properly parameterize the sine-wave model for seasonal
variations in blood-lead that was used in the analysis of the
Repair and Maintenance Study data.  Therefore, a simple linear
trend was fitted to capture the seasonal variations in the
Rochester data between August 31 and November 20, 1993.  Details
concerning the mathematical form of the statistical model can be
found in Appendix A.
    Due to the fact that environmental lead levels are usually
measured with error, both a simple least-squares approach (which
does not account for measurement error) and a statistical
approach that adjusts for measurement error in predictor
variables were used while fitting the statistical model.  Details
concerning the statistical adjustment for errors in predictor
variables can be found in Appendix B.
    The model used makes a simplifying assumption about the
Rochester Lead-in-Dust Study data.  In particular, this analysis
does not attempt to account for the possible effect of
potentially important socioeconomic and behavioral factors.
Investigators in the Rochester Lead-in-Dust Study found that four
covariates were "significantly associated with higher blood lead
levels among children:  Black race, parental reports that

-------
children put  soil in their mouths, single parent  household, and a
higher  ferritin level".1

4.0  STATISTICAL MODELING RESULTS
     The results of fitting the statistical  model  to the data
using an errors in variables approach are reported in Table 1.
Separate models were fitted for the nine environmental-lead (PbE)
measurements  -- lead loading by BRM and wipe sampling for
uncarpeted  floors,  window sills,  and window wells,  as well as
lead concentration by BRM sampling for uncarpeted floors,  window
sills,  and  window wells.  Slope (/i^)  parameter estimates  and
associated  95%  confidence intervals are reported,  as  well as a
measure of  the  proportion of variability explained by the model
(R2)  from a 'least  squares'  fit of the model.
     The relationship between blood-lead concentrations and dust
lead loadings for floors,  window sills, and window wells  are
illustrated graphically in Figures 1 through 9.   The  fitted
regression  curve from the least-squares fit is plotted using a
finely  dashed line,  and the solution from the errors  in variables
model is plotted using a solid line.  The four upper  dashed
curves  in Figures 1 to 9 represent upper 95% tolerance bounds for
the 85th, 90th,  95th,  and 99th percentiles  of the distribution of
children's  blood-lead concentration as a function of  dust-lead
loadings.   The  line type employing the shortest dash  corresponds
to the  85th percentile,  the next  shortest corresponds to  the 90th
percentile, and so on.   The estimated regression  curves and
associated  tolerance bounds were  calculated for children's blood-
lead levels measured near the median sampling date (October 13,
    Department of Pediatrics, Biostatistics, and Environmental Medicine, The
University of Rochester School of Medicine,  (June, 1995),  "The Relation of
Lead-Contaminated  House Dust and Blood Lead Levels Among Urban Children, Final
Report, Volume II, Results and Discussion",  U.S. Department of Housing and
Urban Development  Grant No. MLDP T0001-93.

-------
Table 1.   Results of Fitting the Statistical Model to the Rochester Lead-in-Dust Data
          Using the Errors in Variables Approach
Unit of Measure
Component
Tested
Slope Values
ft
Slope Estimate
95%
Confidence Interval
R2
(a)
Slope Values in Units of ln(0g Pb / dL Blood) / ln(//g Pb / ft2 Sampled) ( BRM Sampler)
Loading
(//g/ft2)
Floors
Window Sills
Window Wells
0.133
0.126
0.149
(0.089,0.177)
(0.087,0.165)
(0.115,0.183)
0.139
0.139
0.160
Slope Values in Units of ln(//g Pb / dL Blood) / ln(//g Pb / ft2 Sampled) (Wipe Samples)
Loading
(//g/ft2)
Floors
Window Sills
Window Wells
0.237
0.198
0.179
(0.155 ,0.319)
(0.133,0.264)
(0.130,0.229)
0.111
0.131
0.108
Slope Values in Units of ln(//g Pb / dL Blood) / ln(//g Pb / g Dust) (BRM Sampler)
Concentration
U/g/g)
Floors
Window Sills
Window Wells
0.104
0.126
0.111
(0.033 , 0.175)
(0.077 ,0.175)
(0.062 ,0.161)
0.033
0.083
0.066
     (a)   The reported R2 values are based on a least squares fit without adjusting for
         the errors in predictor variables.
     (b)   Based on the results of this simple descriptive model, the predicted blood-lead
         concentration for children living in houses with BRM floor lead loadings of 100 and
         200 //g.ft2 would be 7.6 and 8.3 //g/dL respectively for a difference of 0.7 //g/dL.
1993).   Methods for  calculating the  tolerance bounds are detailed
in Appendix  C.   The  tolerance bounds depicted in these figures
represent a  95% upper confidence bound for  the 85th,  90th,  95th
and  99th percentiles of the  distribution of children's blood-lead
concentrations at each dust-lead level, based on the regression
model  results.   Thus,  the highest curve in  Figure  1  corresponds
to a 95% upper confidence bound on the 99th percentile of
children as  a function of BRM floor-lead loading.

-------
5.0  PROTECTIVE DUST LEAD LEVELS
     The dashed curves in Figures 1 to 9 can be used to determine
the levels of interior dust lead that would result in 85%,  90%,
95% and 99% of the distribution of childhood blood-lead
concentrations being below target values with 95% confidence.
This is accomplished by drawing a horizontal line at the blood-
lead level of interest, and then drawing a vertical line at the
point of intersection with the appropriate tolerance bound  curve.
     Table 2 reports such dust-lead loading levels for floors,
window sills and wells based on target blood-lead levels of 10,
15, and 20 /zg/dL.  The results in this table are based on
tolerance bounds calculated using the errors in variables model
at or near the median sampling date (October 13, 1993).
        In  some  cases,  the  tolerance bound  is  higher  than  the
blood-lead level of interest over the entire range of plausible
dust-lead levels.  For example in Figure 1, the tolerance bound
for the 99th percentile of children's blood-lead concentrations
is always above a blood-lead concentration of 10 iig/dL.  In Table
2 these cases have associated dust-lead values that are listed as
"Out of Range".

6.0  EXCEEDANCE PROPORTIONS
        Table  3  provides  estimates of  the proportion  of  children
with blood-lead concentrations exceeding 10, 15, and 20 itg/dL at
various targeted dust-lead loading values for floors, window
sills,  and window wells.   These exceedance proportions,  and
associated 95% confidence intervals were calculated using methods
detailed in Appendix C.  They are based on the errors in
variables model near the median sampling date (October 13, 1993).
        The results  from  this analysis suggest that for  BRM  dust-
lead loadings of 100 /zg/ft2 on  floors,  500  iig/ft2 on  window
sills,  and 800 /ig/ft2 on  window wells,  approximately  31%,  19%,
and 2% of the children sampled in this study would be expected to
have blood-lead concentrations that exceed 10 itg/dL at the median
sampling date (October 13,  1993).

-------
           50:



           40-



      73   30

       U>

      A   20-
       O
      _0
      CD
                                  #•*•  •
                                 *  *
                           10
      100          1000        10000       100000

Floor Pb Loading — BRM (yug/sq. ft)

     *   *   * ObservaHons
     	 Pr«dlct«d (EIV)
     	 Pr»dlci«d (OLS)
     	85% Upper Bound (EIV)
     	90% Upper Bound (EIV)
     	95% Upper Bound (EIV)
     	99% Upper Bound (EIV)
1000000
Figure 1. Rochester Lead-in-Dust Study Floor-Lead Loadings From BRM Sampling -- Estimated Regression Curve
         and Tolerance Bounds for the 85th, 90th, 95th, and 99th Percentiles of Children's Blood-Lead
         Concentrations Based on Errors in Variables Fit.

-------
oo
               50




               40^




          =U   30

          D>


          -Q   201
          O
          O
          m   10
                               10          100         1000         10000      100000

                                  Window Sill Pb Loading — BRM (M9/sq. ft)

                                          *   *   * Observations
                                          	 Pr»dlcf«d (EIV)
                                          	Predicted (OLS)
                                                    85% Upper Bound (EIV)
                                          	90% Upper Bound (EIV)
                                          	95* Upper Bound (EIV)
                                          	99% Upper Bound (EIV)
1000000
    Figure 2. Rochester Lead-in-Dust Study Window Sill-Lead Loadings From BRM Sampling - Estimated Regression
            Curve and Tolerance Bounds for the 85th, 90th, 95th, and 99th Percentiles of Children's Blood-Lead
            Concentrations Based on Errors in Variables Fit

-------
           50 -I



           40



      =0   30]

       D>
-Q
Q_
           20
       O
       O
      m   10
                           10           100         1000        10000       100000

                              Window Well Pb Loading  — BRM (/xg/sq. ft)

                                       *   *   * Observations
                                       	 Predicted (EIV)
                                       	Predicted (OLS)
                                                 85% Uppsr Bound (EIV)
                                                                                    1000000
                                         	90% Upper Bound (EIV)
                                         	95% Upper Bound (EIV)
                                         	99% Upper Bound (EIV)
Figure 3. Rochester Lead-in-Dust Study Window Well-Lead Loadings From BRM Sampling  - Estimated
         Regression Curve and Tolerance Bounds for the 85th, 90th, 95th, and 99th Percentiles of Children's
         Blood-Lead Concentrations Based on Errors in Variables Fit.

