-------
I

 CO
 a>

f



I

i2


f

 Q.


 a
CO

c
o
*5
C*
a>
I
~
9
&
c
£
o
£
5
o
-
i
«•
£
•o.
1
d

'S-
?
j
1
o
c i
_o e
s __
« '
2 §
0 -I
c
•*
i
•5|«r
•wt **"* «NN*
•a « ?
w 4* 3,
> 5 •*
**"• "o
n 0}
^L *3
>> 2
*" eo
z
• 0) O
«il
E | 5
p o &
£|
"S ^ —
111
lt_ i-
ft 5
H w
Z

Time
Following
Intervention
9 ^€
§ "S "5
111

•8*
A W

^1

z
^

111!
(9 X *£ f
D **
£
2
(0
•8
E
1

fx 5 CO
o
o ^
o>
10
CO

Immediately
Post

co co r:
VO CM »
o> u> ®
U
° w
s
CO

o>
o
5
o>
CD
CM °° 5

o
^s
^ in
S

CO

_ "3>
o
fi
3 -c
 0) 0) P
I- «j ^ *~ ro
°? *- ^co
o>
0 o
^^ ILO ^"^ LO
SI 2.
t o>
IO 10

"o 10
s s co e
•a S ' S
|£ 2*
** CM p
co CM a>
r* 

0
^S
£

CO
u)
^
CM at
fl

0« -^
10 S "i
o u
^ U) ^ LD
£ S
CM CM

f ?| «
111 ^1
0) 5 «0 tO O
po 5
P £J
o« 5 ^
10 2. 5
o o
® tG *•* ID
S S
-------
 i
•8
e
o
. «5
i
2
g
!
o
1
!fe
O
1
o
K
TJ

»
O


o
*s
1
I
1
i
o
If
c 5
0 0
«3 K
i-o
i ?
3
I
"S
c
C






"Me
s *"* »••»
1J3
VI
•" w
z
o> o
lit
M 1
0**:
> w -^
*• *o
9 £
> «
*" w

z

Time
Following
Intervention
H|
2> ^) •<•?

1* ^_
O ~
°-\
*~ 09
*
£
"if
• •« 3 *
3 > 0»
sil

>
•o
s
•s

a









CM g
O *° t CO
10 «" - «
C «
O 0

0j£ o^
& S



!-£§ *£
l|s §1
W 5 OJ V O
| £0 *" s







x: o.
s ?
1 1
O) ~5
r- LL

+;
§2S«
| a| >
•£ ? « =
al|fi
£









^- «* CM CM *- IO ,_
10 *- oo co 10 r» .^
oo ^h co cO co T— l—
i- CM r- t- r- i- "

5 S 5 S S S 5
O O O O O  at
£ r. J=. J=
« c c c
5 o o o
| S 5 5
_ l^j flft <(f
** T- «- CM
to
CO
(O
r"



0

CO
CO
>•-
O TJ
•> £
£  to
o 2 2 x: j=
4-> JC. J- *-• 4-"
•S c c S c
ill Si
£ CM <0 2 »
^J-
CO
r»
(>•



2
O

CO
ID
"2 c
£5 -2
m ^ 4-1
~~ oa C
we ?
l°l
3 —














o>
o
CM

5
o


CO

ca
C
O
5
•
-------
 o
 o
CM
CO
e
S
~
I
e
1
e
i
1
£
O
oc
. IB
O
T1
to
O
W*~
I
rt-lnterven
£
«
Q
5 a
II
2 a
4-1
1
||1
t-
9 *
& IB
H w
z

O) O
1||
Value of
Statistic
U/g/ft2)

^
*o —
o «
S *
*~ W
Z
Time
ollowlng
ervention
"•I
o c6 t*v
ill
> CO •*
•sTa
||
z
111!
O * "5 £
.>
3
>*.
0
I
5



5
C3

!?
Months
CO
O
§



5
o

10
r*-
2 Months

10 CO i-
O ^ r»"
't CO CO



2 2 S
O O O

2 * ^
ca en o>
jz j= x:
O O O
S S S
«- CM ^~
CO
id
"t

o
CO
ll
"° c
2
*">
S "o
0 3
*+3
aa
to








o - ^

^
0

° w
S
CO
CN
•5
+rf
*!
p£
E
O 
O)
Months
CO
CO CO o
O 10 o
CO N ^
o
13 ID
0)
0>
CM
"5
|
03
CO CO 4? CM CO P
cn in v- co CM cb
^ o t- co i- g>
*" ^ CM ^ CO CO

,— — ,
0 0

o jjj o ^
O) O)
,fr to
CO CO
>
« 5
(0 4- *^
T3 O O
!«• s
E <°
en 10 co"
O CO SK
CM r>. CM
to n rx
O
»S
O)
to
•o
.5
0
4^
g ^ '3 c
C .2 ~o t T3
5 =5 ^ f 2
m H o <








O
CM



S
O

o
Months
CO
CD

o
o
Automatic
Intervention
o



S
o

O)
Months
CO
in
o
CM

O
«
Randomized
Intervention
E c
2 -B«
JBI yj L _^
§ Q S 2
1 ^
                                                   H-33
-------
CN
CO
.2

§
. 7S
jj

e
£
i
E
o
^••i
*
i
5
5
••'
3
CO
•o
1
3
O
=
CA
V#
|
!
i
i
f^
• 5= m
55 i
c i
1 ^
O "®
> CQ
c
1
"8|-jr

••• (Q Op

> 0) •*
*•
« S
>I
w

z

C38_O
Q •£ 5
ll 2
^ 2
•S|C
111
oS
^* •£
?I
z
Time
Following
Intervention
«co £^
•S IT
3 2 ^*
111

•si
O •£
Q* **
^w
z
nil
3 a • * <»
jjUs
%3 <"«" JJ •«•>
>.
TJ
3
0)
^^
0
E
a

o
Op



CM
tt>
i


to
to


c •* o
•2 § ?
5 S. 6

i^
^
ID
"3
4>4
^ §
1

CM
ID
.2
T3
i
0)
€
i
CO

CA
C
to
'•&

M
1
O
CM

0
0>
10
T3
i
ID
Immediately
Post
\ 6 Months
| 1 2 Months
00
ID
CM

.1
•o
o

S
o
1

5


























o> r-
•<* co
c c
.2 .2
^3 ^3
I 1
CM
CO
Immediately
Post
6 Months
,_
G)
*~

C
CO
^5
CO
1

00
CO
.2
T3
ID

| 1 2 Months








CM
CO
i









(O CO O
u> ^t *
c c c
<0 CD (Q
^3 ^3 ^3
ill
CO
CM
- en «°
B> 2 ^1
•t^ •••• *^
« tt e =
=5 0 0 °
S a- 2 s
1 «, 2

f^,,
^

ra
"•5

1
0)
CM
C
O

ffi


CM P». O
CO ("- ID
C C C
CD 4D CD
^ ^ ^

5
t S 1
P S 1
|Q. 5 S
C fM
E <° -
CO

CO

c
.2
•^
®
S
CO
cii
CB
i


00 CM CM
C C C
^3 ^3 ^3
111
(D
CO
i—
£• * <»
^ 2 £
« S 1 ^
=5 § § o
c °- S 5
1 « «

CO
CM

.2
^3
(D
S
-------
CN
n
X
e


o
"1
i
• *.
9
£
E
!
2
1
o

3
S
oe
CB
2
I
Wmmm
JO
1
I
i
a
terventlon
Lead Resi
e
T
1
•50-
3. S ^
5 * S

^ t-

0 H
B. *S
£ S
"9
z
00
••f c
E * «
Pi 2
0 *
llf
5* 09 *^
*-*b
,|S
w
Z
o» o
HI
o •
•2s
"SSc
» J2 5
Type of
Statistic1


z
e
mi
|x||


•

0
«-•
CO 4-1
11
1
CO
1
O
«.
Months
^
CO
c
.5
i


CO
T3
C
CD
TJ
O
ir
^*
CM
C
ID
T3
CD


§
C
.S


5
c
OJ
CD

01
CO
I"
CD +•*
= CO
|£
E
CO
1
o
(O
Months
CM
2
.S
CD


CM
CO

'c
o
CD

CM
CM
C
CD
i


CO
c
CD
TD
CD


10
c
CO
T(
CD

to
« s
11
1
CO
4-1
O
CO
Months
£
o
10
t—
c
.5
' 1


10

'Z
o
o
CO
„
m
o
£
S
"

Q
3
X












^. p co
J> £
O £ £
U> 10
CM P^
CO
CO
S
CO
CM

ID pSJ
£ £
OS f
10 10
CM PS
00
PS,
2
co
CM






Unabated
homes

Abated
homes

*>

CO
. 0-

0












CM

i

to
CO
Months
CM
U)
PS
O


o>
CO
f~
Interventioi
Group
w
2r cc
0 <
U UJ
£o
OS

•»s
                                                    H-35
-------
Pre-lnt
     O
     CO
     X.
     §
     s.
(M

X
«
     (0

     o
2

I
     £

    c 3>
    JS, £
    +* oc
    S
       IP
        "!*
         w
    O) O
  «f 1
  p S o

  *!!
        >S
        r 
       c  *
              o
              oo
              (0
              1

              0
               §-
        fe 5 o
        ? g -e
              o
             o '-5
         = £ -o
                          H-36
-------
                      APPENDIX t

   AN ASSESSMENT OF DUST-LEAD LEVELS IN CARPETED FLOORS
AND THEIR RELATION TO CHILDREN'S BLOOD-LEAD CONCENTRATION,
    USING DATA FROM THE ROCHESTER LEAD-IN-DUST STUDY
        AND THE HUD GRANTEES PROGRAM EVALUATION
-------
-------
EXECUTIVE SUMMARY TO APPENDIX I

       This appendix presents statistical analyses of data from two lead-exposure studies, the
Rochester (NY) Lead-in-Dust study and the pre-intervention, evaluation phase of the HUD Lead-
Based Paint Hazard Control Grant ("HUD Grantees") Program (data collected through
September, 1997), where the analyses addressed the following:

       •     the need to extend the floor dust-lead loading standard in the §403 rule to include
             carpeted floors, based on the statistical association between carpet dust-lead
             loading and blood-lead concentration

       •     whether a carpet dust-lead loading standard should be different from the §403
            ' uncarpeted floor standard

       •     whether the standard can be expressed assuming wipe dust collection techniques

       •     whether the presence of carpets in a house is associated with reducing blood-lead
             concentration in children within the house (suggesting that carpets may act as a
             mitigator in reducing the bioavailability potential for lead in floor dust).

While the §403 proposed rule recognized the importance of controlling lead in floor dust when
addressing household lead exposures in target housing, it did not suggest a standard to which
carpet dust-lead levels would be compared. Wall-to-wall carpeting is likely to be encountered in
over three-quarters of target homes in which such a risk assessment is to be done.

       Many factors in a child's environment can contribute to the child's blood-lead
concentration, and as a result, it is difficult to isolate the effects of specific factors (such as lead
in carpet dust) with any degree of accuracy. However, in the analyses within this appendix,
increased blood-lead concentrations were statistically significantly associated with increased
household average floor dust-lead loadings, regardless of whether the floors were carpeted or
uncarpeted.  The blood-lead concentration/carpet dust-lead loading relationship did not appear to
differ statistically between housing units having mostly carpeted floors and units with mostly
uncarpeted floors, and it remains significant after accounting for the effects of certain
demographic parameters. While mixed results were observed in analyses that investigated
whether the significance of this relationship remained after taking into account the effects of lead
in other media for which standards were included in the §403 proposed rule (e.g., soil-lead and
window sill dust-lead), there appears to be a sufficient amount of evidence that carpet-dust
sampling should not be ignored in a risk assessment, thereby warranting the need for a carpet
dust-lead loading standard.

       There is evidence in the results presented in this appendix (i.e., when considering various
performance criteria) to suggest that if a carpet (wipe) dust-lead loading standard is added to the
currently-proposed §403 standards, this standard should be set lower than the standard of 50
ug/ft2 for uncarpeted floors.  This evidence includes the following:

                                            l-i
-------
       •      While the blood-lead concentration/dust-lead loading relationship is consistent
              between carpeted and uncaipeted floors, a housing unit's average carpet dust-lead
              loading tends to be approximately 75% of its average dust-lead loading for
              uncarpeted floors, assuming wipe collection techniques.

Adding a carpet dust-lead loading standard of 50 ug/ft2 does not appear to improve the values of
the performance characteristics (e.g., sensitivity, positive predictive value, negative predictive
value) to any degree, regardless of whether or not dust from uncarpeted floors is being evaluated
for lead content at the same time as carpet dust.

       •      When adding a carpeted floor dust-lead loading standard, the sum of the four
              performance characteristics was maximized at a standard of approximately 17
              jig/ft2 in the analysis based on Rochester study data and from 5 to 13 ug/ft2 in the
              analysis based on HUD Grantees evaluation data, regardless of whether or not
              dust from uncarpeted floors is being evaluated for lead content at the same time as
              carpet dust.

When using the Rochester study data to evaluate the performance of a carpet dust-lead loading
standard relative to the performance of an uncarpeted floor standard, without regard to standards
for any other media, these analyses concluded that in order to achieve the same level of
sensitivity observed at an uncarpeted floor dust-lead loading standard of 50 ug/ft2, a carpet dust-
lead loading standard would need to be no higher than approximately 30 ug/ft2. However, other
types of performance criteria did not necessarily set a higher carpet standard in such a bad light.
For example, negative predictive value was similar across the range of candidate standards
(including 50 ug/ft2) regardless of whether the standard represented carpeted or uncarpeted
floors. The outcome of a regression model-based analysis suggested that a carpet dust-lead
loading standard hi the range of 50 ug/ft2 would be at least as protective as an uncarpeted floor
standard at this level, based on the predicted value of blood-lead concentration at which 95% of
children exposed at the standard level would be expected to be below.

       Experts participating in the §403 Dialogue Group meetings indicated that widespread use
of vacuum dust collection methods in risk assessments would not be practical. Furthermore, the
dust standards in the §403 proposed rule assumed that wipe collection methods were being used.
Therefore, a carpet dust-lead loading standard that was not expressed under wipe collection
methods would be very difficult to incorporate by risk assessors.  Based on the findings of this
appendix, no technical reasons were found to suggest that wipe techniques should be excluded as
a candidate dust collection method for carpets.

       Whether considering average dust-lead loadings in a housing unit or loadings for
individual samples, data in the Rochester study suggest that statistically significant (at the 0.05
level) differences were observed between carpeted-floor-dust samples of different dust collection
methods, especially the BRM  vacuum sampler versus the others.  This finding provides evidence
of quantitative differences among the dust collection methods on the amount of lead and dust that
                                           l-ii
-------
is collected from carpeted floors. This implies that floor dust-lead loading standards that may be
applicable to carpets should be tailored to the dust collection method being used.

       la conclusion, a carpeted floor dust-lead standard is most likely needed, not only from a
practicality standpoint, but from a technical one as well. The standard should be based on dust-
lead loadings as measured by the wipe sampling method as wipe sampling is more easily
employed in the field and is even recommended in the HUD Guidelines (USHUD, 1995). There
is some technical evidence that the standard should be lower than the proposed uncarpeted floor
standard of SO ug/ft2, possibly as low as 17 ug/ft2 or 5 ug/ft2, based on analysis of data from the
Rochester study and the HUD Grantees program evaluation, respectively. However, a
recommended standard depends on the specific performance criteria that are of interest, and the
outcomes of characterizing the performance criteria may be associated with considerable data
variability.
                                          l-iii
-------
(This page left blank intentionally.)
                l-iv
-------
11.0   INTRODUCTION

11.1    BACKGROUND

       The U.S. Environmental Protection Agency (EPA) is conducting scientific research in
response to §403 of the Toxic Substances Control Act (TSCA) (Title IV: Lead Exposure
Reduction), as amended within Title X of the Housing and Community Development Act, also
known as the Residential Lead-Based Paint Hazard Reduction Act of 1992. Through §403, EPA
is directed to "promulgate regulations which shall identify... lead-based paint hazards, lead-
contaminated dust, and lead-contaminated soil." On June 3,1998, EPA proposed regulation to
establish standards for lead hazards in  most pre-1978 housing and child-occupied facilities (40
CFR Part 745, "Lead; Identification of Dangerous Levels of Lead; Proposed Rule").  The
standards imposed in this regulation addressed average dust-lead loading (lead amount per unit
area sampled) on uncarpeted floors, average dust-lead loading on window sills, yardwide average
soil-lead concentration, and amount (in square feet) of deteriorated lead-based paint. These
standards, a focal point of the Federal lead program, identify the presence of lead hazards,
defined within TSCA Section 401 as the condition of lead-based paint and the levels of lead-
contaminated dust and soil that "would result" in adverse human health conditions.