-------
M
O

          Q_

          ~O
           O
          £
          DO
               50 H
               40-
               30-
               20-
10-
                0-
                               10
                            100         1000        10000       100000

                      Floor Pb Loading — Wipe (^g/sq. ft)

                            *  *   *  Observations
                           	 Predicted (EIV)
                                      Predicted (OLS)
1000000
                                          	85% Upper Bound (EIV)
                                          	90% Upper Bound (EIV)
                                          	95% Upper Bound (EIV)
                                          	99% Upper Bound (EIV)
    Figure 4.  Rochester Lead-in-Dust Study Floor-Lead Loadings From Wipe Sampling - Estimated Regression Curve
             and Tolerance Bounds for the 85th, 90th, 95th, and 99th Percentiles of Children's Blood-Lead
             Concentrations Based on Errors in Variables Fit

-------
           50 H
           4CH
      TJ   30-
      -Q   20
      Q_
      TJ
       O
       O
      m   10
                           10          100         1000        10000       100000

                               Window Sill Pb Loading — Wipe (//g/sq. ft)

                                        *   *   *  Observations
                                       	 Predicted (EIV)
                                       	 Pr»dlcUd (OLS)
                                                  85% Upper Bound (EIV)
1000000
                                       	90% Upper Bound (EIV)
                                       	95% Upper Bound (EIV)
                                       	99% Upper Bound (EIV)
Figure 5. Rochester Lead-in-Dust Study Window Sill-Lead Loadings From Wipe Sampling - Estimated Regression
         Curve and Tolerance Bounds for the 85th, 90th, 95th, and 99th Percentiles of Children's Blood-Lead
         Concentrations Based on Errors in Variables Fit

-------
               50 -I
to
                               10          100         1000        10000       100000

                                 Window Well Pb Loading — Wipe (/^g/sq. ft)

                                           *   *  *  Observations
                                          	 Pradletcd (EIV)
                                          	Predicted (OLS)
                                                     85% Upper Bound (EIV)
1000000
                                          	90% Upper Bound (EIV)
                                          	95% Upper Bound (EIV)
                                          	99% Upper Bound (EIV)
    Figure 6.  Rochester Lead-in-Dust Study Window Well-Lead Loadings From Wipe Sampling — Estimated
             Regression Curve and Tolerance Bounds for the 85th, 90th, 95th, and 99th Percentiles of Children's
             Blood-Lead Concentrations Based on Errors in Variables Fit

-------
H
OJ
          _Q
          Q-

          -o
           O
          £

          CD
               50 -I
               40
               30
20
10
    	*	

~    	*	3k~ ~* . .  *
                               10
                             100          1000        10000

                      Floor Pb Concentration — BRM
                                                           100000     1000000
                                           *   *
                                                     Observations
                                           	 Predicted (EIV)
                                           	Predicted (OLS)
                                           	85% Upper Bound (EIV)
                                           	90% Upper Bound (EIV)
                                           	95% Upper Bound (EIV)
                                           	99% Upper Bound (EIV)
    Figure 7. Rochester Lead-in-Dust Study Floor-Lead Concentrations  From BRM Sampling - Estimated  Regression
             Curve and Tolerance Bounds for the 85th, 90th, 95th, and  99th Percentiles of Children's Blood-Lead
             Concentrations  Based on Errors in Variables Fit

-------
           50:




           40




      TJ   30

       0)


      -Q   20
       o
       o
      m   10
            0
                                                                       *%* *  *
                           10          100         1000        10000       100000

                             Window Sill Pb Concentration — BRM (y^g/g)

                                       *   *   * Observations
                                      	 Predicted (EIV)
                                                Predicted (OLS)
                                      	85% Upper Bound (EIV)
                                      	90% Upper Bound (EIV)
                                      	95% Upper Bound (EIV)
                                      	99% Upper Bound (EIV)
1000000
Figure 8. Rochester Lead-in-Dust Study Window Sill-Lead Concentrations From BRM Sampling - Estimated
         Regression Curve and Tolerance Bounds for the 85th, 90th, 95th, and 99th Percentiles of Children's
         Blood-Lead Concentrations Based on Errors in Variables Fit

-------
           50 4
           40-
           30-
           20
 O)
3

-Q
Q_
-o
 O
 O
m   10-
            0-
                           10           100         1000        10000      100000

                             Window Well Pb Concentration — BRM (/xg/g)

                                       *   *   *  Observations
                                      	 Predicted (EIV)
                                      	Predicted (OLS)
                                                 85% Upper Bound (EIV)
                                                                                   1000000
                                      	90% Upper Bound (EIV)
                                      	95% Upper Bound (EIV)
                                      	99% Upper Bound (EIV)
Figure 9. Rochester Lead-in-Dust Study Window Well-Lead Concentrations From BRM Sampling - Estimated
         Regression Curve and Tolerance Bounds for the 85th, 90th, 95th, and 99th Percentiles of Children's
         Blood-Lead Concentrations Based on Errors in Variables Fit

-------
Table 2.  Estimated Dust Lead Levels for Floors. Window Sills, and Window Wells at
         Which the 85th, 90th, 95th, and 99th Percentiles of Childhood Blood Lead
         Concentrations Reach 10, 15, and 20/ig/dL (Based on the Errors in Variables
         Approach)
                            BRM Sampler Lead Loading
Sample
Type
Floor
Pb Loading
(0g/ft2)
BRM Sampler
Window Sill
Pb Loading
Oag/ft2!
BRM Sampler
Window Well
Pb Loading
(/ug/ft2)
BRM Sampler
Tolerance
Level
0.85
0.90
0.95
0.99
0.85
0.90
0.95
0.99
0.85
0.90
0.95
0.99
Target Blood-Lead Concentration
lOjig/dL
4
1
Out of Range
Out of Range
89
23
2
Out of Range
10,156
3,779
817
41
15/tg/dL
88
32
5
Out of Range
2,209
770
133
2
144,436
61.592
15,940
982
20/ig/dL
553
223
52
1
15,477
5,938
1,307
45
805,938
362,045
104,728
8,045
Wipe Sampling Lead Loading
Floor
Pb Loading
Gug/ft2>
Wipe Samples
Window Sill
Pb Loading
fog/ft2)
Wipe Samples
Window Well
Pb Loading
fog/ft2)
Wipe Samples
0.85
0.90
0.95
0.99
0.85
0.90
0.95
0.99
0.85
0.90
0.95
0.99
7
3
Out of Range
Out of Range
74
30
7
Out of Range
1,799
732
179
11
40
23
8
Out of Range
578
294
93
7
16,621
7,989
2,420
190
112
67
30
4
1.965
1,068
405
45
66,800
33,992
11,785
1208
                                       16

-------
Table 2.  Continued
                              BRM Sampler Lead Concentration
Sample
Type
Floor
Pb Concentration
(pg/g)
BRM Sampler
Window Sill
Pb Concentration
0/g/g)
BRM Sampler
Window Well
Pb Concentration
(pg/g)
BRM Sampler
Tolerance
Level
0.85
0.90
0.95
0.99
0.85
0.90
0.95
0.99
0.85
0.90
0.95
0.99
Target Blood-Lead Concentration
10/ig/dL
14
Out of Range
Out of Range
Out of Range
593
125
11
Out of Range
1,104
162
8
Out of Range
15j*g/dL
2.450
685
23
Out of Range
16,270
5,586
830
10
49,044
14,463
1,488
6
20j/g/dL
16,581
6,088
1,106
1
104,638
40,827
8,974
228
378,000
131,306
23,482
280
         A result of 'Out of Range' indicates that the tolerance bound for Blood-Pb is always above the
           target level.

           Results are based on a time adjusted analysis held fixed at October 13, which was the median
           sampling date.
                                              17

-------
Table 3.  Estimated Proportion of Children with Blood-Lead Concentrations Greater than
         10, 15, and 20 fjg/dl as a Function of Floor, Window Sill, or Window Well Lead
         Loadings (Based on the Errors in Variables Approach)

                           BRM Sampler Lead Loading
Surface
Tested
Floors
Window
Sills
Window
Wells
Pb
Loading
«J/*t2
5
10
15
20
25
30
35
40
100
200
250
50
100
200
300
400
500
600
700
200
500
750
800
1500
3000
5000
10000
20000
Proportion Greater Than
10/rg/dl
Pr(Pb>10)
0.11
0.15
0.17
0.19
0.20
0.21
0.22
0.23
0.31
0.37
0.39
0.08
0.11
0.14
0.16
0.18
0.19
0.20
0.21
0.00
001
0.02
0.02
0.03
0.05
0.07
0.10
0.15
95% Cl
(0.07,0.16)
(0.10,0.20)
(0.12 , 0.22)
(0.14,0.24)
(0.15,0.26)
(0.16 , 0.27)
(0.17 , 0.29)
(0.18,0.30)
(0.24 , 0.38)
(0.28 , 0.46)
(0.29 , 0.49)
(0.05,0.13)
(0.07,0.16)
(0.10,0.19)
(0.12,0.22)
(0.13,0.23)
(0.14,0.25)
(0.15,0.26)
(0.16,0.27)
(0.00 , 0.02)
(0.00 , 0.04)
(0.01 , 0.05)
(0.01 , 0.05)
(0.01 , 0.07)
(0.03 , 0.09)
(0.04,0.12)
(0.07,0.15)
(0.10,0.20)

IB
Pr(Pb>15)
0.02
0.04
0.04
0.05
0.06
0.06
0.07
0.07
0.11
0.14
0.16
0.01
0.02
0.03
0.04
0.05
0.05
0.06
0.06
0.00
0.00
0.00
0.00
0.00
0.00
0.01
0.02
0.03
Greater Than
pg/dl
95% Cl
(0.01 . 0.05)
(0.02 . 0.06)
(0.02 , 0.07)
(0.03 , 0.08)
(0.03 , 0.09)
(0.04.0.10)
(0.04,0.10)
(0.04,0.11)
(0.07,0.16)
(0.09,0.21)
(0.10,0.23)
(0.00 , 0.03)
(0.01 , 0.04)
(0.02 , 0.06)
(0.02 , 0.07)
(0.03 , 0.08)
(0.03 , 0.09)
(0.03 , 0.09)
(0.04,0.10)
(0.00 , 0.00)
(0.00 . 0.00)
(0.00 , 0.00)
(0.00,0.01)
(0.00, 0.01)
(0.00 , 0.02)
(0.00 , 0.02)
(0.01 , 0.04)
(0.02 , 0.06)
Proportion Greater Than
20//g/dt
Pr(Pb>20)
0.00
0.01
0.01
0.01
0.02
0.02
0.02
0.02
0.04
0.06
0.06
0.00
0.00
0.01
0.01
0.01
0.01
0.02
0.02
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
95% Cl
(0.00.0.01)
(0.00 , 0.02)
(0.00 , 0.02)
(0.00 . 0.03)
(0.01 . 0.03)
(0.01 . 0.04)
(0.01 , 0.04)
(0.01 , 0.04)
(0.02 . 0.07)
(0.03,0.10)
(0.03.0.11)
(0.00.0.01)
(0.00,0.01)
(0.00 . 0.02)
(0.00 , 0.02)
(0.00 , 0.03)
(0.00 . 0.03)
(0.01 , 0.03)
(0.01 , 0.04)
(0.00, 0.00)
(0.00 . 0.00)
(0.00 , 0.00)
(0.00 . 0.00)
(0.00 , 0.00)
(0.00 , 0.00)
(0.00 , 0.00)
(0.00,0.01)
(0.00 , 0.02)
                                       18

-------
Table 3. Continued
                                  Wipe Sampling Lead Loading
Surface
Tested
Floors
Window
Sills
Window
Wells
Pb
Loading
wjrtt2
5
10
15
20
25
30
35
40
100
200
250
50
100
200
300
400
500
600
700
200
500
750
800
1500
3000
5000
10000
20000
Proportion Greater Than
10/ig/dl
Pr(Pb>10)
0.08
0.13
0.17
0.20
0.23
0.25
0.27
0.29
0.44
0.55
0.59
0.08
0.12
0.18
0.22
0.25
0.28
0.30
0.32
0.02
0.05
0.06
0.06
009
0.14
018
0.25
0.32
95% Cl
(0.04.0.13)
(0.09,0.18)
(0.12.0.22)
(0.15,0.26)
(0.18,0.29)
(0.20 , 0.32)
(0.21 , 0.34)
(0.23 , 0.37)
(0.32 , 0.55)
(0.40 , 0.69)
(0.42 . 0.73)
(0.04,0.13)
(0.08,0.18)
(0.13, 0.24)
(0.17,0.28)
(0.19,0.32)
(0.21 , 0.35)
(0.23 , 0.38)
(0.24 , 0.40)
(0.00 , 0.05)
(0.02 , 0.09)
(0.03,0.10)
(0.03,0.11)
(0.06 ,0.14)
(0.10,0.19)
(0.13.0.24)
(0.19,0.31)
(0.25 . 0.41)
Proportion Greater Than
15/fg/dJ
Pr(Pb>15)
0.01
0.03
0.04
0.06
0.07
0.08
0.09
0.1
0.19
0.28
0.31
0.01
0.03
0.05
0.07
0.08
0.09
0.11
0.12
0.00
0.00
0.01
0.01
0.02
0.03
0.05
0.07
0.11
95% Cl
(0.00 . 0.03)
(0.01 , 0.06)
(0.02 , 0.07)
(0.03 , 0.09)
(0.04,0.11)
(0.05,0.12)
(0.06,0.14)
(0.06,0.15)
(0.12,0.28)
(0.16,0.42)
(0.18,0.47)
(0.00 , 0.03)
(0.01 , 0.05)
(0.03 , 0.08)
(0.04,0.10)
(0.05,0.12)
(0.06.0.14)
(0.07,0.16)
(0.07,0.17)
(0.00,0.01)
(0.00 . 0.02)
(0.00 , 0.02)
(0.00 , 0.02)
(0.01 , 0.04)
(0.02 , 0.06)
(0.03 , 0.08)
(0.05,0.12)
(0.07,0.17)
Proportion Greater Than
20/ig/dl
Pr(Pb>20)
0.00
0.01
0.01
0.02
0.02
0.03
0.03
0.04
0.08
0.14
0.16
0.00
0.00
0.01
0.02
0.03
0.03
0.04
0.04
0.00
0.00
0.00
0.00
0.00
0.01
0.01
0.02
0.04
95% Cl
(0.00.0.01)
(0.00 , 0.02)
(0.00 , 0.03)
(0.01 , 0.03)
(0.01 , 0.04)
(0.01 , 0.05)
(0.01 , 0.06)
(0.02 , 0.06)
(0.04,0.14)
(0.07 , 0.24)
(0.08 , 0.28)
(0.00,0.01)
(0.00 , 0.02)
(0.00 , 0.03)
(0.01 , 0.04)
(0.01 , 0.05)
(0.01 , 0.06)
(0.02 , 0.07)
(0.02 , 0.08)
(0.00 , 0.00)
(0.00 , 0.00)
(0.00 , 0.00)
(0.00 , 0.00)
(0.00,0.01)
(0.00 , 0.02)
(0.00 , 0.03)
(0.01 , 0.04)
(0.02 , 0.07)
Results are based on a time adjusted analysis held fixed at October 13, which was the median sampling date.
                                                19

-------
7.0  PARTIAL SIGNIFICANCE OF WINDOW WELL PB MEASUREMENTS
        The final analysis  performed  was  used to  determine
whether the predictive ability of the model improves when adding
window well lead levels to a model which already accounts for
dust lead on floors and window sills.  The analysis begins with a
base model which adjusts for temporal variations in blood-lead
concentrations, and then sequentially adds variables representing
lead in floors, window sills, and window wells to the model.  The
results of this analysis are reported in Table 4.  The slope
parameters and associated 95% confidence intervals in each row
correspond to a statistical model which includes the variables in
the base model and the variable being added to the base model.
The coefficient of determination (R2)  is  provided for each  model
fitted, where floors, sills and wells are added sequentially to
the base model.  The difference in R2 values between the full
model and the base model is also presented.  This value
corresponds to the extra amount of variability explained by the
variable being added to the model.
        The estimated effect  of BRM well  lead on  blood  lead  was
only marginally statistically significant after adjusting for the
effects of floor lead, sill lead, and temporal variation.  Floor
lead appeared to explain most of the variability in blood lead
for both BRM and wipe lead loading,  but was less predictive for
BRM lead concentration. The model with only BRM floor lead
loading and an adjustment for temporal trends explained 16.2% of
the variability in blood-lead concentrations, adding window sill
lead loading to the model explained an additional 4.4% of the
variability, and adding window well lead loading with the other
two factors already in the model explained an additional 5.4% of
the variability.  The amount of extra variability explained by
window sill-lead loading after already accounting for floor-dust
lead loading and temporal variations is calculated by subtracting
the coefficient of determination from the base model (R2=0.162)
                                20

-------
     Table 4.     Results of Fitting Floor, Window Sill, and Window Well Lead Loadings or Concentrations Simultaneously
                   on Blood-Lead Concentrations Based on the Least Squares Approach
Unit of Measure
Sample
Size
Variables
in the
Base Model
Variables
Added to the
Base Model
Slope Values
01 (floors)
(95% CD
01 (SUb)
(95% CI)
01 (Wells)
(95% CI)
R2
for the
Full Model
Additional
Variability
Explained
Slope Values in Units of ln(pg Pb / dL Blood) / ln(jig Pb / f? Sampled) (Dust Samples Collected with BRM Sampler)
Loading
fog/ft2)
BRM Sampler
177



Date
Date, Floors
Date, Floors
Date, Floors,
Window Sills
Floors
Window Sills
Window Wells
Window Wells
0 120
(0 077 . 0 162)
0090
(0.044 . 0 135)
0092
(O.OS1 . 0 134)
0.085
(0.041.0.129)

0.063
(0.023,0.103)

0023
(-0.022. 0 068)


0066
(0 039 , 0.094)
0.058
(0.026 . 0.090)
0.162
0.206
0.256
0.260
0.148
0044
0094
0.054 (W)
0.004 (S)
Slope Values in Units of ln(pg Pb / dL Blood) / ln(/tg Pb / ff Sampled) (Dust Samples Collected with Wipe Samples)
Loading
(Mg/ft2)
Wipe Samples
178
Date
Date, Floors
Date, Floors
Date, Floors,
Window Sills
Floors
Window Sills
Window Wells
Window Wells
0.190
(0.112.0.267)
0138
(O.OS4 . 0 222)
0.140
(O.OS7 , 0.223)
0.125
(0.040 , 0.210)
Slope Values in Units of ln(/tg Pb / dL Blood) / ln(/ig Pb / g Dust)
Concentration
(W5/B)
BRM Sampler
175

Date
Date, Floors
Date, Floors
Date, Floors,
Window Sills
Floors
Window Sills
Window Wells
Window Wells
0.070
(0.013 , 0.127)
0.043
(-0.014, 0 100)
0.054
(-0.003.0.111)
0.041
(-0 017, 0.098)

0.105
(0.038,0.172)

0.067
(-0.013,0.147)


0062
(0023.0.101)
0.040
(-0.007 . 0.087)
(Dust Samples Collected with BRM Sampler)

0076
(0.030.0.122)

0.060
(0.006, 0.114)


0.052
(0.013 . 0.091)
0.026
(-0.019.0071)
0.123
0168
0.168
0.181
0.112
0.045
0.045
0.013 (W)
0 013 (S)

0.047
0.102
0.083
0.108
0.032
0.055
0.036
0006 (W)
0.025 (S)
10
          The reported values of the coefficient of determination (R2) are based on a least squares fit of the descriptive model without adjusting for the errors in predictor variables.
    (w)   Represents the addition variability explained by window wells when it is the last variable added to the model.
    (s)    Represents the additional variability explained by window sills when it is the last variable added to the model.