       The §403 proposed hazard standards did not include a standard for dust-lead levels on
carpeted floors. At the time, EPA did  not have sufficient information on the statistical
relationship between dust-lead from carpets and children's blood-lead concentrations to allow a
standard to be proposed.  However, some researchers have suggested that separate standards for
floor dust-lead loadings on carpeted and uncarpeted floor are likely necessary (e.g., Clark, et al.,
1996). Also, because the §403 proposed rule specifically stated that the floor dust-lead standard
is for uncarpeted floors, additional guidance must be established for risk assessors who encounter
only carpeted floors when collecting dust samples in a home for lead analysis. Such an encounter
is highly likely based on EPA's analysis of publicly-available data collected from the Lead Paint
Supplement of the  1997 American Housing Survey. Based on this analysis, approximately 54
million housing units built prior to 1978 (or 78% of these units) contain some wall-to-wall
carpeting. Of these units, wall-to-wall carpeting is found in a living room in approximately 47
million units and in a bedroom in approximately 46 million units (i.e., rooms in which children
reside and play most frequently, and therefore, would be targeted in a risk assessment).

       This appendix seeks to address the need for a distinct carpeted floor dust-lead standard by
investigating how dust-lead levels on carpeted floors  impact young children's blood-lead
concentration, over and above that captured by  the planned standard for uncarpeted floors. In
addition, this appendix provides some guidance on whether the standard for uncarpeted floors
can be extended to carpeted surfaces, or whether some other standard is more appropriate. While
the scientific literature has attempted to address some of these issues (see USEPA, 1997a)', this
   1  This appendix has its own reference list at the end of the appendix.

                                          1-1
-------
appendix presents the results of statistical analyses on existing data that more clearly and
completely address key issues for §403 rule development.

       This appendix also presents how the results of dust-lead analyses can differ when a wipe
dust collection method (i.e., the method assumed for the dust standards within the §403 rule) is
used to sample dust from carpets versus other techniques (e.g., vacuum). As wipe sampling
tends to perform differently for different substrates, its performance on carpeted surfaces can
vary according to the type of carpet and is likely to be different from uncarpeted surfaces. This
issue must also be addressed when considering an appropriate carpet dust-lead standard.

11.2   OBJECTIVES
       The specific objectives of the statistical analyses presented in this appendix are as
follows:

       1.      Assess the need for a carpeted floor dust-lead loading standard by doing the
              following:

              •      Characterize the relationship between floor dust-lead levels and blood-lead
                    concentration in young children and how this relationship differs for
                    carpeted and uncarpeted floors (with and without adjusting for the effects
                    of key demographic variables and for lead levels in other media
                    represented by standards in the §403 proposed rule).

              •      Determine the added value of including a carpet standard given the current
                    proposed §403 standards for soil, window sills and uncarpeted floors.

       2.      Identify appropriate candidates for carpeted floor dust-lead standards and, in
              particular, whether 50 ug/ft2 (i.e., the proposed uncarpeted floor dust-lead
              standard from the §403 proposed rule) should be considered as one candidate.

       3.      Determine whether the wipe technique is acceptable for sampling dust from
              carpeted floors for evaluating the risk of lead exposure associated with carpet-
              dust, or whether alternative vacuum methods are more appropriate.

The appendix addresses these objectives by presenting the results of statistical analyses on
existing data from two lead-exposure studies: the Rochester (NY) Lead-in-Dust study, and the
pre-intervention, evaluation phase of the HUD Lead-Based Paint Hazard Control Grant ("HUD
Grantees") Program (data collected through September, 1997).

       The conclusions made as a result of the analyses conducted in support of the above
objectives were presented in Section 6.5 of the §403 risk analysis supplement report. For the two
studies whose data are analyzed in this appendix, Section 13 presents relevant information on
study design and data handling that should be considered when interpreting the results and

                                           1-2
-------
conclusions of these analyses. The statistical methods used in these analyses are presented in
Section 14, and detailed results of these analyses are presented in Section 15. Each subsection
within Sections 14 and IS is devoted to addressing one of the above three objectives.

12.0   THE POTENTIAL FOR LEAD EXPOSURE ASSOCIATED
       WITH CARPET DUST

       Several field and laboratory studies documented in the scientific literature have
investigated the nature and magnitude of lead in carpet-dust, as well as how to characterize dust-
lead contamination hi carpets. For example, Adgate et al., 1995, corroborate evidence that
carpets can hold large amounts of dust and soil, thereby increasing the likelihood of carpets being
lead-contaminated relative to other surfaces.  In older, chronically-contaminated carpets,
exposure to lead within the carpet can be delayed over tune as normal cleaning procedures and
activities can gradually bring deeply-embedded lead-dust to the carpet surface (Adgate et al.,
1995). As a result, such carpets can represent a continuing source of lead exposure, even after
other interventions have reduced or eliminated other exposure sources.

       While the performance of wipe techniques to collect carpet-dust can vary across different
types of carpet, Wang et al.,  1995, found that the dust collection efficiency of vacuum techniques
on carpeting can also vary based on factors such as carpet pile height, vacuum velocity, dust
loading within the carpet, and relative humidity2.

       A detailed presentation of the key findings of published studies investigating the
measurement of lead levels in carpet, the relationship of these levels with blood-lead
concentration hi children, and efforts to mitigate lead exposures associated with carpets, is found
in USEPA, 1997a.

13.0  STUDY INFORMATION

       To address the above objectives (Section 11.2), statistical analyses were performed on
data from the Rochester Lead-in-Dust study and on pre-intervention data from the HUD Grantees
program evaluation.  These studies measured lead levels in environmental media such as exterior
soil and ulterior dust collected from carpeted and/or uncarpeted floors, window sills, and window
wells.  Also measured were blood-lead concentrations in resident children.  The final report on
the Rochester study is found in The Rochester School of Medicine and NCLSH, 1995.
Rochester study results addressing specific questions are found in Lanphear et al., 1995;
Lanphear et al., 1996a; Lanphear et al., 1996b; and Emond et al., 1997.  NCLSH and UCDEH,
1998, presents an interim report of data collected in the HUD Grantees program evaluation
through September, 1997.
     Both Adgate et al., 1995, and Wang et al., 1995, document findings from various phases of EPA's Childhood Lead
Exposure Assessment and Reduction Study.

                                           1-3
-------
       Section O.I presents an overview of the designs of these studies, including the dust
collection methods used and types of data collected, and discusses the relevance of using data
from these studies in addressing the objectives of this appendix. The data used to address these
objectives and the data endpoints used in the analyses presented in this appendix are found in
Section 13.2.

13.1   STUDY OVERVIEWS

13.1.1  The Rochester Lead-in-Dust Study

       Performed in 1993, the Rochester Lead-in-Dust study was a cross-sectional lead-exposure
study of 205 children aged 12-31 months who lived in the city of Rochester, New York, and had
no known history of elevated blood-lead concentrations. The objectives of this study were to
evaluate 1) the effect of dust-lead contamination on the blood-lead concentrations of these
children, 2) how this effect differed under differing dust collection methods, 3) whether dust-lead
loadings or concentrations were more predictive of children's blood-lead concentrations, and 4)
which surfaces should be routinely sampled for dust in a risk assessment.

       The study sample consisted of a random sample of children born at three urban hospitals,
where the births were listed within hospital birth registries and occurred from March 1,1991,
through September 30,1992. Thus, the sample was considered representative of the general birth
population of the city of Rochester during this period. However, as the study was conducted in a
single urbanized area, the sample may not be representative of the entire nation.

       The children in the study sample primarily had moderate exposure to lead at their
residence. The geometric mean blood-lead concentration for these children was 6.37 ug/dL,
compared to 3.1 ug/dL for U.S. children aged 1-2 years as estimated by Phase 2 of the Third
National Health and Nutrition Examination Survey (NHANES HI), which was performed from
1991-1994 (CDC, 1997). Approximately 23% of the children in the  study had blood-lead
concentrations of at least 10 ug/dL, and 3% had blood-lead concentrations of at least 20 ug/dL.
This compares to national percentages of children aged 1-2 years (as estimated by Phase 2 of
NHANES ffl) of 6% at or above 10 ug/dL and 0.43% at or above 20 ug/dL (CDC, 1997;
USEPA, 1997b).  Children in this study tended to reside in older housing (84% of the units were
denoted as being built prior to 1940) and to belong to households in the lower-income bracket,
both characteristics of residential environments with a high potential for lead-based paint
hazards. White children and African-American children participated in the study at
approximately equal proportions, each constituting  approximately 42% of the monitored children
in the study.

       Three dust sampling methods were used to collect dust samples in the Rochester study:
the BRM vacuum sampler, the DVM vacuum sampler, and the wipe  method. The BRM vacuum
sampler is a modified, portable version of the high-volume small surface sampler (HVS3;
Roberts et al., 1991), an ASTM standard device for collecting dust "from carpets or bare floors to
be analyzed for lead, pesticides, or other chemical compounds and elements" (ASTM, 1996). -

                                          1-4
-------
The D VM vacuum sampler was developed for use in studies that characterize lead exposure
pathways from environmental media to blood (Que Hee et al., 1985). In sampling carpet-dust,
the DVM vacuum tends to collect only the surface dust that is more readily available to children
(generally particles less than 250 um in diameter), and not the more deeply-embedded dust in the
carpet that the BRM vacuum is capable of sampling. The third method, wipe sampling, collects
dust from a surface by wiping the surface with a premoistened digestible wipe. ("Dttle Ones"
brand baby wipes were used in the Rochester study.) As it can be difficult for the wipe method
to collect dust embedded deeply within carpet fibers, it tends to collect only the most readily
available surface dust from carpets.

       From August to November, 1993, floor-dust samples in the Rochester study were
collected from five rooms within a housing unit: the entryway, child's bedroom, child's principal
play area, kitchen, and living room.  Window sill dust samples were collected within four rooms:
the child's bedroom, child's principal play area, kitchen, and living room.  Window well dust
samples were collected within three rooms: the child's bedroom, child's principal play area, and
kitchen. Within each room, three dust samples were collected side-by-side on a given
component type, with the first sample collected using a wipe, the second using the DVM
vacuum, and the third using the BRM vacuum. For floor-dust samples, information was also
collected on whether or not the floor was carpeted, and if so, the condition of the carpet (good,
average, or poor) and whether the carpet was of high-pile or low-pile.

       Among the data collected in the Rochester study were the following:

       •      lead loading (amount of lead per sample area) in dust samples from floors,
              window sills, and window wells, using each of the three dust collection methods.
              Dust  samples were analyzed using flame atomic absorption (FAA) or graphite
              furnace atomic absorption spectrophotometry (GFAAS).

       •      lead concentration (amount of lead per weight of sample) in dust samples from
              floors, window sills,  and window wells, using the DVM and BRM vacuum
              methods.

       •      lead concentration in soil samples collected from the dripline (foundation) at 186
              housing units and from children's play areas  at 87 units. Soil samples were
              fractionated into fine and coarse soil fractions, both of which were analyzed using
              FAA. The fine soil fraction results were considered in the analyses of this
              appendix.

       •      blood-lead concentration  for participating children, with their blood collected via
              venipuncture and analyzed by GFAAS.

       •      lead levels on up jp 15 painted surfaces in the unit from within the kitchen, child's
              bedroom, child's principal play area, and entryway, as well  as on the exterior.
              The Microlead I portable  x-ray fluorescence (XRF) measurement device was used,

                                           1-5
-------
             but laboratory testing of paint chips was also employed if the XRF could not be
             used or if the result was deemed inconclusive. A rating on the extent of any
             deterioration of the sampled paint (0-5% deteriorated, 5-15% deteriorated, >15%
             deteriorated) was also determined.

       •     demographic information on the household and on the resident children, such as
             income level, age of child, nutritional and feeding information, types of activities,
             and tendency for pica.

       The study units generally had low dust-lead loadings on floor surfaces in this study. The
study-wide geometric mean dust-lead loading for wipe dust samples were 16 jig/ft2 for
uncarpeted floors and 11 ug/ft2 for carpeted floors.

13.1.2 The HUD Grantees Program Evaluation

       In 1993,70 state and local government agencies were awarded grants by the U.S.
Department of Housing and Urban Development (HUD) to "initiate or expand lead-based paint
inspection, abatement, and training certification programs in order to reduce the health hazards
associated with exposure to lead-based paint and lead dust... and to plan and implement cost-
effective testing, abatement, and financing programs, including the testing of innovations that can
serve as models for other jurisdictions interested in addressing this problem..." (HUD, 1992
Notice of Funding Availability). This ongoing national program is known as the HUD Lead-
Based Paint Hazard Control Grant Program in Private Housing, or the HUD Grantees program
evaluation. In this program, enrollment and lead hazard control interventions are still ongoing,
with post-intervention environmental monitoring continuing for up to three years following
interventions.

       The grantees in the HUD Grantees program evaluation are implementing effective, low-
cost intervention and financing programs to control lead-based paint hazards in privately-owned
low- and middle-income housing. As part of a formal evaluation of the program, the fourteen
grantees listed in Table 3-4 of the §403 risk analysis report are also collecting extensive data on
environmental, biological, demographic, housing, cost, and hazard-control aspects of the
intervention activities that they are conducting in this program. This evaluation is intended to
determine the relative cost and effectiveness of the various methods used by states and local
governments to reduce lead-based paint hazards in housing. Among the pre-intervention data
being collected in this evaluation are the following:

       •     lead loadings in dust samples using wipe collection techniques (the DVM vacuum
             sampler was occasionally used on carpets). Carpeted and uncarpeted floors,
             window sills, and window wells were sampled. Sampled rooms included
             entryways, children's principal play room (or living room), kitchen, and up to two
             children's bedrooms. The program directed that two dust samples per surface
             type per room should be taken.
                                          1-6
-------
      •      blood-lead concentration for children between the ages of six months and six
             years (although data exist for children as old as eight years). While the program
             recommended venipuncture collection techniques, some grantees used fingerstick
             methods occasionally. Blood samples were analyzed by GFAAS or by anodic
             stripping voltammetry (ASV).

      •      soil-lead concentration in composite soil samples collected from the dripline
             (foundation) and from children's play areas. Soil sampling was optional in this
             program, collected by only 8 of the 14 grantees.

      *      lead levels on painted surfaces measured to determine the presence of lead-based
             paint. Portable XRF measurement techniques were used, but laboratory testing of
             paint chips was also employed if XRF measurements were indeterminate.

      •      demographic information on the household and on the resident children, such as
             income level, age of house, age of child, and mouthing behavior.

Grantees collecting environmental and blood samples followed specified sampling protocols and
used standard data collection forms developed specifically for this evaluation.

      The pre-intervention data considered in this analysis were collected from February, 1994,
to August, 1997, and therefore provide some of the most recent information on baseline
environmental-lead measurements and their relationship with blood-lead concentration in
children. However, the HUD Grantees data are not meant to be representative of data for the
nation as a whole. The grantees were not selected to achieve a statistical-based sample of
geographic areas of the country. In addition, as it was HUD's desire to emphasize local control
of the individual programs, each grantee participating in the program was given some freedom in
developing their approach to recruitment and enrollment.  Some grantees targeted high-risk
neighborhoods in their enrollment procedure, while others enrolled only homes with a lead-
poisoned child, while still others considered unsolicited applications.  Thus, when interpreting
results of any analyses of data from this program, one should be aware that these data represent
housing units that are more likely to contain lead-based paint hazards or to contain children with
elevated blood-lead concentrations than is the population  as a whole (e.g., higher incidence of
older or low-income housing or sampling from neighborhoods with a history of lead-based paint
hazards).