-------
from the coefficient of determination from the full model
(R2=0.206)  and then multiplying by 100.   The model with only
BRM floor lead concentration and an adjustment for seasonal
trends explained 4.7% of the variability in blood-lead
concentrations, adding window sill lead concentration to the
model explained an additional 5.5% of the variability, and
adding window well lead concentration with the other two
factors already in the model explained an additional 0.6% of
the variability.
     The predictive ability of the model did not improve
significantly when measures of window well dust-lead levels
were added to a model which already accounted for dust lead
on floors and window sills.  The predictive ability of a
regression model will not improve significantly when the
variable added to the model is highly correlated with
another predictor variable.  Therefore,  one possible
explanation for the results seen in Table 4 is that dust
lead measurements from floors, window sills, and window
wells within a house are correlated.  Table 5 demonstrates
estimated Pearson correlation coefficients among natural log
transformed lead levels from children's blood and floor,
window sill and window well dust.   These tables demonstrate
that for each measurement method (BRM Loading, Wipe Loading,
BRM Concentration), lead levels from window sills and window
wells are statistically significantly correlated with each
other, and with children's blood-lead concentrations.  In
addition, the correlations between window sills and wells
are consistently the highest concentrations shown in
Table 5.
                             22

-------
Table 5.   Estimated Pearson Correlation Coefficients Between Natural Log
          Transformed Lead Levels from Children's Blood and Floor, Window Sill
          and Window Well Dust


Blood
Floors
Window
Sills
P
p-value
n
P
p-value
n
P
p-value
n

Blood
Floors
Window
Sills
P
p-value
n
P
p-value
n
P
p-value
n

Blood
Floors
Window
Sills
P
p-value
n
P
p-value
n
P
p-value
n
Floors
Window Sills
Window
Wells
Lead Loadings (BRM Sampler)
0.344
0.001
193
1

0.343
< 0.001
197
0.411
< 0.001
189
1
0.374
<0.001
189
0.271
0.001
179
0.558
<0.001
184
Lead Loadings (Wipe Samples)
0.310
<0.001
197
1

0.338
< 0.001
196
0.395
< 0.001
189
1
0.305
< 0.001
189
0.365
<0.001
182
0.627
<0.001
184
Lead Concentration (BRM Sampler)
0.125
0.084
192
1

0.236
< 0.001
199
0.325
0.047
189
1
0.211
0.004
188
0.225
0.003
177
0.552
<0.001
183
                                     23

-------
    APPENDIX A
STATISTICAL MODELS

-------
The Statistical Model
   The statistical model fitted to the data  in the main
report was  descriptive in nature and appears as follows:

                 logiPbB,) = 4, + ^logfPbE,) «• i2(t,-0) + E,                (1)

where PbBj is the blood-lead  level in iig/dL  for the ith
child, PbEi is the environmental-lead level  for the ith home
in /xg/ft2 for loadings and itg/g  for  concentrations, t± is
the day of  the year  on which  the PbBi measurement was taken,
and E± is the random error term associated with PbBi.   Ei  is
assumed to  follow a  normal distribution with mean zero and
standard deviation aError.   PbE can represent either a lead
loading or  lead concentration in floor,  sill, or well dust.
   00' 0i» 02 and ffError are parameters that are estimated
when the model is fitted.   /30 and /^  are the intercept and
slope of the assumed log-linear  relationship between PbB and
PbE.  Since the dates  of  environmental sampling range from
August 31,  1993 through November 20,  1993  there was not
enough of a range of sample dates to properly parameterize a
sine wave model for  seasonal  variations  in blood-lead.
Therefore, a simple  linear trend was fitted  between August
31 and November 20,  where  j32  is the  slope and 9  is the mean
date of sampling  (October  13,  1993).   ffError  characterizes
the variability in blood-lead left unexplained by the model.

Errors in Variables Solution
   Parameter estimates from a least squares  regression model
for variables that are measured  with error are usually
biased towards zero.   Appendix B  provides  details on the
statistical methodology used  to  correct  the  parameter
estimates to reduce  this bias, and Appendix  D details how
measurement error in composite dust  lead measurements was
estimated for use in these adjusted  models.
                             A-l

-------
   Tables El and E2 in Appendix E provide the parameter
estimates and  associated  standard  errors  for the intercept
and slope  (j30 and j^)  from the descriptive model as
estimated by both ordinary  least squares  and the errors  in
variables statistical approach.  The  errors  in  variables
solution assumes that the measurement error  in  dust  samples
is known, and  fixed at the  values  presented  in  Appendix  D.
The ordinary least squares  solution assumes  that there is no
measurement error in the dust lead variables (i.e.
ff Concentration =  ' •
   A comparison of the slope parameter for dust lead (/Sj)
between the  least squares and errors  in variables  solution
demonstrates that for variables that  have a  strong
relationship with blood- lead  (such as floor  wipe lead
loading) , the bias in j8x attributable to measurement error
can be substantial.  However, when the relationship between
the predictor variable and  the response variable is weak
(such as floor- lead concentration) the bias  in /3X
attributable to measurement error is  not as  large.
   To facilitate  a comparison between the  assumed known
measurement error and zero measurement error, all
statistical results are presented in  the Appendices for both
the least squares and the errors in variables solution.

Alternate Statistical Models
   The descriptive model  presented in the  main body of the
report makes a simplifying assumption about  the data from
the Rochester Lead in Dust Study.  The model, as fitted,
does not account for the effects of potentially important
socioeconomic and behavioral factors.  Investigators in the
Rochester Lead-in-Dust Study found that four covariates in
particular were "significantly associated with higher blood
lead levels in children: Black Race,  parental reports that
                             A-2

-------
   children put soil in  their  mouths,  single parent  household,
   and higher  ferritin level".2
      No  attempts  have  been  made  at this time to  investigate
   changes in  the relationship between blood-lead concentration
   and measures of  lead  in dust that  would be  caused by  the
   presence of these important covariates  in the log-linear
   regression  models.
   Department of Pediatrics, Biostatistics, and Environmental Medicine. The University of Rochester School of Medicine, (June,
1995), "The Relation of Lead Contaminated House Dust and Blood Lead Leveies Among Urban Children. Final Report, Volume II,
Results and Discussion", U.S. Department of Housing and Urban Development Grant No. MLDP T0001-93

                                     A-3

-------
                 APPENDIX B

ALTERNATIVE REGRESSION PARAMETER ESTIMATION IN
      THE PRESENCE OF MEASUREMENT ERROR

-------
                           Appendix B
                  Regression Parameter Estimation in the
                     Presence of Measurement Error
     Let

                          Y = Xj8 + e                       (1)

where

     Y  = a nxl  vector containing the n values of  the
          dependent variable;
     X  = a nxp  matrix where each column contains  the n
          values of one independent variable in the
          regression model (in a model with an intercept
          term,  one of the columns would be a column of
          ones);
     jS  = a pxl  vector of regression coefficients; and
     c  = a nxl  vector of random error terms.

In a standard  regression model it is assumed that  X  is a
matrix of fixed  and known constants, 0 is a vector of fixed
and unknown constants,  and e is distributed as MVN(0,a2I)
where MVN(fi,E) represents a multivariate normal distribution
with mean vector \L and covariance matrix Z.  Estimates of
regression parameters for this standard regression model are
obtained as follows:

                        ft  = (X'X)'1 X'Y

               o2  =  (Y'Y -  p i'X'X p ^  / (n-p)            (2)

                   C a v( p  ^ =  a2  (X'X)'1


                             B-l

-------
      In the presence of measurement error,  it is assumed
that


                          Y = U0 + e                      (3)


where
     U  =  a  nxp matrix of fixed but unknown constants
           representing the values of the independent
           variables  if measured without error;

     X  =  U  +  A,  and

     A  =  a  nxp matrix of the  random measurement errors
           associated with each of the observed  values of the
           independent  variables.
Y and  e are as defined  above.   It  is  assumed that A is

distributed as MVN(0,EA) where EA is known and A is

stochastically independent of  e.   Under  this measurement

error model, estimates  of regression  parameters  are obtained
as follows:



                    p   = (X'X - nZ^r1 X'Y


         o2  =  (Y'Y -  0 '(X'X  -  (n-p)EA) p )  / (n-p)      (4)


                  C( » )  =  a2  (X'X -  nEj'1



These estimators are equivalent to those recommended in

Equations  (2.2.11) and  (2.2.12) by Fuller  (Measurement Error
Models. 1987).

     It can be shown that
      (la) The difference between  [(X'X  - n£&) / n] and  [U'U
          / n] converges in probability to zero as n-»co;
                             B-2

-------
      (Ib) The difference between  [(X'X  -  (n-p)E4) /  (n-p)]
          and  [U'U /  (n-p)] converges in probability to  zero
          as n-»oo; and                              •*
      (2)  The difference between  [X'Y / n] and  [U'Y  / n]
          converges in probability to zero as n-»co.
Using these facts, it follows that the differences between
the regression parameter estimates of Equation  (4) and
(Y'Y -
C a ( 0
P
)
(U'U)'1
'U'U
= o2
U'Y
j ) /
(U'U)'1
(n-p)
                                                          (5)
converge in probability to zero as n-*x>.
     Note that the estimators in Equation  (5) are equivalent
to those of Equation  (2) except that X has been replaced  by
U.  Thus, if U were known, the estimators of Equation  (5)
would be used.  However, since U is unknown, the
asymptotically equivalent estimators of Equation  (4) are
used.  For the purpose of making inferences involving  the
unknown parameters j8 and a2,  it is assumed that the
estimators of Equation  (4) have the same distribution  as
those of Equation  (5).
     In this application of obtaining regression parameter
estimates in the presence of measurement error, the
covariate represents a weighted average of several observed
dust-lead levels that are measured with error.  Therefore,
the estimate of measurement error in an individual sample
obtained from Appendix D must be divided by the number of
individual samples that were used to construct the dust-lead
variable being used in the regression model.  Thus each
diagonal entry of EA (^(ii>) becomes  the  estimate  of
                            B-3

-------
measurement error from Appendix D divided by the number of
individual dust samples (of a given component type) that
were collected from the ith house.
                            B-4

-------
               APPENDIX C

    TOLERANCE BOUNDS AND CONFIDENCE
INTERVALS FOR PERCENTILES AND EXCEEDANCE
   PROBABILITIES IN A REGRESSION SETTING

-------
                          Appendix C
      Tolerance Bounds and Confidence Intervals for Percentiles
        and Exceedance Probabilities in a Regression Setting
     Assume the standard regression model  of Equation  (1) of
Appendix B.  Estimates of regression  parameters are obtained
as follows:

                        j?  = (X'X)-1 X'Y                    (1)

                o2  =  (Y'Y  -  0  'X'X i  )  /  (n-p)


A 95% upper tolerance bound on  (l-q)% of the distribution of
Y for values of the independent variables  given by x0 is

                       TO = x0'  p  + k a                     (2)


where

               k - L1/2 to.ss.n.pEft'Ml-q) / L1/2],             (3)

L = x0'  (X'X)'1^,, is the leverage of the vector x0/  and tUiV[5]
is the  orth  percentile of the noncentral t  distribution with
v degrees of freedom and noncentrality parameter 6.
Similarly,  a 95% confidence interval  for the (l-q)th
percentile  of the distribution of Y for values  of the
independent variables given by x0 is  given by

                            (TL,  T0)                         (4)

where
                              C-l

-------
                      TL = X0' J9  + kL a

                      TU = X0' ft  + kw a                    (5)
A 95% confidence  interval for qy =  Prob(Y>y)  is
where qL is the value of q  for which TL=y and q0 is the
value of q for which T0=y.
     Under the measurement  error model of Equation (3) of
Appendix B, Equations  (2) through (6) above still apply.
However, under this  measurement  error model,  values of /?*,
crA, and L should be  calculated as follows:

                     "ft  =  (X'X -  nEJ'1 X'Y

         o2  = (Y'Y -  ft  '(X'X -  (n-p)EA)  ft  )  /  (n-p)

                    L = XQ'  (X'X - nEJ^Xn
                             C-2

-------
          APPENDIX D

ESTIMATION OF MEASUREMENT ERROR
      VARIANCE COMPONENTS

-------
                         APPENDIX D
             Details on Measurement Error Estimation

   The statistical models which adjust for measurement error
in predictor variables  assume  that the variability due to
sampling and chemical analysis of  dust samples is fixed and
known.  Several sources of  data were  considered for
providing information about the variability in dust sample
results due to measurement  error  including information from
the Rochester Lead-in-Dust  Study,  the R&M study,  data from
the Lead Abatement  Effectiveness  Study in Milwaukee,  and
data  from the Comprehensive Abatement Performance Study
(CAPS).  These data represent  side-by-side field duplicate
vacuum dust samples from floors,  and  can  be evaluated using
the following variance  components  model:

                In(Dust^) = ln(fi)  + Pi +  E^

where Dust^  is  the jth  (first or second)  lead-loading or
lead-concentration  result from the ith side-by-side sample,
H is  the geometric  mean of  Dustij  among all side-by-side
pairs, P± is the random  effect associated with the  ith side-
by-side pair, and Eij  is the random within-pair error term
associated with Dustij •   pi  ^s  assumed to  follow a normal
distribution with mean  zero and variance  ff2Between'  an<* Eij is
assumed to follow a normal  distribution with mean zero and
variance a2Error.
   ^Between characterizes the variability between pairs, and
cr2Error characterizes the variability attributed to
measurement error in each source of data.   Table Dl provides
estimates of <72Error found from each source of data.
                             D-l

-------
   Estimates of variability in Side-by-Side dust sample
results  from the Rochester Lead-in-Dust Study were used in
the statistical models which account  for measurement error
in predictor variables.  These values were chosen to promote
internal consistency within the data  being analyzed.
Table 01.    Estimates of Variability In Side-by-Side Dust Sample Results
           Attributable to Measurement Error.
Data Source
Rochester
Lead-in-Dust
Study
Milwaukee
Repair and
Maintenance
CAPS
Sampling
Method
BRM
Wipe
BRM
BRM
Wipe
Component
•type
Floors
Window
Sills
Window
Wells
Floors
Window Sills
Window Wells
Kitchen
Floor
Interior
Entryways
Floors
Measure
Loading
Concentration
Loading
Concentration
Loading
Concentration
Loading
Loading
Loading
Loading
Concentration
Loading
Concentration
Loading
"Error
(In Std Dev)
1.107
1.382
1.494
1.745
2.872
2.201
0.764
0.801
2.383
1.022
1.151
1.279
0.624
0.56
Number of
Fairs
22
22
15
14
14
14
22
16
15
42
12
35
                              D-2

-------
               APPENDIX E




PARAMETER ESTIMATES FOR STATISTICAL MODELS

-------
Table E1.  Results off Fitting Statistical Models to BRM Lead-Loading Data from the
           Rochester Lead-in-Dust Study
Statistical
Approach
Least
Squares
Errors in
Variables
Component
Tested
Floors
Window
Sills
Window
Wells
Floors
Window
Sills
Window
Wells
Parameter Estimates
for Dust
/»o(a)
setfo)
1.540
(0.072)
1.290
(0.111)
1.064
(0.143)
1.471
(0.076)
1.108
(0.123)
0.345
(0.179)
*,*
seOS,)
0.109
(0.020)
0.09S
(0.018)
0.077
(0.014)
0.133
(0.022)
0.126
(0.020)
0.149
(0.017)
Parameter
Estimate for
Time Effect
ti*
setf^
-0.004
(0.002) •
•0.004
(0.002)
-0.004
(0.002)
-0.004
(0.002)
-0.005
(0.002)
-0.005
(0.002)
Estimate of
Error
Variance
Dmjig/dL)]2
0.332
0.332
0.330
0.322
0.316
0.279
       (a)  Intercept values reported in units of ln(pg Pb/dL Blood).
       w  Slope values reported in units of ln(/ig Pb/dL Blood) / ln(pg Pb/fi? sampled).
       (c)  Time effect reported in units of -——^L,..-,-,  •
                       *               date -10/13/93
                                         E-l

-------
Table E2. Results of Fitting Statistical Models to Wipe Lead-Loading Data from the
           Rochester Lead-in-Dust Study
Statistical
Approach
Least
Squares
Errors in
Variables
Component
Tested
Floors
Window
Sills
Window
Wells
Floors
Window
Sills
Window
Wells
Parameter Estimates
for Dust
V«
seOV
1.367
(0.113)
1.014
(0.165)
1.162
(0.159)
1.179
(0.127)
0.797
(0.181)
0.328
(0.218)
*,*
setf,)
0.172
(0.036)
0.157
(0.030)
0.080
(0.018)
0.237
(0.042)
0.198
(0.033)
0.179
(0.025)
Parameter
Estimate for
Time Effect
&
se(ft)
-0.004
(0.002)
-0.004
(0.002)
-0.004
(0.002)
-0.004
(0.002)
-0.004
(0.002)
-0.004
(0.002)
Estimate of
Error
Variance
Dnfog/dL)]2
0.339
0.336
0.349
0.325
0.324
0.306
       W  Intercept values reported in units of ln(pg Pb/dL Blood).

       w  Slope values reported in units of ln(/tg Pb/dL Blood) / ln(pg Pb/ft2 sampled).