13.2   DATA HANDLING

      For the analyses presented in this appendix, Rochester study data were obtained in
electronic format directly from the Rochester study team.  Pre-intervention data collected in the
HUD Grantees program evaluation through September, 1997, were obtained from the University
of Cincinnati. Outlier screens and logic checks were performed on the HUD Grantees data prior
to analysis, and unusual data values were checked for accuracy and corrected if necessary.
                                          1-7
-------
Version 6.12 of the SAS® System was used to manage the data and conduct all data summaries
and statistical analyses presented in this appendix.

       Data for all 205 housing units in the Rochester study and for 395 housing units across 13
of the 14 HUD grantees were included in the analyses presented in this appendix. As the effects
of carpeting on the relationship between lead-based paint hazard and children's blood-lead
concentration were to be investigated in this appendix, analyses of HUD Grantees data involved
only those housing units which had data on both of the following:

       •     blood-lead concentration for at least one resident child, where the blood samples
             were obtained by venipuncture, and

       •     floor dust-lead loadings, where the type of floor surface (carpeted, uncarpeted)
             and the dust collection method (wipe or DVM) were specified.

In addition, to ensure the integrity of the relationship between environmental-lead and blood-lead
measurements in a given unit, the following blood-lead concentration data were omitted from the
analysis of HUD Grantees data:

       •     data for children who had earlier treatment for lead poisoning, such as chelation

       *     data for children residing in the unit for less than three months

       •     data for children not residing in the unit until after dust samples were collected

       *     data for children whose blood was sampled more than four months after dust
             sample collection.

Data for all Rochester study units were considered in the analyses in this appendix, as the
Rochester study design allowed for more detailed analyses on relationships between dust-lead
measurements for different dust collection methods.

       The analyses presented in this appendix assumed that each housing unit in both studies
was associated with a blood-lead concentration for a single child. This was true for units in the
Rochester study,  but some units in the HUD Grantees program evaluation had blood-lead
concentrations for multiple children. For these units, data for only the youngest child 12 months
and older were considered. If all children in a unit were younger than 12 months, data for the
oldest child was selected.  In one instance, when these criteria did not yield a single child (e.g.,
twins born on the same day), a child was selected randomly from those meeting the criteria.

       When reviewing the data more closely (Appendix 12), some of the HUD grantees
frequently reported the same dust-lead loading value across different locations or housing units.
Although not confirmed, this value is likely an estimated lead level that is below a limit of
detection and is equal to the detection limit divided by the square root of two. In the analyses

                                           1-8
-------
presented in this appendix, these values were treated as actual values rather than censored values.
However, excessive numbers of data points that represent not-detected lead levels can impact
underlying data assumptions relevant to the statistical analyses and can introduce considerable
bias to the analysis results.

       In each study, the floor dust-lead measurements for each housing unit were categorized by
dust collection method, measurement type (loadings or concentrations), and whether the sample
was taken from a carpeted or an uncarpeted surface. These categories are presented in Table
13-1.  Floor dust-lead measurements could be placed into ten categories in the Rochester study
and three categories in the HUD Grantees program evaluation.  For each housing unit, the area-
weighted arithmetic average of floor dust-lead loadings (i.e., each measurement is weighted by
the area of the sample) was calculated for each dust collection method used and floor surface
type sampled in the unit. In addition, within the Rochester study, the mass-weighted arithmetic
average of floor dust-lead concentrations (i.e., each measurement is weighted by the mass of the
sample) was calculated for each vacuum dust collection method used and floor surface type
sampled in the unit. While floor dust-lead loading as measured by the wipe method was the
primary floor-dust endpoint used in the statistical analyses, descriptive statistics were reported in
Appendix 12 for all three sampling methods and both measurement types (loading and
concentration). Typically, all available interior floor-dust measurements in the unit, including
measurements from rooms other than those specified within the study design, were used in
calculating these endpoints. However, in the Rochester study, data for dust samples from
exterior surfaces such as driveways and porches were not included.

       Table 13-2 contains additional endpoints used in the statistical analyses that were
calculated from data in these two studies. As indicated in this table, dust-lead measurements on
window components were summarized within each unit by taking area-weighted averages (for
loadings) or mass-weighted averages (for concentrations) by dust collection method. Only dust-
lead data for windows located in a kitchen, play area, living room, or bedroom were considered
in the Rochester study. When calculating the endpoint representing paint-lead level, lead
measurements corresponding to intact paint were set to zero (as intact paint was not considered
to pose a lead hazard), and the 75th percentile of all paint-lead measurements in the unit (i.e., the
level  where  75% of the measurements were below it) was determined. The "lead-based paint
hazard score" is a measure of both the extent of deteriorated lead-based paint in either the interior
or the exterior of the unit and paint pica tendencies in the resident child. The endpoints in Table
13-2 were among those considered as predictors of blood-lead concentration in developing the
empirical model used in the §403 risk analysis (USEPA, 1997b).
                                           1-9
-------
Table 13-1.    Types of Floor Dust-Lead Samples and Measurements Taken in the Two
                Studies
Measurement
f
Dust-lead
loading
Dust-lead
concentration
^ -> Sample Type v " -
*" *••*" * '*-.***
, . "•?' " ' ' .? «•- * ' ',* -i-"*" *""•»• jf" <.~?
Wipe dust collection on carpeted floors
BRM (vacuum) dust collection on carpeted floors
DVM (vacuum) dust collection on carpeted floors
Wipe dust collection on uncarpeted floors
BRM (vacuum) dust collection on uncarpeted floors
DVM (vacuum) dust collection on uncarpeted floors
BRM (vacuum) dust collection on carpeted floors
DVM (vacuum) dust collection on carpeted floors
BRM (vacuum) dust collection on uncarpeted floors
DVM (vacuum) dust collection on uncarpeted floors
:: Data Objected
in the Rochester
? Study?- , .
• *: . '*•!-'
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
, Data CoUectid
* -inline HOD"* ,"
': l&ssijsz&r*.
Evaluation?
Yes
No
Yes
Yes
No
No
No
No
No
No
Table I3-2.    Definitions of Additional Endpoints Included in Data Summaries and/or Used
                in Statistical Analyses Within This Appendix
      Endpoint
                                                    :De1iiutiGii of.' .Endpoint
                           Based on Rochester Study Data '
                                             Based on HUD Grantees Program Evaluate
                                                                                 Data
 Percentage of floor
 area consisting of
 carpeted surfaces
Percentage of total sampled floor area
consisting of carpeted surfaces (determined
across all dust collection methods as well as for
each method)
Percentage of total sampled floor area
consisting of carpeted surfaces (determined
across all dust collection methods as well as
for each method)
                    Percentage of total sampled carpeted floor area
                    corresponding to high-pile versus low-pile carpet
                    (calculated only for units with carpet dust
                    sample data)
  Lead levels on
  window sills
Area-weighted arithmetic average of dust-lead
loadings on window sills (determined separately
for wipe, DVM, BRM)
                    Mass-weighted arithmetic average of dust-lead
                    concentrations on window sills (determined
                    separately for DVM, BRM)
Area-weighted arithmetic average of wipe
dust-lead loadings on window sills
                                                  1-10
-------
                                             Table 13-2.  (cont.)
     -Endpoffit

 Lead levels on
 window wells
Area-weighted arithmetic average of dust-lead
loadings on window wells (determined
separately for wipe, DVM, BRM)
Area-weighted arithmetic average of wipe
dust-lead loadings on window wells
                      Mass-weighted arithmetic average of dust-lead
                      concentrations on window wells (determined
                      separately for DVM, BRM)
 Lead levels in soil
Average soil-lead concentration (fine soil
fraction only) across dripline and play areas, or
for only one area if no data exist for the other
area
Defined in the same manner as for the
Rochester study data, but no separation of
sample into size fractions was done
  Lead levels in
  interior paint*
75th percentile of interior XRF paint-lead
measurements in the unit, with the XRF
measurement for a given surface reset to zero
when the measurement exceeded 1.0 mg/cm2
but the paint on the surface was considered
intact, or when the measurement was below
1.0 mg/cmj
Defined in the same manner as for the
Rochester study data.
  Lead levels in
  exterior paint1
75th percentile of exterior XRF paint-lead
measurements in the unit, with the XRF
measurement for a given surface reset to zero
when the measurement exceeded 1.0 mg/cm2
but the paint on the surface was considered
intact, or when the measurement was below
1.0 mg/cm1
Defined in the same manner as for the
Rochester study data.
  Lead-based paint
  hazard score (i.e.,
  extent of a lead-
  based paint hazard)
         if no deteriorated lead-based paint
         exists in the unit, or the child exhibits
         no paint pica
         if deteriorated lead-based paint is
         present in the unit, and the child
         exhibits paint pica rarely
         if deteriorated lead-based paint is
         present in the unit, and the child
         exhibits paint pica at least sometimes
         if no deteriorated lead-based paint
         exists in the unit, or the child puts
         fingers or other objects in his/her
         mouth less than once/week or not at
         all
         if deteriorated lead-based paint is
         present in the unit, and the child puts
         fingers or other objects in his/her
         mouth several times/week
         if deteriorated lead-based paint is
         present in the unit, and the child puts
         fingers or other objects in his/her
         mouth several times/day or more.
 Other demographic
 endpoints '
Ownership status (owner- vs. renter-occupied),
household annual income, age of child, parents'
education, cleaning frequency, mouthing
behavior, family history of lead,  race, gender.
Ownership status (owner- vs. renter-occupied),
household annual income, age of house, age of
child, mouthing behavior, race, season of
measurement, gender, grantee.
1 The 75* percentile is that value for which 75% of the observed XRF measurements in a housing unit are lower (XRF
measurements exceeding 1.0 mg/cm2 for surfaces covered with intact paint were reset to 0 prior to determining the 75*
percentile).
1A household's lead-based paint hazard score incorporates information on the presence of deteriorated lead-based paint in
the unit and paint pica behavior in the child whose blood is tested for lead levels. The score was determined separately for
the interior and exterior of the unit.
3 See Table 14-1 for more details on these endpoints.
                                                       1-11
-------
       The databases for both studies included a variable identified as the year in which the
housing unit was built  This variable, which is either a specified year (Rochester study) or a
category representing a range of years (HUD Grantees), has historically been an important
indicator of the presence and magnitude of lead-based paint hazard. (Lead in residential paint
was only gradually phased out before its ban in 1978, plus paint films deteriorate over time.)
However, the year specified in the Rochester study data may be unreliable, as the Rochester study
team has indicated that  it was taken from public tax assessor records. It is possible that the tax
assessment records of some units actually contain a later year in which a certain event, such as
extensive remodeling, was performed that can affect tax assessments.  Therefore, information on
age of unit was not used in the analysis of Rochester study data.

14.0   METHODS

       This section presents the statistical methods that were developed to address the objectives
in Section 11.2. The results  of applying these methods to data from the Rochester study and/or
the HUD Grantees evaluation are detailed hi Section 15 of this appendix.

14.1   ASSESSING THE NEED FOR A CARPETED FLOOR
       DUST-LEAD LOADING STANDARD

       In the §403 proposed rule, EPA proposed a standard of 50 ug/ft2 for uncarpeted floor
dust-lead loading measured  using the wipe method (Section II. 1). However, risk assessors may
encounter situations where nearly all of the floor in a unit is covered by carpeting, or the only
uncarpeted floor is in an area where lead exposure to children may be minimal (e.g., bathroom).
Clearly, in these situations, any floor-dust samples would come from carpeted floors. Therefore,
a standard would be needed against which to compare these carpeted floor dust-lead
measurements.

       One may argue, however, that if no association is found to exist between carpeted floor
dust-lead loading and blood-lead concentration, then sampling dust from carpets during a risk
assessment (and, therefore, the need for a carpet dust-lead standard) may not be necessary.
Section 14.1.1 presents various methods used to examine whether a statistically significant
association exists between carpeted floor dust-lead loading and blood-lead concentration, both
adjusting for and not adjusting for relevant demographic variables,  and how this association
compares with that where the floor dust is assumed to have come from uncarpeted floors.

       As documented in Section II. 1, the §403 proposed rule included standards for lead in dust
from uncarpeted floors  and window sills, as well as for lead in soil  and for deteriorated paint.
Exceeding any of these standards will trigger the need for  certain interventions in a housing unit.
Nevertheless, certain housing units containing children with high blood-lead concentrations may
not exceed any of these standards, but perhaps would exceed a properly-established standard for
lead in carpet dust. To determine the need for a carpet dust-lead loading standard in the context
of the §403 proposed standards, Sections 14.1.2 and 14.1.3 portray modeling and non-modeling
                                          1-12
-------
approaches, respectively, for evaluating the added benefit that a carpet dust-lead standard may
bring to the set of proposed standards.

14.1.1 Investigating the Association Between Dust-Lead Loading and
       Blood-Lead Concentration for Carpeted and Uncarpeted Floors

       This subsection presents methods for examining the relationship between area-weighted
arithmetic average floor dust-lead loading and children's blood-lead concentration without
considering other environmental-lead sampling. (See Section 14.1,2 for a similar analysis which
does control for other environmental-lead sampling.)  Correlation coefficients and regression
models that account for effects of demographic covariates were used to assess the relationship
between blood-lead and dust-lead for both carpeted and uncarpeted floors.

       Unless otherwise mentioned, the following approaches were taken within each method
described in this subsection:

       •      The analyses were applied separately to carpeted and uncarpeted floor dust-lead
              loading data (assuming wipe dust collection techniques).

       •      Average household dust-lead loadings and blood-lead concentrations were log-
              transfonned, as typically the underlying distributions of these data parameters tend
              to follow a normal distribution more closely upon taking a log-transformation.

       *      When floor dust-lead loadings were assumed to be from carpeted surfaces, the
              data for each housing unit were weighted by the proportion of total floor wipe
              sample area in the unit that was carpeted. (This proportion acted as a surrogate
              for the proportion of actual floor area in the unit that was carpeted.)

       *      When floor dust-lead loadings were assumed to be from uncarpeted surfaces, the
              data for each housing unit were weighted by the proportion of total floor wipe
              sample area in the unit that was uncarpeted.  (This proportion acted as a surrogate
              for the proportion of actual floor area in the unit that was uncarpeted.)

       14.1.1.1. Correlations Between Floor Dust-Lead Loading and Blood-Lead
Concentration. Pearson correlation coefficients between log-transformed average dust-lead
loading and log-transformed blood-lead concentration were calculated for carpeted floors and
uncarpeted floors separately, in order to assess the degree of linear relationship between these
variables for both types of floor surfaces.  Scatterplots of these data were also generated to
further explain the nature of the relationship for both surfaces.

       14.1.1.2. Univariate Regression of Blood-Lead Concentration on Floor Dust-Lead
Loading. The log-linear relationship between average floor dust-lead loading and blood-lead
concentration was investigated by fitting the following regression model (separately for carpeted
and uncarpeted floors):

                                           1-13
-------
                         log(PbBj)  =   n + a*log^bD{) + e.
(1)
where PbB; represents the blood-lead concentration for the child in the i* housing unit, PbD; is
the observed average dust-lead loading (from either carpeted or uncarpeted floors, depending on
the model fit) for the 1th housing unit, u and a are parameters representing the intercept and slope
of the model, respectively, and e; represents error not explained by the model and is presumably
characterized by a normal distribution with mean zero and standard deviation a.  When fitting the
model to HUD Grantees data, separate intercepts (u) were estimated for the different grantees but
not separate slopes (a), as preliminary analyses had determined that there was no significant
improvement to the model by considering grantee-specific slopes. A statistically non-zero slope
(a) suggests that the average dust-lead loading is significantly associated with blood-lead
concentration by the methods used in the model fitting.

       Note that model (1) does not take into account the effects that lead exposure in other
media or the effects of certain demographic variables may have on blood-lead concentration. If
these effects are highly correlated with the effect of floor dust-lead loading, then a portion of the
effect of floor dust-lead loading on blood-lead concentration that is observed from fitting model
(1) may actually be the result of these other factors. Therefore, the degree of association between
the floor dust-lead loading and blood-lead concentration in these regressions is not necessarily
the degree to which floor dust-lead loading causes a change in blood-lead concentration.

       As it was desired to express blood-lead concentration as a function of observed dust-lead
loading, the model fitting does not adjust for measurement error in the dust-lead loading
measurement.

       14.1.1.3. Comparing the Dust-Lead Loadino/Blood-Lead Concentration Relationship
Between Homes With Mostly Carpeted Floors and  Homes With Mostly Uncarpeted Floors.
Most housing units in the Rochester study and HUD Grantees evaluation had floor-dust samples
taken from both carpeted and uncarpeted floors.  Thus, it was difficult for an analysis of these
data to isolate the role that carpeting had on the relationship between lead in floor-dust and
children's blood-lead levels. One approach taken to investigate the role of carpeting was to
consider how this relationship differed between two groups of housing units in each study:

       •     units where floor-dust was sampled from mostly carpeted floors (i.e., > 50%
             carpet-dust samples, by area)

       •     units where floor-dust was sampled from mostly uncarpeted floors (i.e., < 50%
             carpet-dust samples, by area)

(Units where total sampled floor area consisted of equal proportions of carpeted and uncarpeted
floors were omitted from this analysis.) The underlying assumption here was that if the majority
                                           1-14
-------
of sampled floor area in a housing unit was from a single floor surface type, then a resident
child's floor dust-lead exposure derived mostly from that surface type.

       For each housing unit, let pc; equal the proportion of the total floor wipe area sampled in
the 1th housing unit that was carpeted.  Then for each study, the following model was fitted twice,
once for each of two definitions for the predictor variable relating average floor dust-lead loading
in a household:
       log(PbBj)   =  u + a*log(PbDj)  + p0*SURFj + P1*SURFi*log(PbDi ) + £s      (2)
where, in each fit, SURF{ equals 0 or 1 depending on whether pc, is less than or greater than 50%,
respectively, and PbBs represents the blood-lead concentration for the child in the i* housing unit.
The two possible definitions of log(PbDi*) were as follows:

       Fit#l: Surface Majority. Here, log(PbDj') equals the log-transformed average dust-lead
       loading for the floor surface type which makes up the majority of the sampled floor area:

              log(PbDj*)    =      log(PbDj for carpeted surfaces) if pCj > 0.5
                                  log(PbDj for uncarpeted surfaces) if p^ < 0.5

       In this model fit, the ith housing unit was weighted by pq if pc; > 0.5 and by (1-pCj) if pc{
       <0.5.

       Fit #2: Weighted Average.  Here, log(PbDj*) equaled a weighted average of average
       carpeted-floor dust-lead loading and average uncarpeted-floor dust-lead loading in a
       household, with the weights determined by pq:
                     j*) = pc, * log(PbDi(carpeted)) + (1-pCi)* log(PbDj(uncarpeted))

       Equal weight was given to all housing units in this model fit.

Therefore, the first fit only considered dust-lead data for the surface type having the majority of
sample area (and each housing unit was weighted by the proportion of total sample area
representing this surface type), while the second fit considered an overall household average
across both types of floor surfaces.

       The parameters of most importance when interpreting these analysis results were the
parameters |30 and p,. These parameters are "effect modifiers" that represent the change in the
intercept (u) and slope (a), respectively, when homes have greater than 50% of floor-dust
sampled from carpeted floors.  If both P0 and p, are not significantly different from zero, then
these results imply that the statistical relationship between blood-lead concentration and floor
dust-lead loading does not differ significantly between homes that are mostly carpeted and homes
that are mostly uncarpeted.

                                          1-15
-------
       As in model (1), when fitting model (2) to HUD Grantees data, separate intercepts (n)
were estimated for the different grantees, but not grantee-specific slopes.

       14.1.1.4.  Investigating the Association Between Floor Dust-Lead Loading and
Blood-Lead Concentration. Controlling for Demographic Variables. It is possible that even if
one concludes from fitting models (1) and (2) that the association between floor dust-lead
loading and blood-lead concentration is statistically significant, the significance may actually be
due to confounding effects of certain demographic variables such as income, age of house, etc.
In this analysis, the demographic variables listed in Table 14-1 were considered as predictor
variables in an expanded version of model (1) from Section 14.1.1.1. Certain variables from
Table 14*1 were added to the regression model using stepwise selection techniques, and the
household's average floor dust-lead loading was added to the model last This approach,
therefore, evaluated the degree of association between floor dust-lead loading and blood-lead
concentration after adjusting for the effects of important demographic variables.
       The expanded version of model (1) takes the form
"  I*  + £ Pk*2^ + o*log(PbD4)  + et
        k
                                                                                   (3)
where Z^ denotes the value (for the ith housing unit) of the kth in a series of selected
demographic variables, p\ denotes the slope parameter associated with Z^, and the remaining
notation is the same as for model (1) above. Model (3) was fit twice: once using carpeted floor
dust-lead loading when determining PbD; and once using uncarpeted floor dust-lead loading.

       When fitting model (3) to the HUD Grantees data, separate intercepts (ji) for the different
grantees were included among the pool of demographic variables in Table 14-1 that were
considered in the stepwise procedure rather than being forced into the model. Therefore, the
stepwise procedure was allowed to choose which grantees had significantly different intercepts
from the others.

14.1.2 Investigating the Association Between Carpeted  Floor Dust-Lead
       Loading and Blood-Lead Concentration, Controlling for Other
       Environmental-Lead Sampling

       The §403 proposed rule set standards for lead in dust from uncarpeted floors and window
sills, lead levels in soil, and the amount of deteriorated lead-based paint within a household. To
investigate the extent to which a carpeted floor dust-lead loading standard may address that
portion of a child's total lead exposure that is not attributable to the environmental-lead levels
addressed by the proposed standards, the contribution of carpeted floor dust-lead loading
measurements to the prediction of blood-lead concentration, over and above the contributions of
the lead measures that were compared to the §403 standards, was evaluated. The data analysis
consisted of two parts:

                                          1-16
-------
Table 14-1.   Demographic Variables Considered in Stepwise Regressions Examining the
               Association Between Floor Dust-Lead Loading and Blood-Lead Concentration
• Study"




Rochester








HUD
Grantee




Demographic Variable^
Age
Education
Cleaning Frequency 1
Income
Mouthing Behavior z
Lead in Family History
Paint Pica Hazard
Race
Sex
Rent/Own
Age
Income
Mouthing Behavior 3
Paint Pica Hazard
Race
Sex
Year Home Built
Season
' ", 4'^
Child age and square of child age (considered jointly)
0 = sHigh School, 1 = > High School
(Frequency of Sweeping -f- Frequency of Vacuuming +
Frequency of Cleaning Window Wills + Frequency of Wet
Mopping)/1 6
0 = £$15,500 per year, 1 = > $15,500 per year
(Mouth on Window Sill + Pacifier + Soil Pica + Sucks
Thumb)/16
0 = No, 1 = Yes
= 0 if the sum of interior LBP hazard score and exterior LBP
hazard score (Table 13-1 ) equals 0 or 1
= 1 if the sum equals 2, 3, or 4
0 = Non-white, 1 = White
0 - Female, 1 = Male
0 = Own, 1 = Rent
Child age and the square of child age (considered jointly)
0 = £$15,500 per year, 1 = > $1 5,500 per year
(Fingers in Mouth + Toys in Mouth)/6
= 0 if the sum of interior LBP hazard score and exterior LBP
hazard score (Table 13-1) equals 0 or 1
= 1 if the sum equals 2, 3, or 4
0 = Non-white, 1 = White
0 = Female, 1 = Male
0 = Pre-1940, 1 = Post-1940
0 = Fall/Winter, 1 = Spring/Summer
1 Each of the four frequency variables in the sum has possible values 0 = Never, 1 = Less than once per month, 2 =
Monthly. 3 = Bimonthly, 4 = More than once per week.  Thus, the sum ranges from 0 to 1 and was not calculated if data
for any of the terms in the sum were not available.

2 Each of the four mouthing variables in the sum has possible values 0 = Never, 1  = Rarely, 2 = Sometimes, 3 = Often, 4
 = Always. Thus, the sum ranges from 0 to 1 and was not calculated if data for any of the terms in the sum were not
available.

3 Each of the mouthing variables in the sum has possible values 0 =  Less than once per week or never, 1 = Several times
a week, and 2 •= Several times a day or more. Thus, the  sum ranges from 0 to 1 and was not calculated if data for any of
the terms in the sum were not available.
                                                1-17
U.S. EPA Headquarters Library
        Mail code 3201
1200 Pennsylvania Avenue NW
    Washington  DC 20460
-------
              Model (1) in Section 14.1.1 was expanded to consider other environmental-lead
              measures as predictor variables that were selected by stepwise regression
              procedures. These measures were dust-lead loadings for both uncarpeted floors
              and window sills, soil-lead concentration, and paint condition (as represented by
              the paint pica hazard variable). Then, carpeted floor dust-lead loading was added
              to this expanded model in order to assess its association with blood-lead
              concentration after adjusting for these other predictor variables:
                         =  u
                                                                                   (4)
              Same as #1, but the demographic variables in Table 14-1 were also included in the
              stepwise regression procedure as potentially significant predictor variables in the
              expanded model prior to adding carpeted floor dust-lead loading:
                   =  U
                                                                                   (5)
In these two models, for the ith housing unit, Xu denotes the product of log-transformed area-
weighted average uncarpeted floor dust-lead loading and the proportion of sampled floor-dust
that was uncarpeted, X2J denotes log-transformed area-weighted average window sill dust-lead
loading, X34 denotes log-transformed average soil-lead concentration, X44 denotes paint pica
hazard (Table 14-1), Z^ denote the kth in a series of selected demographic variables, and the
remaining terms are as defined for the previous models presented in this section.

      In models (4) and (5), the area-weighted average carpeted floor dust-lead loading was
multiplied by the proportion of sampled floor area that was carpeted and, as mentioned in the
definition of X,, the area-weighted average uncarpeted floor dust-lead loading was multiplied by
the proportion of sampled floor area that was uncarpeted. In model (1), the relationship between
blood-lead concentration and floor dust-lead loading was modeled separately for carpeted and
uncarpeted floors, and observations were weighted by the proportion of sampled floor area that
was carpeted (when considering carpeted floor dust-lead data) or uncarpeted (when considering
uncarpeted floor dust-lead data). In models (4) and (5), carpeted and uncarpeted floor dust-lead
loadings are included in the same model.  Multiplying these values by the proportion of sampled
floor area that was carpeted and uncarpeted, respectively, achieved a similar goal as the
weighting in model (1): carpeted (uncarpeted) floor dust-lead loading measurements taken from
homes where more of the floor was carpeted (uncarpeted) were given more influence in the
model fit.

      As soil sampling was optional in the HUD Grantees program, models (4) and (5) were
fitted to the HUD Grantees data both with and without soil-lead concentration included in the list
of predictor variables in the stepwise regression procedure. When fitting the model to HUD
                                          1-18
-------
Grantees data, separate intercepts (u) for the different grantees were included in the pool of
potential predictors but were not forced into the model. The stepwise procedure was allowed to
choose which grantees had significantly different intercepts from the others.

14.1.3 Performance Characteristics Analysis

       While the model-based analyses in Sections 14.1.1 and 14.1.2 can provide useful results,
these results may depend highly on the form of the model, the set of predictor variables included
in the model, and how these variables were defined and measured.  To reduce the level of
dependence that these factors may have on the outcome of these analyses, the non-modeling,
performance characteristics analysis approach documented in Section 6.1 of the §403 risk
analysis supplement report was also applied to data from the two studies. (See Section 6.1 for
details on the features of this approach.) Considering results of both this approach and the
model-based approach can provide a more complete perspective on findings to support the
analyses' common underlying objective to characterize the relationship between blood-lead
concentration and carpeted floor dust-lead loading and the need for a carpet dust-lead loading
standard.

       Of interest in the performance characteristics analysis was how the performance of a
given set of standards for lead in dust (uncarpeted floors and window sills) and soil might be
improved by adding a carpeted floor dust-lead loading standard. For example, performance
would improve if the carpet dust-lead loading standard triggers an intervention for some homes
containing children with elevated blood-lead concentrations that had not been previously
triggered by the other standards, while at the same time not triggering other homes that do not
contain elevated blood-lead children. The deteriorated lead-based paint standards in the §403
proposed rule were not considered in this analysis as no measurements were made in either study
that could be directly compared to these standards.

       In this analysis, the performance characteristics of the §403 proposed standards (dust and
soil) were initially calculated. Then, the change in performance when including a carpeted floor
dust-lead loading standard was evaluated for a range of such carpet dust-lead standards. The
candidate carpet standard that achieved the largest total of sensitivity, specificity, positive
predictive value (PPV), and negative predictive value (NPV) was then identified. However, the
individual characteristics were also of interest.  For example, if it is particularly important to
have few false positives (i.e., triggering homes that do not contain elevated blood-lead children),
then one would wish to maximize specificity. On the other hand, if a classification that results in
few false negatives is most desired (i.e., not triggering homes that contain elevated blood-lead
children), then one would maximize sensitivity. Plots of each of the four performance
characteristics and their total were provided to allow visual inspection of performance over a
range of candidate carpeted floor dust-lead loading standards.

       As discussed earlier, evaluating the need for a carpeted floor dust-lead loading standard
must also consider situations where housing units with only carpeted floors are encountered. To
evaluate the need for carpet dust-lead loading standards in this type of environment, it was

                                           1-19
-------
desired to perform the performance characteristics analysis on data for only those housing units
having exclusively carpeted floors. However, the two studies considered in these analyses did
not identify homes in this manner. While homes having floor-dust samples taken only from
carpets could be considered as an approximation, few such homes existed in either study.
Instead, the additional performance characteristics analyses were performed on all homes, but
floor dust-lead loading data were considered only for carpeted floors.  The results of these
analyses (which considered carpet, soil, and window sill dust standards) were then compared to
the results of analyses where carpet dust-lead was not considered (i.e., only soil-lead and window
sill dust-lead standards were considered) to determine if the addition of a carpeted floor standard
provided any performance benefit when floor dust sampling was assumed to be entirely from
carpeted floors.

       Note that while this performance characteristics analysis addressed the issue of the need
for a carpet dust-lead loading standard, it also addressed what this standard may be and whether
it should be different from the uncarpeted floor dust-lead loading standard of 50 jag/ft2 specified
in the §403 proposed rule. These latter areas are components of the second and third objectives
of this analysis, which are addressed further in Sections 14.2 and 14.3.

W.2   DETERMINING A CARPETED FLOOR DUST-LEAD
       LOADING STANDARD

       The results of applying the analysis method in Section 14.1.3 provide initial information
on objective #2, which was to consider appropriate candidates for carpeted floor dust-lead
loading standards, and in particular, whether the proposed uncarpeted floor dust-lead loading
standard of SO ug/ft2 should be considered a candidate standard. Applying the approaches
presented in this section provided additional information on addressing this objective. Three
approaches are presented:

       •     a comparison of average dust-lead loadings between carpeted and uncarpeted
             floors in the same housing unit, to determine whether the two averages within a
             home differ significantly (Section 14.2.1)

       •     regression modeling to predict the blood-lead concentration at which 95% of
             children are expected to be below at a given floor dust-lead loading, and how this
             blood-lead concentration differs when the dust-lead loading is assumed to be for
             carpeted versus uncarpeted floors (Section 14.2.2)

       •     performance characteristics analyses to evaluate a carpeted floor dust-lead loading
             standard whose performance was  similar to or better than that of the proposed
             standard for uncarpeted floors (Section 14.2.3).

In each of these three analyses, only data from the Rochester study were considered. As the
grantees participating in the HUD Grantees program evaluation targeted homes with children at
high risk for elevated blood-lead, applying these  analyses to the HUD Grantees data could yield

                                           1-20
-------
misleading conclusions when attempting to make inferences on the entire population based on
the results. In contrast, the Rochester study is at best representative of a typical urban population.

14.2.1 Comparing Average Dust-Lead Loadings Between
       Carpeted and Uncarpeted Floors in a Housing Unit

       In this analysis, average (wipe) dust-lead loadings between carpeted and uncarpeted floors
were compared within housing units having both types of floor surfaces.  A paired t-test was used
to make this comparison (i.e., a one-sample t-test on the differences between the log-transformed
area-weighted average floor dust-lead loadings for carpeted and uncarpeted floors within a unit).
This test determined whether the differences were significantly different from zero, or
equivalently, whether the geometric mean of the ratio of carpeted to uncarpeted (untransfonned)
area-weighted averages within a unit was significantly different from one. Non-significance
implied that (wipe) dust-lead loadings were similar between the two floor surfaces within a
housing unit, suggesting that a dust-lead loading standard for uncarpeted floors may be
reasonably implied, unchanged, to carpeted floors as well.