       (c)  Time effect reported in units of
Intug/dL)
                                       date -10/13/93
                                         E-2

-------
Table E3. Results of Fitting Statistical Models to BRM Lead-Concentration Data
           from the Rochester Lead-in-Dust Study
Statistical
Approach
Least
Squares
Errors in
Variables
Component
Tested
Floors
Window
Sills
Window
Wells
Floors
Window
Sills
Window
Wells
FlEtrsunctcr Estinmtcs
for Dust
jS8W
seOV
1.493
(0.178)
1.215
(0.167)
1.301
(0.177)
1.201
(0.233)
0.8S4
(0.202)
0.832
(0.232)
*»
setf,)
0.058
(0.027)
0.080
(0.020)
0.060
(0.019)
0.104
(0.036)
0.126
(0.025)
0.111
(0.025)
Parameter
Estimate for
Time Effect
#
setfj)
-0.004
(0.002)
-0.005
(0.002)
-0.004
(0.002)
-0.005
(0.002)
-0.006
(0.002)
-0.005
(0.002)
Estimate of
Error
Variance
WfffOLSf
0.376
0.352
0.370
0.369
0.337
0.353
       (a)  Intercept values reported in units of ln(pg Pb/dL Blood).
       w  Slope values reported in units of ln(pg Pb/dL Blood) / ln(/ig Pb/g Dust).
       (c)  Time effect reported in units of   . .      „. ..„   .
                      ^               date - 10/13/93
                                         E-3

-------
Table E4.  Results of Fitting Floor, Window Sill, and Window Well Lead Loadings
           (BRM Sampler) Simultaneously on Blood-Lead Concentrations for the
           Rochester Lead-in-Dust Study.
Statistical
Approach
Least
Squares
Errors in
Variables
Component
Tested
Floors
Window
Sills
Window
Wells
Floors
Window
Sills
Window
Wells
Parameter Estimates
for Dust
fc{"
sefcV
0.876
(0.148)
-0.226
(0.197)
*«
seOS,)
0.08S
(0.022)
0.023
(0.023)
0.058
(0.016)
0.125
(0.024)
-0.219
(0.043)
0.297
(0.037)
Parameter
Estimate for
Time Effect
*"
seQSj)
-0.005
(0.002)
-0.004
(0.002)
Estimate of
Error
Variance
pnfeg/dL)]2
0.300
0.233
       (a)  Intercept values reported in units of ln(/tg Pb/dL Blood).
       w  Slope values reported in units of ln(/ig Pb/dL Blood) / ln(^g Pb/ft2 sampled).
       (c)  Time effect reported in units of  	—	  .
                                     date -10/13/93
                                       E-4

-------
Table E5.     Results of Fitting Floor, Window Sill, and Window Well Lead Loadings
               (Wipe Samples) Simultaneously on Blood-Lead Concentrations for the
               Rochester Lead-in-Dust Study.
Statistical
Approach
Least
Squares
Errors in
Variables
Component
Tested
Floors
Window
Sills
Window
Wells
Floors
Window
Sills
Window
Wells
Parameter Estimates
for Dust
V
setfg)
0.790
(0.188)
(d)
*i«
setf,)
0.125
(0.043)
0.067
(0.041)
0.040
(0.024)



Parameter
Estimate for
lime Effect
&*>
se<&)
-0.004
(0.002)

Estimate of
Error
Variance
Onfcg/dL)]2
0.330

        (a)  Intercept values reported in units of ln(/xg Pb/dL Blood).

        w  Slope values reported in units of ln(/ig Pb/dL Blood) / ln(pg Pb/fi2 sampled).

        (c)  Time effect reported in units of         «„,„»«.»  •
                                         date - 10/13/93

        (d)  The simultaneous fitting of dust-lead from floors, window sills and window wells was
           conducted on a subset of the data which had non-missing values for all three variables.
           The variability of the observed dust-lead loadings from window sills and window wells in
           this restricted subset of the data was less than the estimate of variability attributed to
           measurement error that was being used as input to the  errors in variables regression
           models.  Attempts to compute the errors in variables solution under these circumstances
           resulted in nonsensical parameter estimates with associated negative variances. Therefore,
           it was inappropriate to provide these parameter estimates for the errors in variables
           solution to the simultaneous fitting of wipe dust-lead loadings from floors, window sills
           and window wells.
                                           E-5

-------
Table E6  Results of Fitting Floor, Window Sill, and Window Well Lead
           Concentrations (BRM Sampler) Simultaneously on Blood-Lead
           Concentrations for the Rochester Lead-in-Dust Study.
Statistical
Approach
Least
Squares
Errors in
Variables
Component
Tested
Floors
Window
Sills
Window
Wells
Floors
Window
Sills
Window
Wells
Parameter Estimates
for Dust
ft*
setf,,)
0.870
(0.244)
0.617
(0.289)
V*
seQy
0.041
(0.030)
0.060
(0.028)
0.026
(0.023)
0.048
(0.043)
0.1S2
(0.080)
-0.032
(0.070)
Parameter
Estimate for
Time Effect
**
setfj)
-0.006
(0.002)
-0.008
(0.002)
Estimate of
Error
Variance
Dmjtg/dL)]2
0.36S
0.352
       (a)  Intercept values reported in units of ln(/ig Pb/dL Blood).
       w  Slope values reported in units of ln(pg Pb/dL Blood) / ln(/tg Pb/g Dust).
       (c)  Time effect reported in units of   . .   ^L^,.,.  .
                     ^              date -10/13/93
                                       E-6

-------
                APPENDIX F

 PROTECTIVE DUST LEAD LEVELS AND EXCEEDANCE
PROBABILITIES FOR ERRORS IN VARIABLES SOLUTION

-------
Table F1.  Estimated Dust Pb Loadings for Floors, Window Sills, and Window Wells
          at Which the 85th, 90th, 95th and 99th Percentiles of Childhood Blood
          Pb Concentrations Reach 10, 15, and 20 A/g/dL (Based on the Errors in
          Variables Regression Models of the Rochester Lead-in-Dust Study Data)
Sample
-type
Floor
Pb Loading
(Mg/ft2)
BRM Sampler
Window Sill
Pb Loading
Gig/ft2)
BRM Sampler
Window Well
Pb Loading
(/*g/ft2)
BRM Sampler
Tolerance
Level
0.85
0.90
0.95
0.99
0.85
0.90
0.95
0.99
0.85
0.90
0.95
0.99
Target Blood-Lead Concentration
10 jtg/dL
4
1
Out of Range
Out of Range
89
23
2
Out of Range
10,156
3,779
817
41
15/tg/dL
88
32
5
Out of Range
2,209
770
133
2
144,436
61,592
15,940
982
20/tg/dL
553
223
52
1
15,477
5,938
1,307
45
805,938
362,045
104,728
8,045
Floor
Pb Loading
Gtg/ft2)
Wipe Samples
Window Sill
Pb Loading
0*g/ft2)
Wipe Samples
Window Well
Pb Loading
(/tg/ft2)
Wipe Samples
0.85
0.90
0.95
0.99
0.85
0.90
0.95
0.99
0.85
0.90
0.95
0.99
7
3
Out of Range
Out of Range
74
30
7
Out of Range
1,799
732
179
11
40
23
8
Out of Range
578
294
93
7
16,621
7,989
2,420
190
112
67
30
4
1,965
1.068
405
45
66,800
33,992
11,785
1,208
                                    F-l