I4.2.2 Regression Modeling Approach

       In this analysis, model (1) of Section 14.1.1.2 was fitted to the Rochester study data to
predict blood-lead concentration as a function of average floor dust-lead loading for a given
surface type (carpeted, uncarpeted), with separate model fittings being performed for each
surface type. However, unlike the approach taken in Section 14.1.1.2, the observations included
in the model fittings were not weighted. As these model fittings were used to evaluate the need
for a separate dust-lead loading standards between carpeted and uncarpeted floors, an unweighted
analysis was used as such standards would be compared directly to a household average and not
to a weighted version.

       Within each regression model fitting, an upper 95% prediction bound on blood-lead
concentration was calculated over the range of average floor dust-lead loadings. Then, for a
given dust-lead loading, the blood-lead concentration was identified below which 95% of the
population of children exposed to that average dust-lead level would be expected to fall. The
results were compared between model fits (i.e., between carpeted and uncarpeted floors).  If the
bound on blood-lead concentration for carpeted floors using a standard of 50 ug/ft2 was not much
higher than the bound for uncarpeted floors using that same standard, then this provided evidence
that using this same standard for carpeted floor dust-lead loadings would be at least as protective
of children as the same standard for uncarpeted floor dust-lead loadings.

14.2.3 Performance Characteristics Analysis Approach

       The approach taken in this performance characteristics analysis is the same as that
documented in Section 14.1.3, but only average dust-lead loadings on carpeted or uncarpeted
floors were compared to candidate standards when determining whether an intervention was
triggered in a given housing unit (i.e., window sill dust-lead loadings and soil-lead concentrations

                                           1-21
-------
were not considered). The analysis calculated the four performance characteristics described in
Section 6.1 of the §403 risk analysis supplement report under a variety of alternative values of
the dust-lead loading standard for carpeted and uncarpeted floors. Each of the four
characteristics, as well as their total, were plotted versus the candidate floor dust-lead loading
standards to illustrate the differences in performance of candidate standards between carpeted
and uncarpeted floors. The goal was to identify a carpeted floor dust-lead loading standard
whose performance in this analysis was similar to or better than that of the proposed standard of
SO ug/ft2 for uncarpeted floors. In this way, similar levels of protection may be achieved by floor
dust-lead loading standards regardless of surface type.

14.3   DETERMINING AN APPROPRIATE METHOD FOR
       SAMPUNG CARPET DUST

       The dust-lead loading data analyzed by the methods in Sections 14.1 and 14.2 were for
samples collected using wipe techniques. However, other methods have been developed for
collecting dust samples as part of a risk assessment.  Different dust collection methods can
collect different types of dust samples containing different amounts of lead. This can have a
major effect on the observed relationship between dust-lead levels in the collected samples  and
blood-lead concentration. Therefore, objective #3 of this analysis was to investigate how the
effect of floor dust-lead levels on children's blood-lead concentration may depend on the dust
collection method being used and how the results differ between carpeted and uncarpeted floors.
This section documents the methods used to conduct statistical analyses on Rochester study and
HUD Grantees evaluation data in support of this objective. Other studies that have investigated
these issues and their findings have been documented in USEPA, 1997a.

       Floor dust-lead data for samples collected using the BRM vacuum, DVM vacuum, and
wipe techniques exist within the Rochester study database. For the HUD Grantees program
evaluation, only wipe dust-lead loading data were available for both carpeted and uncarpeted
floors, while very limited data on DVM dust-lead loadings for carpeted floors were collected.

14.3,1 Investigating the Association Between Floor Dust-Lead
       Levels and Blood-Lead Concentration for Different
       Sampling Methods

       Pearson correlation coefficients between average dust-lead levels and blood-lead
concentration were computed for BRM and DVM vacuum sampling and for wipe sampling, for
both dust-lead loading and concentration and for both carpeted and uncarpeted floors. Then,
univariate regressions of blood-lead concentration on average floor dust-lead, using model (1) of
Section 14.1.1.2, were fitted to data for all three dust collection methods according to each
combination of measurement type (loading, concentration) and surface type (carpeted,
uncarpeted). In the correlation and regression analyses, dust-lead data for a given household
were weighted by the percent of total floor sample area for the given dust collection method that
was carpeted (or uncarpeted, depending on the model fit).
                                          1-22
-------
14.3.2 Determining the Relationships of Average Dust-Lead
       Levels Between Sampling Methods

       This analysis investigated how dust-lead levels, as well as the relationship between dust-
lead loadings and concentrations, differed between dust collection methods and how these
comparisons differed between carpeted and uncarpeted floors. This analysis was performed only
on Rochester study data, as the HUD Grantees evaluation had virtually all carpet dust samples
collected via wipe methods.

       This analysis made statistical comparisons between the following pairs of dust-lead
measurements, with each comparison being done separately for carpeted and uncarpeted floors
(i.e., a total of 6x2=12 comparisons):

       •     Average BRM dust-lead loading versus average DVM dust-lead loading
       •     Average BRM dust-lead concentration versus average DVM dust-lead
             concentration
       •     Average BRM dust-lead loading versus average wipe dust-lead loading
       *     Average DVM dust-lead loading versus average wipe dust-lead loading
       •     Average BRM dust-lead loading versus average BRM dust-lead concentration
       •     Average DVM dust-lead loading versus average DVM dust-lead concentration

Each comparison consisted of plotting the data, then calculating Pearson correlation coefficients
on the log-transformed data to evaluate the linear relationship between the two (log-transformed)
measurements.  When calculating the Pearson correlation coefficients, each data point was
weighted by the proportion of total floor area in the housing unit sampled by the given dust
collection methods that corresponded to the particular surface type (carpeted or uncarpeted). For
example, when  calculating the correlation coefficient between BRM and DVM carpet dust-lead
loadings, each data point was weighted by the proportion of total floor area sampled by the BRM
and DVM that was carpeted. Each calculated correlation coefficient was tested for significant
difference from zero. The results for carpeted surfaces were then compared to those for
uncarpeted surfaces.

14.3.3 Investigating the Relationship in Lead Loadings of Side-by-Side
       Dust Samples Collected by  Different  Methods

       The Rochester study sampling design included taking dust samples from three adjoining
(side-by-side) areas, where each dust collection method (BRM, DVM, wipe) was used to collect
one of the three samples, hi this analysis, it was of interest to determine how measured dust-lead
loadings differed among side-by-side samples (and, therefore, among different dust collection
methods). This comparison was based on within-location variability (as well as sampling and
analysis variability), as opposed to the unit-to-unit variability used to make comparisons in the
analyses described in the previous subsections.  The analysis was done on data for carpeted
surfaces and uncarpeted surfaces separately, allowing for comparisons between the two surface
                                          1-23
-------
types.  This analysis was performed only on Rochester study data, as the HUD Grantees program
evaluation did no side-by-side sampling.

       In the Rochester study, floor-dust samples were identified according to the room in which
they were collected and the collection method used; the dust samples within a room were
assumed to be collected from adjacent, side-by-side areas. The lead loading data for these
samples were used in fitting the following regression model to predict the dust-lead loading
under one dust collection method (method A) as a function of die loading under a second method
(method B):
                    log(PbDAjp   =  \i + c^logtPbDBjp  + Ht + e-                   (6)
where PbDAy is the dust-lead loading for the floor-dust sample collected by method A in the jth
room within the ith housing unit, PbDE^ is the dust-lead loading for the floor-dust sample
collected by method B at the jth room within the ith housing unit, and Hj is the random effect of
the ith housing unit on PbDAjj. Thus, model (6) was used to predict the dust-lead loading for a
sample under one collection method as function of the observed dust-lead loading for the
adjacent sample of another collection method.  The model controls for two types of variation:
variation due to sampling in different housing units, and variation due to sampling in different
rooms within a housing unit As it was desired to express the dust-lead loading under one
method as a function of the observed dust-lead loading of another method, the model fitting did
not adjust for measurement error in the dependent variable.

       For every dust collection method that was assigned as method A, model (6) above was
fitted four times, once for each combination of surface type (carpeted floors, uncarpeted floors)
and for the remaining two dust collection methods that could be assigned as method B.

       In model (6) above, the intercept ji represents a constant underlying multiplicative bias in
the results of the two collection methods, while the slope o represents the extent to which the
bias is constant across the range of loadings. Intercepts significantly different from zero suggest
the presence of a bias, while slopes significantly different from one suggest that the bias changes
with the magnitude of the measurements. Therefore, the estimates of the intercept and slope
parameters are reported for each model fitting, as well as results of significance tests.

       A more statistically rigorous procedure for converting dust-lead loadings from one dust
collection method to another is found in USEPA, 1997c.

15.0   RESULTS

       Detailed results of the statistical methods documented in Section 14 as applied to data
from the Rochester study and the HUD Grantees program evaluation (Section 13) are presented in
this section. To allow the reader to easily refer to details on the statistical methods behind a
particular set of results, the sections and subsections within this section are titled and organized

                                          1-24
-------
in the same way as in Section 14, where the methods were presented. Each subsection (Sections
15.1 through 15.3) corresponds to one of the three appendix objectives presented in Section 11.2.
Conclusions made from these results are found in Section 6.5 of the §403 risk analysis
supplement report.

       Note that individual results presented hi this section may differ from similar results
presented in previously-published documents on these two studies. This is due to differences in
the statistical methods used in this appendix, in the subsets of data included in the analysis, and
hi any transformations and summary calculations performed on the data prior to analysis.

       Descriptive statistics of the data analyzed hi this section are presented hi Appendix 12.

15.1   ASSESSING THE NEED FOR A CARPETED FLOOR
       DUST-LEAD STANDARD

See Section 14.1 and its subsections for details on the statistical methods associated with the
results presented in this section.

15.1.1 Investigating the Association Between Dust-Lead  Loading and
       Blood-Lead Concentration for Carpeted and Uncarpeted Floors

       15.1.1.1.  Correlations Between Floor Dust-Lead Loading and Blood-Lead
Concentration. Figure 15-1 contains four plots, each depicting blood-lead concentration versus
household average (wipe) floor dust-lead loading for each combination of surface type (carpeted,
uncarpeted) and study.  Each point within the plots represents a single housing unit.

       The plots hi Figure 15-1 show some positive correlation between dust-lead loadings and
blood-lead concentration, but the level of variability hi these relationships is high for both studies
and surface types.

       For each plot in Figure 15-1,  a Pearson correlation coefficient was calculated on the data
hi the plot to quantify the extent of a linear relationship between log-transformed blood-lead
concentration and log-transformed average floor dust-lead loading. The correlation coefficients
for each study and particular surface type (carpet, uncarpeted) are presented in Table 15-1. This
table indicates the following:

       •      For the Rochester study, statistically significant correlation was observed at the
              0.01 level between blood-lead concentration and average dust-lead loading when
              sampling from uncarpeted floors and at the 0.05 level when sampling from
              carpeted floors.

       •      For the HUD Grantees program evaluation, statistically significant correlations
              were observed at the  0.01 level between blood-lead concentration and average
              dust-lead loading when sampling from both carpeted and uncarpeted floors.

                                           1-25
-------
                   10     100     1000    10000  100000
                                                       100.0
                                                         0.1      I      10     100    1000    10000  100000
                                                                          puil-LMd loading
                                                                       HUO -
   100
                   10     100     1000    10000  100000
                     Otat-LMd Loading
                                                       10.0
                                                                     a a o ttftoo a
                                                                       10     100    1000    10000  10OOOO
Figure 15-1.   Plots of Blood-Lead Concentration versus Household Average (Wipe) Floor
                Dust-Lead Loading, for Each Combination of Floor Surface Type (Carpeted,
                Uncarpeted) and Study

Table 15-1.    Pearson Correlation Coefficients of Log-Transformed Average (Wipe) Dust-
                Lead Levels with Log-Transformed Blood-Lead Concentration, for Carpeted
                and Uncarpeted Floors
Surface Type • ;
Carpeted Floors '
Uncarpeted Floors ^
•;.; ••• ';•" Rochester Study : =:
0.190* (179)
0.313** (193)
HUD Grantees Program Evaluation
0.308** (226)
0.335** (390)
1 Correlation coefficients are calculated on unit-wide area-weighted average dust-lead loadings, where averages are taken
across all samples in a housing unit of the given surface type (carpet or non-carpet). The average for a given housing unit is
weighted by the proportion of total floor wipe sample area in the unit represented by carpeted (uncarpeted) surfaces in
calculating the correlation coefficient for carpeted (uncarpeted) floors.

* Significant at the 0.05 level.
** Significant at the 0.01 level.
                                                1-26
-------
Results in Table 15-1 differ slightly from correlation coefficients reported in the Rochester study
report (the Rochester School of Medicine and NCLSH, 1995), primarily due to the form of the
dust-lead parameter (this analysis used a log-transformed weighted arithmetic average of
untransformed data, while the Rochester study report used an untransformed unweighted average
of log-transformed data).

        15.1.1.2.  Univariate Regression of Blood-Lead Concentration  on Floor Dust-lead
Loading. To further investigate the relationship between floor dust-lead loading and blood-lead
concentration, model (1) in Section 14.1.1.2 was fitted separately to each set of data determined
by the four plots in Figure 15-1. Table 15-2 presents the estimated slope and intercept terms for
the two model fits to the Rochester data, and the estimated slope terms for the two model fits to
the HUD Grantees evaluation data. (Recall that the latter two model fits had grantee-specific
intercepts, whose estimates are not included in Table 15-2). Table 15-2 also includes the standard
errors associated with each estimate. The column marked "baseline" hi Table 15-2 is the
exponentiation of the intercept term (for the Rochester study data fits) and represents a baseline
geometric mean blood-lead concentration before any floor dust-lead effects impact the value.
Statistically significant slope estimates (denoted by asterisks in Table 15-2) imply that the
predictor variable is significantly associated with blood-lead concentration.
Table 15-2.   Estimates of Intercept and Slope Parameters (and their Standard Errors)
               Associated With Regression Models That Predict Blood-Lead Concentration
               Based on Average (Wipe)  Floor Dust-Lead Loading
Study
Rochester Study '
HUD Grantees
Program
Evaluation 2

Floor Surface
. Type- ,
Carpeted
Uncarpeted
Carpeted
Uncarpeted
Number of
.. Units.
179
193
226
390
Estimates (Standard Errors)
, Intercept (//)
., f £ ~
1.53(0.11)
1.39(0.12)
Baseline >
(e°;^g/dL)
4.61
4.03
. '••• • • •". • -:;" ' .-..•- :•
. , . Slope (a)
0.103* (0.040)
0.1 74* •(0.038)
0. 1 60* • (0.048)
0.117** (0.030)
 1 The regression model takes the form logfPbB,) = fJ + adogtPbD,)) + ef, or equivalently, PbB, =expb) x (PbDJa x expfe,),
 where PbB, is the blood-lead concentration for the child in the ith housing unit, e, refers to the random error associated with
 the model-based blood-lead concentration for the ith unit, and remaining notation is specified in the column headings. For a
 specific surface type, results for the ith unit are weighted by the proportion of total floor sampling area represented by the
 given surface type.

 1 The regression model takes the form log(PbB,) = y, + a* log (Potty + es, where  PbB, represents the blood-lead
 concentration for the selected child in the ith housing unit within the jth grantee, PbD5 corresponds to the observed average
 floor dust-lead loading for the ith housing unit within the jth grantee (for the given  surface type), and a and YI are parameters
 representing the slope of the model and the intercept for the jth grantee, respectively. The residual error left unexplained by
 the model is denoted by e,. The model is weighted by the proportion of total floor sampling area represented by the given
 surface type.

 *  Significantly different from zero at the 0.05 level.
 ** Significantly different from zero at the 0.01 level.
                                              I-27
-------
       Results from Table 15-2 are as follows:

       *      For each model fit, the slope estimate was positive and statistically different from
              zero at the 0.05 level, implying that increased blood-lead concentrations were
              significantly associated with increased values of the dust-lead predictor variable.