-------
Table F2. Estimated Dust Pb Concentrations for Floors, Window Sills, and Window
         Wells at Which the 85th, 90th, 95th and 99th Percentiles of Childhood
         Blood Pb Concentrations Reach 10, 15, and 20//g/dL (Based on the
         Errors in Variables Regression Models of the Rochester Lead-in-Dust
         Study Data)
Sample
Type
Floor
Pb Concentration
G
-------
Table F3.  Estimated Proportion of Children with Blood Pb Concentrations Greater
          than 10, 15, and 20^g/dL as Predicted by Errors in Variables Regression
          Models of Blood Pb versus BRM Lead Loadings from Floors. Window Sills
          and Wells
Surface
Tested
Floors
Window
Sills
Window
Wells
Pb
Loading
Mg/ft*
5
10
IS
20
25
30
35
40
100
200
250
50
100
200
300
400
500
600
700
200
500
750
800
1500
3000
5000
10000
20000
Proportion Greater Than
lOfig/di
Pr(Pb>10)
0.11
0.15
0.17
0.19
0.20
0.21
0.22
0.23
0.31
0.37
0.39
008
Oil
0.14
0.16
0.18
0.19
0.20
021
0.00
0.01
0.02
0.02
003
005
0.07
010
015
95% CI
(0.07,0.16)
(0 10 . 0.20)
(0.12,0.22)
(0.14,0.24)
(0.15 , 0.26)
(0.16,0.27)
(0.17,0.29)
(0.18,0.30)
(0.24 . 0.38)
(0.28 , 0.46)
(0.29 , 0 49)
(0.05 , 0 13)
(007.0.16)
(0.10.0.19)
(0.12,0.22)
(0.13 . 0.23)
(0.14,0.25)
(0 15 , 0.26)
(0.16,0.27)
(0 00 , 0.02)
(0 00 . 0 04)
(0.01 , 0.05)
(0.01 , 0.05)
(0.01 , 0 07)
(0.03 , 0.09)
(0.04.0.12)
(0.07,0.15)
(0.10,0.20)
Proportion Grcnter Th&n
15pg/dl
Pr(Pb>15)
0.02
0.04
0.04
0.05
0.06
006
0.07
0.07
0.11
0.14
016
0.01
002
0.03
0.04
0.05
005
006
0.06
0.00
0.00
000
000
000
0.00
0.01
0.02
0.03
95% CI
(0.01 , 0.05)
(0.02 . 0.06)
(0.02 . 0.07)
(0 03 , 0.08)
(0.03 , 0.09)
(0.04 , 0.10)
(004,0.10)
(0.04,0.11)
(007,0.16)
(0 09 , 0.21)
(0.10 , 0.23)
(0.00 , 0.03)
(0.01 , 0.04)
(0.02 , 0.06)
(0 02 , 0.07)
(0.03 , 0.08)
(0.03 , 0.09)
(0.03 , 0.09)
(0.04 , 0.10)
(0.00 , 0.00)
(0 00 , 0 00)
(0 00 , 0.00)
(0.00 , 0.01)
(0.00 , 0.01)
(0.00 , 0.02)
(0.00 , 0.02)
(0.01 , 0.04)
(0.02 . 0.06)
Proportion Greater Than
20pg/dl
Pr(Pb>20)
0.00
0.01
0.01
0.01
0.02
0.02
0.02
0.02
0.04
0.06
006
0.00
0.00
0.01
0.01
0.01
001
0.02
002
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
95% CI
(0.00 , 0.01)
(0.00 , 0.02)
(0.00 , 0.02)
(0.00 . 0 03)
(0.01 , 0.03)
(0.01 . 0.04)
(0.01 . 0.04)
(0.01 , 0.04)
(0.02 , 0.07)
(0.03 , 0.10)
(0.03,0.11)
(0.00.001)
(0.00 , 0.01)
(0 00 , 0.02)
(0.00 , 0.02)
(0.00 , 0.03)
(0.00 , 0.03)
(0.01 , 0.03)
(0.01 , 0.04)
(0.00, 0.00)
(0.00 , 0.00)
(0.00 , 0.00)
(0.00 . 0.00)
(0.00 . 0.00)
(0.00 , 0.00)
(0 00 , 0.00)
(0.00 , 0 01)
(0.00 . 0.02)
                                    F-3

-------
Table F4.  Estimated Proportion of Children with Blood Pb Concentrations Greater
          than 10,  15, and 20 //g/dL as Predicted by Errors in Variables Regression
          Models of Blood Pb versus Wipe Lead Loadings from Floors, Window Sills
          and Wells
Surface
Tested
Floors
Window
Sills
Window
Ufollc








Pb
Loading
Mgffl2
5
10
IS
20
25
30
35
40
100
200
250
50
100
200
300
400
500
600
700
200
500
750
800
1500
3000
5000
10000
20000

10
Pr(Pb>10)
0.08
0.13
0.17
0.20
0.23
0.25
0.27
0.29
0.44
0.55
0.59
0.08
0.12
018
0.22
0.25
0.28
0.30
032
0.02
0.05
006
0.06
0.09
0.14
018
0.25
032
Greater Than
Kg/dl
959, Cl
(0.04,0.13)
(0.09 , 0.18)
(0.12.0.22)
(0.15 , 0.26)
(0.18 , 0.29)
(0.20 , 0.32)
(0.21 , 0.34)
(0.23 , 0.37)
(0.32 , 0.55)
(0.40 , 0.69)
(0.42 , 0.73)
(0.04,0.13)
(0.08,0.18)
(0.13 , 0.24)
(0.17.0.28)
(0.19 , 0.32)
(0.21 , 0.35)
(0.23 , 0.38)
(0.24 , 0.40)
(0 00 , 0.05)
(0.02 . 0.09)
(0.03 , 0.10)
(003 ,0 11)
(0.06.0.14)
(0.10.0.19)
(0.13,024)
(0.19,031)
(0.25 . 0 41)
Proportion Greater Than
15pg/dl
Pr(Pb>15)
0.01
0.03
0.04
0.06
0.07
0.08
0.09
0.1
0.19
0.28
0.31
0.01
0.03
0.05
0.07
0.08
0.09
0.11
0.12
0.00
0.00
0.01
001
0.02
0.03
0.05
007
Oil
95* CI
(0.00 . 0.03)
(0.01 . 0.06)
(0.02 , 0.07)
(0.03 , 0.09)
(0.04,0.11)
(0.05 , 0.12)
(0.06,0.14)
(0.06 , 0.15)
(0.12 , 0.28)
(0.16,0.42)
(0.18 . 0.47)
(0.00 . 0.03)
(0.01 , 0.05)
(0.03 . 0.08)
(0.04,0.10)
(0.05.0.12)
(0.06.0.14)
(0.07.0.16)
(0.07 , 0.17)
(0.00 , 0.01)
(0 00 . 0.02)
(0.00 , 0.02)
(0.00 , 0.02)
(0.01 , 0.04)
(0.02 . 0 06)
(0.03 . 0.08)
(0.05 . 0.12)
(0.07 , 0.17)
Proportion Greater Than
2008/41
Pr(Pb>20)
0.00
0.01
0.01
0.02
0.02
0.03
0.03
0.04
0.08
0.14
0.16
0.00
0.00
0.01
0.02
0.03
0.03
0.04
0.04
0.00
0.00
0.00
0.00
0.00
0.01
0.01
0.02
0.04
95% CI
(0.00 , 0.01)
(0.00 , 0.02)
(0.00 , 0.03)
(0.01 , 0.03)
(0.01 ,004)
(0.01 , 0.05)
(0.01 , 0.06)
(0.02 , 0.06)
(0.04.0.14)
(0.07 , 0.24)
(0.08 . 0 28)
(0.00 . 0.01)
(0.00 , 0.02)
(0.00 , 0.03)
(0.01 , 0.04)
(0.01 , 0.05)
(0.01 . 0.06)
(0.02 , 0.07)
(0 02 , 0.08)
(0.00 , 0.00)
(0.00 , 0.00)
(0.00 , 0.00)
(0.00 . 0 00)
(0 00 , 0.01)
(0.00 , 0.02)
(0.00 . 0.03)
(0.01 . 0 04)
(0.02 . 0.07)
                                    F-4

-------
               APPENDIX G

PROTECTIVE DUST LEAD LEVELS AND EXCEEDANCE
 PROBABILITIES FOR LEAST SQUARES SOLUTION

-------
Table G1. Estimated Dust Pb Loadings for Floors, Window Sills, and Window Wells
         at Which the 85th, 90th, 95th and 99th Percentiles of Childhood Blood
         Pb Concentrations Reach 10, 15, and 20 /ig/dL (Based on the Least
         Squares Regression Models of the Rochester Lead-in-Dust Study Data)
Sample
Type
Floor
Pb Loading
(pg/ft2)
BRM Sampler
Window Sill
Pb Loading
dig/ft2)
BRM Sampler
Window Well
Pb Loading
(^g/ft2)
BRM Sampler
Tolerance
Level
0.85
0.90
0.95
0.99
0.85
0.90
0.95
0.99
0.85
0.90
0.95
0.99
Target Blood-Lead Concentration
10/tg/dL
2
Out of Range
Out of Range
Out of Range
41
5
Out of Range
Out of Range
1,797
173
4
Out of Range
15 pg/dL
108
31
3
Out of Range
3.138
764
65
Out of Range
339,046
59,738
3,079
4
20 /ig/dL
956
319
54
2
38,277
10,884
1,456
12
> 1 Million
> 1 Million
131,989
442
Floor
Pb Loading
(/ig/ft2)
Wipe Samples
Window Sill
Pb Loading
(/tg/ft2)
Wipe Samples
Window Well
Pb Loading
(/tg/ft2)
Wipe Samples
0.85
0.90
0.95
0.99
0.85
0.90
0.95
0.99
0.85
0.90
0.95
0.99
4
1
Out of Range
Out of Range
49
14
2
Out of Range
230
16
Out of Range
Out of Range
41
22
5
Out of Range
687
291
64
2
42,597
8,025
359
Out of Range
189
95
31
1
3,079
1,441
426
23
714,210
167,331
15,874
37
                                   G-l

-------
Table G2. Estimated Dust Pb Concentrations for Floors, Window Sills, and Window
         Wells at Which the 85th, 90th, 95th and 99th Percentiles of Childhood
         Blood Pb Concentrations Reach 10, 15, and 20//g/dL (Based on the
         Least Squares Regression Models of the Rochester Lead-in-Dust Study
         Data)
Sample
Type
Floor
Pb Concentration
(Mg/g)
BRM Sampler
Window Sill
Pb Concentration
0*g/g)
BRM Sampler
Window Well
Pb Concentration
(Mg/g)
BRM Sampler
Tolerance
Level
0.85
0.90
0.95
0.99
0.85
0.90
0.95
0.99
0.85
0.90
0.95
0.99
Target Blood-Lead Concentration
10/ig/dL
Out of Range
Out of Range
Out of Range
Out of Range
118
6
Out of Range
Out of Range
45
Out of Range
Out of Range
Out of Range
15/tg/dL
5,33
584 '
Out of Range
Out of Range
28,655
5,438
196
Out of Range
127,010
13,934
83
Out of Range
20pg/dL
107,321
21,623
1,333
Out of Range
442,459
108,134
10,705
16
> 1 Million
647,078
31,183
1
                                   G-2