       »      For the Rochester study, dust-lead loadings were significant predictors of blood-
              lead concentration for carpeted floors at the 0.05 level (p-value = 0.0110) and for
              uncarpeted floors at the 0.01 level (p-value £ 0.0001).

       *      For the HUD Grantees program evaluation, dust-lead loadings were significant
              predictors of blood-lead concentration at the 0.01 level for both carpeted (p-value
              = 0.0010) and uncarpeted (p-value s 0.0001) floors.

Therefore, the results of this analysis indicate that dust-lead loadings from both carpeted and
uncarpeted floors are statistically significant predictors of blood-lead concentration, in the
absence of other potentially significant (and possibly confounding) predictors. The same
conclusion holds whether one considers data from the Rochester study or the HUD Grantees
program evaluation.

       15.1.1.3. Comparing  the Dust-Lead Loading/Btood-Lead Concentration Relationship
Between  Homes With Mostly Carpeted Floors and Homes With Mostly Uncaroeted Floors.
To illustrate whether the relationship between blood-lead concentration and floor dust-lead
loading differs significantly between homes that are mostly carpeted (i.e., more than 50% of the
total floor area wipe-sampled for dust is carpeted) and homes that are mostly uncarpeted, Table
15-3 presents the results of fitting model (2) of Section 14.1.1.3 according to the procedures
specified in that section. Recall from Section 14.1.1.3 that the dust-lead loading variable in model
(2) had one of two possible definitions: the average floor dust-lead loading based on samples
taken only from the surface type with the higher total sample area ("surface majority"), and a
weighted average of the average carpeted and uncarpeted floor dust-lead loadings ("weighted
average").

       The key results in Table 15-3 are found within the columns labeled "P0" and "P,", as these
model parameters represent whether the intercept and slope parameters in the model differ
between homes having floor dust samples collected from mostly carpeted floors and homes
having floor dust samples collected from mostly uncarpeted floors.  Note that none of the rows of
Table 15-3 indicate that the estimates of P0 and pt are significantly different from zero. The
results in Table 15-3 suggest that for each study, regardless of whether the floor dust-lead loading
variable follows the "surface majority" or "weighted average" definition in this analysis, there is
no statistically significant difference in the relationship between blood-lead concentration and
average floor dust-lead loading between houses with mostly carpeted floors and houses with
mostly uncarpeted floors.  This supports the hypothesis that carpeted and uncarpeted floor dust-
lead loadings predict blood-lead concentration in a similar manner.
                                           1-28
-------
Table 15-3.   Estimates of Intercept and Slope Parameters (and Their Standard Errors)
               Associated With Fitting Model (2) to Predict Blood-Lead Concentration
               Based on an Average (Wipe) Floor Dust-Lead Loading Which Emphasizes the
               Roor Surface Type With the Larger Sample  Area
Study
Rochester
Study '
HUD
Grantees
Program
Evaluation 2
Definition
of:Dust-:
Lead
Variable
Surface
Majority
Weighted
Average
Surface
Majority
Weighted
Average
#- :
Units
149

363
Estimate- (Standard Error) - ~ - __ •" ~
Intercept (fi)
1.627"* (0.262)
1.538" (0.274)

* -Change'in .
• Intercept, for
Units Having > '
50% FTopr:Dust
Samples from
Carpets (W
-0.281 (0.335)
-0.314 (0.353)
-0.111 (0.264)
-0.063 (0.271)
Slope (a)
*"*
0.137(0.078)
0.1 70* (0.085)
0.1 24" (0.034)
0.1 35'* (0.037)
•^"r A V ** >" ^r
| Change in Slope
for Units Having
7 >; 50% Boor-,
; - Oust Samples.
^from Carpets IB,)
0.025(0.108)
0.036(0.116)
0.057 (0.082)
0.032 (0.084)
' The regression model takes the form loglPbB,) = // + aplog(PbDjp) + p\,*SURF, + P,*1og(PbDi*)*SURFi + e,, where
is the blood-lead concentration for the child in the ith housing unit, PbDj* is the dust-lead loading variable as defined in
Section I4.1.1.3 in the ith unit, SURF, equals one if floors were sampled mostly from carpets in the ith unit, and zero if floor-
dust sampling was mostly from uncarpeted surfaces, e, refers to the random error associated with the model-based blood-
lead concentration for the ith unit, and remaining notation is specified in the column headings.

2 The regression model takes the form loglPbB,) = v, + a*log(PbD,*> +  Bo* SURF, + pMlog(PbO,*)*SURF; + e, where PbB,
represents the blood-lead concentration for the selected child in the ith housing unit within the jth grantee, PbD,*
corresponds to the observed floor dust-lead loading as  defined in Section 14.1.1.3 for the ith housing unit within the jth
grantee, and a and v, are parameters representing the slope of the model and the intercept for the jth grantee, respectively.
The residual error left unexplained by the model is denoted by e,. SURF, equals one if floors were sampled  mostly from
carpets, and zero  if floor-dust sampling was mostly from uncarpeted surfaces in the ith unit within the jth grantee.
Remaining notation is specified in the column headings.

* Significantly different from zero at the 0.05 level.
**  Significantly different from zero at the 0.01 level.
        15.1.1.4.  Investigating the Association Between Floor Dust-Lead Loading and
Blood-Lead Concentration, Controlling for Demographic Variables. The previous sections
investigated the association between floor dust-lead loading and blood-lead concentration
without considering the effects on blood-lead concentration of other potentially influential
variables. In this section, model (3) from Section 14.1.1.4 was fitted to the study data, which
extends model (1) used to generate the results in Section 15.1.1.2 above by adding other
potentially influential demographic variables as predictor variables using stepwise regression
techniques.  The effect of average dust-lead loading on blood-lead concentration was assessed
only after taking into account the effects of these other demographic variables (which do not
represent the set of all such important variables).  See Table 14-1  for a listing and definitions of
the demographic variables considered in this analysis.
                                               1-29
-------
       Tables 15-4 and 15-5 present the results of fitting model (3) to the Rochester study data
and the HUD Grantees program evaluation data, respectively. The tables list those demographic
variables from Table 14-1 that were selected for the model due to having significant effects on
blood-lead concentration data, along with their corresponding slope estimates.  The slope
estimates corresponding to average dust-lead loading is in the last row of these tables, as this
variable was added last to model (3).

       Both analyses concluded that regardless of whether carpeted or uncarpeted floors were
being considered, average floor dust-lead loading was a statistically significant predictor of
blood-lead concentration even after adjusting for other important demographic variables, with an
increase in floor dust-lead loading associated with an increase in blood-lead concentration. Other
findings when analyzing the Rochester study data (Table 15-4) included the following:

       •       The race, sex, and education variables (Table 14-1) were statistically significant
              predictors of blood-lead concentration.

       •       When dust-lead loadings from only carpeted floors were considered, mouthing
              behavior (putting mouth on window sill, use of pacifier, soil pica, thumb-sucking)
              was a statistically significant predictor of blood-lead concentration, with a greater
              propensity of mouthing behavior corresponding to higher blood-lead
              concentration.
       *      When dust-lead loadings from only uncarpeted floors were considered, paint/pica
              hazard was a statistically significant predictor of blood-lead concentration with a
              larger potential for paint pica hazard corresponding to higher blood-lead
              concentration.

Other findings when analyzing the HUD Grantees evaluation data (Table 15-5) included the
following:

       •      More differences among the grantee-specific intercepts were observed when dust-
              lead loadings were considered for uncarpeted floors versus carpeted floors.  Note,
              however, that the model fitting which considered carpeted floor dust-lead loadings
              involved data for 161 fewer housing units, as some grantees had few or no
              carpeted floor dust-lead loading data.

       •      When dust-lead loadings from only carpeted floors were considered, the only
              significant demographic variable other than grantee differences was the
              seasonably variable, with measurements in spring and summer associated with
              larger values of blood-lead concentration.

       •      When dust-lead loadings from only uncarpeted floors were considered, income,
              race, and mouthing behavior were found to be statistically significant predictors of
              blood-lead concentration.

                                           1-30
-------
Table 15-4.   Parameter Estimates (and their Standard Errors) Associated With Fitting
              Model (3) to Rochester Study Data to Predict Blood-Lead Concentration
              Based on Average Floor Dust-Lead Loading
1 • " ~ Carpeted ROOT, dust . ; ~ - • ;•
Parameter

•Intercept
Race
Sex
Education
Mouthing Behavior
r
Log Floor Dust-Lead
Loading
Estimate (Stfl»*cif m )
'"' * '" Parfimnl

1.843(0.121)
-O.430 (0.089)
-0.614(0.154)
-0.300 (0.088)
0.536 (0.262)
\ -
0.087 (0.034)
P-vakie
— - ~w?

'-,„ »„- •
Parameter


$0.0001
sO.0001
sO.0001
0.0009
0.0428
Intercept
Race
Sex
Education
Paint Pica Hazard
•j ParamotBf Addod L&st

0.0117
R2 of final model: 0.334
Number of data points (housing units): 1 76
Log Floor Dust-Lead
Loading
Estimate. {Strivi. Error }
*,
1.787(0.133)
-0.322(0.101)
-0.513 (0.194)
-0.188 (0.1 OO)
0.441 (0.152)
' ' -'-V
0.101 (0.037)
"4 ~-
P-value
-," •
£0.0001
0.0018
0.0091
0.0626
0.0042
'-
0.0065
R2 of final model: 0.277
Number of data points (housing units): 192
(see footnote below)

Table 15-5.   Parameter Estimates (and their Standard Errors) Associated With Fitting
              Model (3) to Data from the HUD Grantees Program Evaluation to Predict
              Blood-Lead Concentration Based on Average Floor Dust-Lead Loading
Carpeted Floor Dust . - -
_.- Parameter

Intercept
California
Cleveland
New York City
Minnesota
.Season
Estimate (Stdf Error) ••,
P-vahiB:


1.628(0.163)
-0.730 (0.259)
0.326(0.133)
-0.400 (0.218)
•0.348 (0.120)
0.217I0.1O4)
*0.0001
0.0052
0.0152
0.0673
0.0042
0.0378


Log Floor Dust-
Lead Loading
0.160(0.046)


0.0006
R2 of final model: 0.246
Number of data points (housing units): 226
. ' • Uncarpeted Floor' Dust
Parameter " •
r Stepwise Regression
Intercept
California
Cleveland
New York City
Alameda County
Baltimore
Vermont
Income
Race
Mouthing
UdedLast
Log Floor Dust-
Lead Loading
Estimate (Std. Error)
11 • •

1.83(0.131)
-0.899(0.174)
0.231 (0.136)
-0.597(0.141)
-0.505(0.116)
-0.225 (0.108)
0.518 (0.218)
-0.180(0.071)
-0.317(0.091)
0.191 (0.091)
'::• ' ' 	 •
0.110(0.029)
P-value .

£0.0001
sO.0001
0.0896
sO.0001
sO.0001
0.0375
0.0180
0.0123
0.0005
0.0364

0.0002
R1 of final model: 0.29O
Number of data points (housing units): 387
1 Parameters are accepted into the model with a significance level (adjusted for other terms in the model) of 0.10 or lower
and are removed from the model when their significance level (adjusted for other terms in the model) is higher than 0.10.
                                            1-31
-------
Note that these analyses ignored the contribution to the prediction of blood-lead concentration
made by other environmental-lead variables such as soil-lead concentration and window sill dust-
lead loading. The next section will address effects in the presence of these additional variables.

15.1.2 Investigating the Association Between Carpeted Floor Dust-Lead
       Loading and Blood-Lead Concentration, Controlling for Other
       Environmental-Lead Sampling
                                                                                     •
       To investigate the contribution that average carpeted floor dust-lead loading may have on
predicting blood-lead concentration, over and above the contributions of the lead measures
(uncarpeted floor dust, window sill dust, soil-lead concentration) that can be compared to the
current §403 standards, models (4) and (5) of Section 14.1.2 were fitted to the Rochester and
HUD Grantees data. As described in Section 14.1.2, stepwise regression procedures were used to
select predictor variables, with the candidate predictor variables corresponding to uncarpeted
floor dust-lead loading, window sill dust-lead loading, soil-lead concentration, and paint pica
hazard for model (4), and these variables plus the demographic variables in Table 14-1 for model
(5).  Once these other variables were selected for the model, the carpeted floor dust-lead loading
variable was added to the model. Data for only those housing units having floor dust-lead
loading data for both carpeted and uncarpeted surfaces were included in this analysis.

       Tables I5-6a and I5-6b present the results of fitting models (4) and (5), respectively, to
data from the Rochester study. According to these tables, once the effects of other important
factors were accounted for in both models, the additional effect of average carpeted dust-lead
loading on blood-lead concentration was not statistically significant. (Both p-values were •
considerably higher than 0.10.) ID contrast, soil-lead concentration  and uncarpeted dust-lead
loadings had highly significant effects on blood-lead concentration in both model fits.

       Tables I5-7a and I5-7b present the results of fitting models (4) and (5), respectively, to
data from the HUD Grantees program evaluation. Recall that since soil sampling was optional in
this evaluation, the models were fitted both with and without considering soil-lead concentration
as a candidate predictor variable. In contrast to the findings of the Rochester data analysis
{Tables I5-6a and I5-6b), once the effects of other important factors (including soil-lead
concentration) were accounted for in the models, the additional effect of average carpeted dust-
lead loading on blood-lead concentration was significant at the 0.05 level. When soil-lead
concentration was excluded from the models, the additional effect of average carpeted dust-lead
loading on blood-lead concentration achieved statistical significance at the 0.10 level but not at
the 0.05 level.

       Thus, the analyses involving models (4) and (5) provide disparate results between the two
studies concerning the significance of any added effect that carpeted floor dust-lead loading may
have on blood-lead concentration once the effects of other important environmental-lead and
demographic predictors have been taken into account.  While this may suggest that the role of
lead in carpet dust on increased blood-lead concentration in children may be marginal, one must
                                           1-32
-------
Table !5-6a.  Parameter Estimates (and their Standard Errors) Associated With Fitting
               Model (4) to Rochester Study Data to Predict Blood-Lead Concentration
               Based on Average Carpeted Floor Dust-Lead  Loading After Adjusting for
               Other Environmental Sampling
Parameter ~ < ,
x. Estimate; (Standard Error)
.. '-- ,T»-value?^
. Parameters Selected by Stepwise Regression*
Intercept
Log Soil-Lead Concentration
Log Window Sill Dust-Lead Loading
Log Floor Dust-Lead Loading (Uncarpeted)
Paint Pica Hazard
0.371 (0.251)
0.107 (0.038)
0.074 (0.037)
0.257 (0.064)
0.372 (0.167)
0.1417
0.0052
0.0486
0.0001
0.0271
• Parameter Added Last - '•-„--
Log Floor Dust-Lead Loading (Carpeted)
0.015(0.059)
0.7938
R2 of final model: 0.287. Number of data points (housing units): 152
1 A p-value of 0.0001 indicates a p-value of s. 0.0001.
1 Parameters were accepted into the model with a significance level (adjusted for other terms in the model) of 0.10 or lower
and were removed from the model when their significance level (adjusted for other terms in the model) exceeded 0.10.
Table !5-6b.  Parameter Estimates (and their Standard Errors) Associated With Fitting
               Model (5) to Rochester Study  Data to Predict Blood-Lead Concentration
               Based on Average Carpeted Floor Dust-Lead Loading After Adjusting for
               Other Environmental and Demographic Variables
- Parameter '.'"..
: - i Par amtf*tprg l ^gilo/H