-------
Table G3. Estimated Proportion of Children with Blood Pb Concentrations Greater
          than 10. 15. and 20 /ig/dL as Predicted by Least Squares Regression
          Models of Blood Pb versus BRM Lead Loadings from Floors, Window Sills
          and Wells
Surface
Tested
Floors
Window
Sills
Window







Pb
Loading
pgltf
5
10
15
20
25
30
35
40
100
200
250
50
100
200
300
400
500
600
700
200
500
750
800
1500
3000
5000
10000
20000
Proportion Greater Than
lOpg/dl
Pr(Pb>10)
0.13
0.16
0.18
0.19
0.20
0.21
0.22
0.23
0.29
0.33
0.35
0.11
0.13
0.16
0.17
0.19
0.20
0.20
0.21
0.06
0.07
0.08
0.08
010
Oil
0.13
0.15
0.17
95% CI
(0.08 , 0.18)
(0.11 ,0.21)
(0.13 . 0.23)
(0.14 . 0.25)
(0.15,0.26)
(0.16 . 0.27)
(0.17 , 0.28)
(0.18,029)
(0.22 . 0.36)
(0.25 . 0.42)
(0.26 , 0.44)
(007,0.16)
(0.09 ,0.18)
(0.11 ,0.21)
(0.13.0.23)
(0 14 , 0.24)
(0 15 , 0.25)
(0.15 , 0.26)
(0.16 , 0.27)
(0.03.0.11)
(004.0.13)
(0.05 , 0 13)
(0.05,0.14)
(0.06 . 0 15)
(0.07 , 0.17)
(0.09 . 0.18)
(011.020)
(0.13 . 0.23)
Proportion Greater Than
Upg/dl
Pr(Pb>15)
0.03
0.04
0.05
0.05
0.06
0.06
0.07
007
0.1
0.13
0.14
0.02
0.03
0.04
0.05
0.05
0.06
006
006
0.01
0.01
0.01
0.01
0.02
0.03
0.03
004
0.05
95% CI
(0.01 , 0.06)
(0.02 , 0.07)
(0.03 . 0.08)
(0.03 . 0.09)
(0 04 . 0.09)
(0.04 , 0.10)
(0.04,0.11)
(0.04,011)
(0.06 , 0.15)
(0.08 , 0.19)
(0.08 , 0.20)
(0.01 . 0.05)
(0.01 , 0.06)
(0.02 , 0.07)
(0.03 , 0.08)
(0.03 . 0 08)
(0.03 , 0.09)
(0 04 , 0.09)
(0.04 , 0.10)
(0 00 , 0.03)
(0 00 , 0.03)
(0.00 . 0.04)
(0.00 , 0.04)
(0.01 , 0.04)
(0 01 , 0.05)
(0.01 . 0.06)
(0.02 , 0.07)
(0.03 , 0.08)
Proportion Greater Than
20fig/dl
Pr(Pb>20)
001
0.01
0.01
0.01
0.02
0.02
0.02
0.02
0.04
0.05
0.05
0.00
0.01
0.01
0.01
0.01
0.02
002
0.02
0.00
0.00
0.00
0.00
0.00
0.00
0.01
0.01
0.01
95% CI
(0.00 . 0 02)
(0.00 . 0.02)
(0.00 , 0.03)
(0 01 , 0.03)
(0.01 . 0.04)
(0.01 , 0.04)
(0.01 . 0.04)
(0.01 . 0.04)
(0.02 , 0.06)
(0.02 , 0.09)
(0.03 , 0.09)
(0.00 , 0.01)
(0.00 . 0 02)
(0.00 , 0.02)
(0.00 . 0.03)
(0.00 , 0.03)
(0.01 , 0.03)
(0.01 . 0.04)
(0.01 , 0.04)
(0.00 . 0.01)
(0 00 , 0.01)
(0.00 . 0 01)
(0.00 . 0.01)
(0.00 . 0.01)
(0.00 , 0.01)
(0 00 . 0 02)
(0.00 . 0.02)
(0.00 , 0.03)
                                    G-3

-------
Table G4. Estimated Proportion of Children with Blood Pb Concentrations Greater
          than 10, 15, and 20//g/dL as Predicted by Least Squares Regression
          Models of Blood Pb versus Wipe Lead Loadings from Floors, Window Sills
          and Wells
Surface
Tested
Floors
Window
Sills
Window
Wells







Pb
Loading
Jig/ft2
5
10
IS
20
25
30
35
40
100
200
250
50
100
200
300
400
500
600
700
200
500
750
800
1500
3000
5000
10000
20000
Proportion Greater Than
lOjig/d!
Pr(Pb>10)
0.11
0.15
0.18
0.20
0.22
0.24
0.26
0.27
0.36
044
0.47
0.10
0.14
0.19
0.22
024
026
0.28
0.29
0.09
0 12
0.13
0.13
015
017
0.19
0.22
0.25
95% CI
(0.06,0.17)
(0.11,0.21)
(0.13 , 0.24)
(0.16.0.26)
(0.17 , 0.28)
(0.19.030)
(0.20 , 0.32)
(0.21 . 0.34)
(0.27 , 0.47)
(0.32 , 0.58)
(0.33 , 0.61)
(0.06 , 0.16)
(0 10 , 0 19)
(0.14 , 0.24)
(0.17 , 0.28)
(0.18,031)
(0.20 . 0.33)
(0.21 . 0.35)
(0.22 . 0.37)
(005.0.15)
(0.07 , 0 17)
(0.08 . 0.18)
(0.09 ,0.19)
(0.10 , 0.21)
(0.13,023)
(0 14 , 0 25)
(0.16.0.28)
(0.19.0.32)
Proportion Greater Than
ISpg/dl
Pr(Pb>15)
0.02
0.04
0.05
0.06
0.07
0.08
0.09
0.09
0.15
0.20
0.22
0.02
0.03
0.05
007
0.08
0.09
0.10
0.10
002
0.03
0.03
0.03
0.04
005
0.06
007
008
95% CI
(0.01 , 0.05)
(0.02 , 0.07)
(0.03 , 0.08)
(0.04 . 0.10)
(0.04.0.11)
(0.05.0.12)
(0.05 , 0.13)
(0.06 , 0.14)
(0.09 , 0.22)
(0.12 , 0.31)
(0.13 , 0.34)
(0.01 , 0.04)
(0.02 , 0.06)
(0.03 , 0.08)
(0.04 . 0.10)
(0.05 , 0.12)
(0.05,0.13)
(0.06 , 0 14)
(0.06 , 0.16)
(0.01 . 0.04)
(0.01 . 0.05)
(0.01 , 0.06)
(0.01 , 0.06)
(0.02 . 0.07)
(0.03 , 0 08)
(0.03 . 0.09)
(0.04.0.11)
(0.05.0.13)
Proportion Greater Than
20/ig/dl
Pr(Pb>20)
0.00
0.01
0.01
0.02
0.02
0.03
0.03
0.03
0.06
0.09
0.10
0.00
0.01
0.01
0.02
0.03
0.03
0.03
0.04
0.00
0.00
0.01
0.01
0.01
0.01
0.02
0.02
0.03
95% CI
(0.00,001)
(0 00 , 0.02)
(0.00 . 0.03)
(0.01 . 0.04)
(0.01 . 0.04)
(0.01 , 0.05)
(0.01 , 0.05)
(0.01 . 0.06)
(003,0.11)
(0.04 , 0 16)
(0.05,0.19)
(0.00 , 0.01)
(0.00 , 0.02)
(0.00 . 0.03)
(0.01 , 0.04)
(0.01 , 0.05)
(0.01 , 0.05)
(0.02 . 0 06)
(0.02 . 0.07)
(0 00 , 0.01)
(0.00 . 0.02)
(0.00 , 0.02)
(0.00 . 0.02)
(0 00 . 0.02)
(0.00 , 0.03)
(0.01 . 0.03)
(0.01 , 0.04)
(0.01 , 0.05)
                                    G-4

-------
              APPENDIX H

PLOTS COMPARING THE ERRORS IN VARIABLES
MODEL RESULTS FOR BRM AND WIPE SAMPLING

-------
              50 ^
              40
              30
          TJ
          §  20
          CD
               10
                              10
100
1000
10000
                                       Floor Lead Loading (yug/sq. ft.)

                                     Predicted EIV — BRM   	Predicted EIV — Wipe
100000
1000000
Figure H1.   Rochester Lead-in-Dust Study Floor Lead Loadings - Estimated Regression Curve for Errors in
            Variables Model From BRM and Wipe Sampling

-------
to
                  50 •{
                  40
                  30
             -Q
             Q_
              8   20
             m
                  10-
                                 10          100         1000         10000        100000

                                      Window Sill Lead Loading (/zg/sq. ft.)
1oooooo
                                        Predicted EIV — BRM   	Predicted EIV — Wtp«
   Figure H2.   Rochester Lead-in-Dust Study Window Sill-Lead Loadings - Estimated Regression Curve for Errors
               in Variables Model From BRM and Wipe Sampling

-------
              50 H
              40
              30
          -O
          EL
          TJ
          |  20
          DO
               10
                             10
      100         1000         10000        100000
Window Well Lead Loading (yug/sq. ft.)
1000000
                                    Predicted EIV — BRM   	Predicted EIV — WIp.
Figure H3.   Rochester Lead-in-Dust Study Window Well-Lead Loadings - Estimated Regression Curve for Errors
            in Variables Model From BRM and Wipe Sampling.

-------