Intercept
Log Soil-Lead Concentration
Log Floor Dust-Lead Loading (Uncarpeted)
Race
Paint/Pica Hazard
Age4
Estimate (Standard Error) [ P-value1
ted by Stepwise Regression2
0.624 (0.267)
0.117(0.033)
0.223 (0.053)
-0.441 (0.077)
0.243(0.156)
0.178 (0.086)
0.0207
0.0004
0.0001
0.0001
0.12163
0.0411
..; •' ' ••>•:;;•• • i'i'.. Parameter Added Last ; . :::. •;•;; . •• -:v,.;.
Log Floor Dust-Lead Loading (Carpeted)
0.037 (0.050) | 0.4657
R2 of final model: 0.399. Number of data points (housing units): 157
1 A p-value of 0.0001 indicates a p-value of * 0.0001.
1 Parameters were accepted into the model with a significance level (adjusted for other terms in the model) of 0.10 or lower
and were removed from the model when their significance level (adjusted for other terms in the model) exceeded 0.10.
3 These variables had a p-value £ 0.10 prior to adding Log Floor Dust-Lead Loading (Carpeted}), but their p-value exceeded
0.10 when Log Floor Dust-Lead Loading (Carpeted) was added to the model and when age was added to the model rather
than age-squared.
* The Stepwise procedure chose age-squared rather than age, but age was added to the model instead.
                                               I-33
-------
Table !5-7a.  Parameter Estimates (and their Standard Errors) Associated With Fitting
               Model (4) to HUD Grantees Evaluation Data to Predict Blood-Lead
               Concentration Based on Average Carpeted Floor Dust-Lead Loading After
               Adjusting for Other Environmental Sampling
Soil-Lead Concentration Included as a Possible
Predictor Variable
Parameter
Estimate (Std. Error)
P-value
- .Soil-Lead Concentration Excluded as a, Possible :: ,
- . J Predictor Variable '"^- ".*%?;,
: . Parameter ,
Estimate {Std. Error)
P-value
. Parameters Selected By Stepwise Regression1 •'
Intercept
Log Soil-Lead
Concentration
0.091 (0.711)
0.288 (0.099)
0.8985
0.0061
"
Cleveland
Minnesota
0.037 (0.290)
0.215(0.311)
0.8999 2
0.4932 *
" - " , • '/ - '.'-„'•'
Intercept
&
Log Window
Sill Dust-Lead
Loading
Log Floor Dust-
Lead Loading
(Uncarpeted)
California
Cleveland
1 .440 (0.234)
- -s_ ';'• -
0.031 (0.031)
0.133(0.055)
-0.784 (0.264)
0.479 (0.144)
^0.0001
"
0.3265 2
0.0167
0.0034
0.0011
" ''T-.P- :..:'::v:S'- • ^-igss-' '•
New York City
-0.204 (0.245)
0.4044 2
• ' ' •'' '., 'Parameter Added Last, .:-' -.'.'.. ./:'';%•' ! :-?:?£f:> :''^'p£ J'
Log Floor Dust-
Lead Loading
(Carpeted)
0.215(0.093)
0.0260
R2 of final model: 0.330
Number of data points (housing units): 42
Log Floor Dust-
Lead Loading
(Carpeted)
0.143 (0.074)
0.0541
R2 of final model: 0.1 80
Number of data points (housing units): 220
1 Parameters were accepted into the model with a significance level (adjusted for other terms in the model) of 0.10 or lower
and were removed from the model when their significance level (adjusted for other terms in the model) exceeded 0.10.
2 These variables had a p-value & 0.10 prior to adding Log Floor Dust-Lead Loading (Carpeted)), but their p-value exceeded
0.10 when Log Floor Dust-Lead Loading (Carpeted) was added to the final model.
                                              1-34
-------
Table !5-7b.  Parameter Estimates (and their Standard Errors) Associated With Fitting
               Model (5) to HUD Grantees Evaluation Data to Predict Blood-Lead
               Concentration Based on Average Carpeted Floor Dust-Lead Loading After
               Adjusting for Other Environmental and Demographic Variables
Soil-Lead Concentration Included as a Possible .
Predictor Variable
- Parameter
Estimate (Std: Error)
: P-valiie .
- SoB-Lead Concentration Excluded as a Possible
L „- . -•_-, Predictor .Variable ;, '"?-'
Parameter
Estimate (Std. Error)
P-value
/ - Parameters Selected By Stepwise Regression1
Intercept
Log Soil-Lead
Concentration

Cleveland
Minnesota
-0.174(0.792)
0.298 (0.100)
..',•••. ' .-;,••- 	 !: ' "• >:. : -j.-'-r::-1
• ' : .-""; • ' • - ::-:.r:., =:- -'"-i- '•'•••"•'•• '"'-:"?
0.8277
0.0053
;';;;v;::Tij;c^;-:M- ' ••
=r"' . • :::•'.*'""!'" '. ' '
;;fe::;;;||i|||;;
'•.:.' : : '•'.•• • .. •' •-.':.-..' i'1"'"" " •••- '•" '
0.103(0.304)
0.275 (0.322)
0.7377 2
0.3982 2
- -- ,-• •
Mouthing
0.222 (0.286)
0.4425 2
.
Intercept
1.770(0.274)
sO.0001
, "' ' ' " -I", ."* . ~ -
Log Window
Sill Dust-Lead
Loading
Log Floor Dust-
Lead Loading
(Uncarpeted)
California
Cleveland
0.023 (0.032)
0.117(0.056)
-0.777 (0.266)
0.421 (0.149)
0.4657 3
0.0382
0.0039
0.0053
•••:Y:'''-. !'" '•'.... :"\ ' • .^::::;' ; • ;i;!;J: ;- .'.:!-:i'':l'.:;:"-v' V :
New York City
Rhode Island
Vermont
-0.270(0.251)
0.368 (0.220)
0.374 (0.362)
0.2827 2
0.0961
0.3030 2
"•.'•",.' " . /<'". ~ .•
Income
Race
Age3
-0.119(0.110)
-0.241 (0.126)
-0.044 (0.035)
0.2806 2
0.0576
0.2181 2
Parameter Added Last . • '. • ..•.-.,' "•: /o''.'- ;. *••;.••
Log Floor Dust-
Lead Loading
(Carpeted)
0.203 (0.094)
0.0379
R2 of final model: 0.342
Number of data points (housing units): 42
Log Floor Dust-
Lead Loading
(Carpeted)
0.137(0.074)
0.0663
R2 of final model: 0.21 3
Number of data points (housing units): 218
1 Parameters were accepted into the model with a significance level (adjusted for other terms in the model) of 0.10 or lower
and were removed from the model when their significance level (adjusted for other terms in the model) exceeded 0.10.
2 These variables had a p-value £ 0.10 prior to adding Log Floor Dust-Lead Loading (Carpeted)), but their p-value exceeded
0.10 when Log Floor Dust-Lead Loading (Carpeted) was added to the model and when age was added to the model rather
than age-squared.
3 The stepwise procedure chose age-squared rather than age. but age was added to the model instead.
                                               1-35
-------
keep in mind that differences in the types and definitions of variables measured between the two
studies (i.e., candidates for predictor variables in these models) also play a key role in the
outcome of the model fits.

15.1.3 Performance Characteristics Analyses

       As discussed in Section 14.1.3, the results presented in this subsection are based on a non-
modeling analysis approach whose objective was to evaluate the need to add a carpet (wipe)
dust-lead loading standard to the set of dust and soil standards in the §403 proposed rule (i.e., SO
ug/ft2 for uncarpeted floors, 250 jig/ft2 for window sills, 2000 ppm for soil), and to investigate
possible recommended values for such a standard.  Section 6.1 of the §403 risk analysis
supplement report defines the four performance characteristics (sensitivity, specificity, positive
predictive value, negative predictive value) which were the focus of this analysis and how they
are calculated and interpreted.

       The four performance characteristics (expressed as percentages) were calculated over a
range of candidate carpet dust-lead loading standards from 0 to 100 ug/ft2, where the carpet
standard  was added to the set of dust (uncarpeted floors, window sills) and soil standards from
the §403  proposed rule. (Recall that the proposed paint standards were not considered in this
analysis.) The results are plotted as "performance curves" within Figures 15-2 (based on
Rochester study data) and 15-3 (based on HUD  Grantees evaluation data). These two figures
each contain six plots: one for each of the four performance characteristics, one for the sum of
the four performance characteristics, and one containing the four performance characteristics
superimposed on the same plot. (The vertical axis labels distinguish the plots from each other.)

       Each plot in Figures 15-2 and 15-3 contains a horizontal dashed line which denotes the
calculated value of the given performance characteristic when no candidate carpet dust-lead
loading standard is considered. When the performance curve lies above this horizontal dashed
line, this implies that any of the corresponding values of the carpet dust-lead loading standards,
when added to the set of dust and soil standards in the §403 proposed rule, would result in a
higher value of the given performance characteristic, and therefore, improved performance based
on this performance criterion. Each plot in Figures 15-2 and 15-3 contains a vertical dashed line at
50 Hi/ft2 (i.e., the proposed standard for uncarpeted floors) to illustrate the value of the
performance characteristic if both the carpeted and uncarpeted floor dust-lead loading standards
were set equal to 50 ug/ft2. An additional vertical dashed line is provided at the candidate carpet
dust-lead loading standard that leads to the maximum value of the sum of the four performance
characteristics: 17 ug/ft2 based on analysis of the Rochester study data (Figure 15-2) and 5 ug/ft2
based on analysis of the HUD Grantees evaluation data (Figure 15-3), thereby representing a
possibly  "optimal" value for the standard. An additional vertical dashed line is provided at 13
      within the plots in Figure 15-3, for reasons to be discussed later in this section.
                                           1-36
-------
i «
  *>
                                              f
                   «
                  floor O
Figure 15-2.   Values of the Performance Characteristics As a Function of Candidate
              Carpeted Floor Dust-Lead Loading Standards, Based on Analysis of the
              Rochester Study Data, Where the Set of Standards Also Includes the
              Uncarpeted Floor, Window Sill, and Soil Standards Proposed in the S403
              Proposed  Rule

(See text for the connotations of the horizontal and vertical dashed lines in these plots. The §403 proposed
standards were 50jt/g/ft2 for uncarpeted floors, 250/yg/ft2 for window sills, and 2000 ppm for soil.)
                                           1-37
-------
                                             I
I
       «   a
Figure 15-3.  Values of the Performance Characteristics As a Function of Candidate
             Carpeted Floor Dust-Lead Loading Standards, Based on Analysis of the HUD
             Grantees Evaluation Data, Where the Set of Standards Also Includes the
             Uncarpeted Floor, Window Sill, and Soil Standards Proposed in the §403
             Proposed Rule

(See text for the connotations of the horizontal and vertical dashed lines in these plots. The S403 proposed
standards were 50 pg/ft2 for uncarpeted floors, 250 //g/ft2 for window sills, and 2000 ppm for soil.)
                                          1-38
-------
       Note that in Figures 15-2 and 15-3, the sensitivity performance profile always falls above
the horizontal dashed line, while the specificity performance profile always falls below the
horizontal dashed line. This is because when a carpet dust-lead loading standard is added to
existing standards, it cannot decrease the total number of housing units being triggered by the
entire set of standards. Thus, the added standard will not decrease sensitivity, but it will not
increase specificity.  Equivalently, the added standard will not increase the false negative rate,
but it will not decrease the false positive rate. Therefore, in evaluating the benefit of adding a
carpet dust-lead loading standard, one must consider whether the improvements in some
performance characteristics, such as sensitivity and the false negative rate, outweigh the losses in
others, such as specificity and the false positive rate. As a result, the other two performance
characteristics, positive predictive value (PPV) and negative predictive value (NPV), play more
important roles hi the evaluation.

       In cases where only carpeted floors exist hi a housing unit for dust sampling within a risk
assessment, a carpet dust-lead loading standard would be needed, but not an uncarpeted floor
standard. To investigate the need for such a standard in this type of scenario, the sensitivity/
specificity analysis was repeated by ignoring the uncarpeted floor dust-lead loading standard.
That is, the analysis considered the added benefit associated with adding a carpet dust-lead
loading standard to the set of standards given by window sill dust-lead loading (250 ug/ft2) and
soil-lead concentration (2000 ppm). Figures 15-4 and 15-5 contain plots of the performance
characteristic curves in the situation where the uncarpeted floor dust-lead loading standard is not
used.

       Some of the performance characteristics values plotted in Figures 15-2 through 15-5 are
detailed within Tables 15-8 (for the Rochester study data analysis) and 15-9 (for the HUD
Grantees data analysis). These tables contain calculated values of the four performance
characteristics, their sum, and the percentage of housing units triggered for intervention, for the
following sets of standards:

       •      The standards specified in the §403 proposed rule, without regard to carpet

       •      The standards specified in the §403 proposed rule, plus a carpet dust-lead loading
              standard of 50 ug/ft2 (i.e., the same as the uncarpeted floor dust-lead loading
              standard)

       •      The standards specified in the §403 proposed rule, plus a carpet dust-lead loading
             standard of either 17 fig/ft2 (for the Rochester study data), 5 ug/ft2 (for the HUD
             Grantees data), or 13 jig/ft2 (for the HUD Grantees data) (i.e., "optimal" values of
             the standard)

       •      The standards specified in the §403 proposed rule, without regard to carpeted or
             uncarpeted floors
                                           1-39
U.S. EPA Headquarters Library
       Mail code 3201
1200 Pennsylvania Avenue NW
   Washington DC 20460
-------
                                             i »
Figure 15-4.  Values of the Performance Characteristics As a Function of Candidate
             Carpeted Floor Dust-Lead Loading Standards, Based on Analysis of the
             Rochester Study Data, Where the Set of Standards Also Includes the
             Window Sill and Soil Standards Proposed in the S403 Proposed Rule

(See text for the connotations of the horizontal and vertical dashed lines in these plots. The §403 proposed
standards considered here are 250 j/g/ft2 for window sills and 2000 ppm for soil.)
                                          1-40
-------
    —1
                    i   >o  . •   10
                    IOM-UM Uattg andM
   70
   CO
 i  »
Figure 15-5.   Values of the Performance Characteristics As a Function of Candidate
              Carpeted Floor Dust-Lead Loading Standards, Based on Analysis of the HUD
              Grantees Evaluation Data, Where the Set of Standards Also Includes the
              Window Sill and Soil Standards Proposed in the §403 Proposed Rule

(See text for the connotations of the horizontal and vertical dashed lines in these plots. The §403 proposed
standards considered here are 250 j/g/ft2 for window sills and 2000 ppm for soil.)
                                           1-41
-------
Table 15-8.   Values of the Performance Characteristics for Specified Sets of Standards,
             Based on Analysis of Rochester Study Data
•'_ • ; ":
Set of Standards
Soil standard = 2000 ppm
Sill standard = 250 //g/ft2
Uncarpeted floor standard = 50 jug/ft2
NO CARPET STANDARD
Soil standard = 2000 ppm
Sill standard = 250 //g/ft2
Uncarpeted floor standard = 50 //g/ft2
Carpeted floor standard = 50 //g/ft2
Soil standard = 2OOO ppm
Sill standard = 250 //g/ft2
Uncarpeted floor standard = 50 //g/ft2
Carpeted floor standard = 17 //g/ft2
NO UNCARPETED FLOOR STANDARD
Soil standard = 2000 ppm
Sill standard = 250 //g/ft2
NO CARPET STANDARD
NO UNCARPETED FLOOR STANDARD
Soil standard = 2000 ppm
Sill standard = 250 //g/ft2
Carpeted floor standard = 50 //g/ft2
NO UNCARPETED FLOOR STANDARD
Soil standard = 2000 ppm
Sill standard = 250 //g/ft2
Carpeted floor standard = 17 //g/ft2
;_! _ ' '
Sensitivity
64.6%
66.7%
85.4%

60.4%

62.5%

81.3%

Specificity
60.3%
59.6%
52.6%

62.2%

61.5%

54.5%
.
PPV
33.3%
33.7%
35.7%

33.0%

33.3%

35.5%
"j i. ;
NPV
84.7%
85.3%
92.1%

83.6%

84.2%

90.4%
Sum of.
the 4
Values .
242.9
245.3
265.8

239.2

241.5

261.7
%.of ,
- Homes
Triggered
45.6%
46.6%
56.4%

43.1%

44.1%

53.9%
                                         I-42
-------
Table 15-9.   Values of the Performance Characteristics for Specified Sets of Standards,
             Based on Analysis of HUD Grantees Evaluation Data
•
- Set of Standards
Soil standard = 2000 ppm
Sill standard = 250/ig/ft2
Uncarpeted floor standard = 50 //g/ft2
NO CARPET STANDARD
Soil standard = 2000 ppm
Sill standard = 250/yg/ft2
Uncarpeted floor standard = 50 /ig/ft3
Carpeted floor standard = 50//g/ft2
Soil standard = 2000 ppm
Sill standard = 250^/g/ft2
Uncarpeted floor standard = 50 /yg/ft2
Carpeted floor standard = 13//g/ft2
Soil standard = 2000 ppm
Sill standard = 250/vg/ft2
Uncarpeted floor standard = 50 //g/ft2
Carpeted floor standard = 5 /ig/ft2
NO UNCARPETED FLOOR STANDARD
Soil standard = 2000 ppm
Sill standard = 250 //g/ft2
NO CARPET STANDARD
NO UNCARPETED FLOOR STANDARD
Soil standard = 2000 ppm
Sill standard = 250 fig/ft*
Carpeted floor standard = 50^/g/ft2
NO UNCARPETED FLOOR STANDARD
Soil standard = 2000 ppm
Sill standard = 250#g/ft2
Carpeted floor standard = 13//g/ft2
NO UNCARPETED FLOOR STANDARD
Soil standard = 2000 ppm
Sill standard = 250*/g/ft2
Carpeted floor standard = 5 0g/ft2
jtT »*• »*
.Sensitivity,
78.7%
79.3%
90.2%
94.8%

70.1%

71.3%

85.1%

89.7%
•*ji"« a-" < >. i 't
jt''*^- ^i"
'Specificity
43.4%
42.1%
30.3%
25.8%

52.0%

50.7%

37.1%

31.7%

PPV
52.3%
51.9%
50.5%
50.2%

53.5%

53.2%

51.6%

50.8%
_ *''
NPV -
72.2%
72.1%
79.8%
86.4%

68.9%

69.1%

75.9%

79.5%
Sum of
the 4
Values
246.6
245.4
250.8
257.2

244.5

244.3

249.7

251.7
Jsrdf '
Homes
Triggered
66.3%
67.3%
78.7%
83.3%

57.7%

59.0%

72.7%

77.7%
                                        I-43
-------
       *      The standards specified in the §403 proposed rule, without regard to uncarpeted
              floors, plus a carpet dust-lead loading standard of 50 ug/ft2

       *      The standards specified in the §403 proposed rule, without regard to uncarpeted
              floors, plus a carpet dust-lead loading standard of either 17 ug/ft2 (for the
              Rochester study data), 5 ug/ft2 (for the HUD Grantees data), or 13 ug/ft2 (for the
              HUD Grantees data).

Tables 15-10 and 15-11 provide the 2x2 performance characteristic tables corresponding to each
set of standards specified in Tables 15-8 and 15-9, respectively.  In these tables, numbers in italics
indicate an incorrect risk assessment (either a false positive or a false negative), while those
underlined indicate a correct assessment.

       The analyses presented in this subsection (for both studies) indicate that adding a carpeted
floor dust-lead loading standard of 50 ug/ft2 to the standards in the §403 proposed rule for soil,
window sills and uncarpeted floors did little, if anything, to change the values of the four
performance characteristics. (This can be seen, for example, in the plots within Figures 15-2 and
15-3 by noting that at a carpet dust-lead loading standard of 50 ng/ft2, the performance curves are
approximately at the horizontal dashed line.) This supports the hypothesis that the performance
of the standards would not be affected by adding a carpet dust-lead loading standard equal to the
proposed uncarpeted floor dust-lead loading standard (50 ug/ft2) when both surfaces are available
to sample within a housing unit.  When the uncarpeted floor dust-lead loading standard is not
considered (e.g., in housing units where all floor surfaces are carpeted), the same conclusion is
made (see Figures 15-4 and 15-5). These findings support the hypothesis that adding a carpeted
floor dust-lead standard of 50 ug/ft2 to the currently-proposed §403 standards may not provide a
sufficient level of improved performance to warrant its addition.

       Other candidate carpet dust-lead loading standards that  are lower than 50 ug/ft2 appear to
improve performance of the §403 proposed standards for dust and soil if they are added. These
other candidate standards ranged from 5 ug/ft2 to 17 ug/ft2, depending on the dataset being
analyzed.  For analyses involving the Rochester study data (Figures 15-2 and 15-4; Tables 15-8
and 15-10), the results indicated the following:

       •      The candidate carpet standard resulting in the most unproved performance of the
              proposed §403 standards (for dust and soil) was 17 fig/ft2. Adding this standard to
              the proposed §403 standards increased sensitivity by 20.8 percentage points, PPV
              by 2.4 percentage points, and NPV by 7.4 percentage points, while it decreased
              specificity by 7.7  percentage points (see first and third rows of Table 15-8).
              Adding this standard triggered 22 additional housing units in the Rochester study,
              10 of which contained children with elevated blood-lead concentrations (Table 15-
              10).
                                            1-44
-------
Table 15-10.  Results of Performance Characteristics Analyses for the Sets of Standards
             Included in Table I5-8, Based on Analysis of Rochester Study Data

(PbB = Blood-Lead Concentration)
Soil standard - 2000 ppm
Sill standard = 250 //g/ft2
Uncarpeted floor standard = 50 //g/ft2
NO CARPET STANDARD
Soil standard = 2000 ppm
Sill standard = 250 //g/ft2
Uncarpeted floor standard = 50 //g/ft2
Carpeted floor standard = 50 //g/ft2
Soil standard = 2000 ppm
Sill standard = 250 //g/ft2
Uncarpeted floor standard = 50 /ig/ft2
Carpeted floor standard » 1 7 //g/ft2
NO UNCARPETED FLOOR STANDARD
Soil standard = 2000 ppm
Sill standard = 250 //g/ft2
NO CARPET STANDARD
NO UNCARPETED FLOOR STANDARD
Soil standard = 2000 ppm
Sill standard - 250 //g/ft2
Carpeted floor standard = 50 //g/ft2
NO UNCARPETED FLOOR STANDARD
Soil standard = 2000 ppm
Sill standard = 250 //g/ft2
Carpeted floor standard = 17 //g/ft2

PbB 2 10
«j/dL?


Yes
No
Total

PbB 2 10
//g/dL?


Yes
No
Total

PbB 210
//g/dL?
Yes
No
Total

PbB i 10
//g/dL?


Yes
No
Total .' :. ' •

PbB*10
//g/dL?
Yes
No
Total

PbB i 10
//g/dL?
Yes
No
Total
At least one standard exceeded?
No
17
94
111
Yes
31
62
93
At least one standard exceeded?
No
16
93
•;.;':, 109? - '
Yes
32
63
'•? • 95 - I
At least one standard exceeded?
No
7
82
89
Yes
11
74
115
At least one standard exceeded?
No
19
97
116
Yes
29
59
*;• 88 ••
At least one standard exceeded?
No
18
96
114
Yes
m
6O
90
At least one standard exceeded?
No
9
85
94
Yes
39
71
1 10
Total
48
156
2O4
Total

48"
156
204
Total
''
48
156
204
Total

48V. •
156
':••• 204
Total

48
156
204
Total

48
156
204
                                         I-45
-------
Table 15-11.  Results of Performance Characteristics Analyses for the Sets of Standards
             Included in Table 15-9, Based on Analysis of HUD Grantees Evaluation Data
(PbB = Blood-Lead Concentration)
\
Soil standard = 2000 ppm
Sill standard = 250 //g/ft2
Uncarpeted floor standard = 50 //g/ft2
NO CARPET STANDARD
Soil standard - 2000 ppm
Sill standard = 250 //g/ft2
Uncarpeted floor standard = 50 //g/ft2
Carpeted floor standard = 50 //g/ft2
Soil standard = 2000 ppm
Sill standard = 250 //g/ft2
Uncarpeted floor standard = 50 //g/ft2
Carpeted floor standard = 13 //g/ft2
Soil standard = 2000 ppm
Sill standard = 250 //g/ft2
Uncarpeted floor standard - 50 //g/ft2
Carpeted floor standard = 5 //g/ft2
NO UNCARPETED FLOOR STANDARD
Soil standard = 2000 ppm
Sill standard = 250 //g/ft2
NO CARPET STANDARD
NO UNCARPETED FLOOR STANDARD
Soil standard = 2000 ppm
Sill standard — 250 //g/ft2
Carpeted floor standard = 50 //g/ft2
':?•••• -.. •
PbBilO
//g/dL?

..?
Yes
No
Total
•:"^- ' '
PbB i10
//g/dL?
L '•':•

Yes
No
••'••"•' Total" ".'
• i
PbBilO
//g/dL?

"•'I
Yes
No
Total
•. : "•« . < , •• ', ' - . *••
PbBilO
//g/dL?
Yes
No
.•:.•• Total; - ;

PbB ^10
//g/dL?
Yes
No
Total

PbB i10
//g/dL?
Yes
No
Total
At least one standard exceeded?
No
37
96
133
Yes
132
725
262
At least one standard exceeded?
No
35
93
129
Yes
138
128
266 -,:::: -
At least one standard exceeded?
No
17
67
84
Yes
152
154
311
At least one standard exceeded?
No
9
57
66
Yes
165
164
329
At least one standard exceeded?
No
52
115
167
Yes
122
106
228
At least one standard exceeded?
No
50
112
162
Yes
124
1O9
233
Total
174
221
'395
Total

.;.ijj74/- :
221
' ^3?5 -1
Total
'. ••':-'• - -----
174
- ;.;221 ,.
395
Total
. r .
. ;:174'- .:•:
-;221 ;
395
Total

174
221
395
Total

174
221
395
                                         1-46'
-------
                                  Table 15-11.  (cont.)
NO UNCARPETED FLOOR STANDARD
Soil standard = 2000 ppm
Sill standard = 250 //g/ft2
Carpeted floor standard = 13 //g/ft2
NO UNCARPETED FLOOR STANDARD
Soil standard = 2000 ppm
Sill standard = 250 jug/ft2
Carpeted floor standard = 5 //g/ft2
• ;•'• ' ',.
PbBilO
//g/dL?


Yes
No
• • Total- - ••' ' -

PbB^IO
A/g/dL?
Yes
No
Total
At least one standard exceeded?
No
26
82
108
Yes
H8
139
287
At least one standard exceeded?
No
18
70
88
Yes
156
1S1
307
Total

174
221
395
Total

174
221
395
       •      When not considering the proposed uncarpeted floor standard of 50 Mg^ft2. adding
              a carpeted floor standard of 17 pg/ft2 to the §403 proposed standards for soil and
              window sills increased sensitivity by 20.9 percentage points, PPV by 2.5
              percentage points, and NPV by 6.8 percentage points, while it decreased
              specificity by 7.7 percentage points (see fourth and sixth rows of Table K-8). As
              in the previous bullet, adding this standard triggered 22 additional housing units in
              the Rochester study, 10 of which contained children with elevated blood-lead
              concentrations (Table 15-10).

Thus, results of the analyses on Rochester study data suggest that improved performance
characteristics, particularly sensitivity, are achieved with a carpeted floor standard of 17 ug/ft2
without a large decrease in specificity. If this increased performance is considered important
enough, then a carpeted floor standard (set sufficiently low enough) would be warranted for all
homes.

       The above results based on analysis of the HUD Grantees evaluation data (Figures 15-3
and 15-5; Tables 15-9 and 15-11) include the following:

       •      The candidate carpet standard resulting in the most improved performance of the
              proposed §403 standards (for dust and soil) was 5 ug/ft2. Adding this standard to
              the proposed §403 standards increased sensitivity by 16.1 percentage points and,
              NPV by 14.2 percentage points, while it decreased specificity by 17.6 percentage
              points and PPV by 2.1 percentage points (see first and fourth rows of Table 15-9).
              Adding this standard triggered 67 additional housing units in the HUD Grantees
              evaluation,  28 of which contained children with elevated blood-lead
              concentrations (Table 15-11).
                                           1-47
-------
       *     The lower-right plot within Figure 15-3 indicates that a carpeted floor standard of
              13 ug/ft2 achieves some gain in overall performance without observing as large of
             a decrease in specificity as occurs with the candidate standard of 5 ug/ft2. Adding
             this standard triggered 49 additional housing units in the HUD Grantees
             evaluation, 20 of which contained children with elevated blood-lead
             concentrations (Table 15-11). Therefore, if a large loss hi specificity outweighs the
             gain in sensitivity and NPV that is observed with the candidate standard of 5
             ug/ft2, then the alternative standard of 13 ug/ft2 may be of more interest.

       •     When not considering the proposed uncarpeted floor standard of 50 ug/ft2, adding
             a carpeted floor standard of 5 ug/ft2 to the §403 proposed standards for soil and
             window sills increased sensitivity by 19.6 percentage points and NPV by 10.6
             percentage points, while it decreased specificity by 20.3 percentage points and
             PPV by 2.7 percentage points (see fifth and eighth rows of Table 15-9). Adding
             this standard triggered 79 additional housing units hi the HUD Grantees
             evaluation, 34 of which contained children with elevated blood-lead
             concentrations (Table 15-11).

       •     When not considering the proposed uncarpeted floor standard of 50 ug/ft2, adding
             a carpeted floor standard of 13 ug/ft2 to the §403 proposed standards for soil and
             window sills had a slightly lower increase in sensitivity and NPV than adding a
             standard of 5 ug/ft2, but the decrease in specificity was only 14.9 percentage
             points (Table 15-9).

These results indicate that improved sensitivity and NPV were achieved by adding a carpeted
floor standard of 5  ug/ft2, but a considerable decrease in specificity was also observed. Less of a
loss in specificity, with only a minor loss of improvement in the other performance
characteristics, was achieved when the candidate carpet standard was increased to 13 ug/ft2. If
this increased performance is considered important enough, then a carpeted floor standard (set
sufficiently low enough) would be warranted for all homes.

15.2   DETERMINING A CARPETED FLOOR DUST-LEAD
       LOADING STANDARD

See Section 14.2 and its subsections for details on the statistical methods associated with the
results presented in this section.

15.2.1 Comparing Average Dust-Lead Loadings Between
       Carpeted and Uncarpeted Floors in a Housing Unit

       A total of 168 housing units in the Rochester study had wipe dust-lead loading data for
both carpeted and uncarpeted floors. When considering the ratio of a housing unit's average
dust-lead loading for carpeted floors versus uncarpeted floors, the geometric mean of these ratios
across the 168 housing units was 0.745, indicating that the average dust-lead loading for carpeted

                                          1-48
-------
floors was roughly 75% of the unit's average for uncarpeted floors. This geometric mean had a
95% confidence interval of (0.62,0.90), implying that the geometric mean was significantly
different from one (i.e., equal averages between carpeted and uncarpeted floors within a unit) at
the 0.05 level based on a paired t-test on the log-transformed averages. Only 36% of the 168
housing units had ratios which exceeded one (i.e., had average carpeted floor dust-lead loadings
that exceeded the average for uncarpeted floors).

       For the Rochester study, Figure 15-6 portrays a housing unit's area-weighted average
dust-lead loadings for carpeted floors versus its average for uncarpeted floors. The solid line in
Figure K-6 represents equality in the averages between the two surface types. This plot indicates
that the average loadings from uncarpeted floors are generally higher than for carpeted floors.

15.2.2 Regression Modeling Approach

       Figure 15-7 presents the upper 95% prediction bounds on the curve that results from
fitting model (1) of Section 14.1.1.2 to the Rochester study data to predict blood-lead
concentration as a function of average floor wipe dust-lead loading.  As the model was fitted
separately for carpeted floor dust-lead loading data and uncarpeted floor data (with equal weight
given to each housing unit), one set of prediction bounds exist for each surface type. Vertical
dashed lines are included at dust-lead loadings of 17 ug/ft2 and 50 ug/ft2, corresponding
respectively, to the "optimal" carpet dust-lead loading standard identified in the performance
characteristics analysis of Section 15.1.3 on the Rochester study data and to the §403 proposed
standard for uncarpeted floors.

       The confidence bounds in Figure 15-7 represent predicted blood-lead concentrations for
which approximately 95% of children would fall below.  For example, Figure 15-7 indicates that
approximately 95% of children exposed to an average carpeted dust-lead loading of 50 ug/ft2 (the
proposed uncarpeted floor standard) are expected to have blood-lead concentrations below 22.4
ug/dL. In contrast, approximately 95% of children exposed to an average uncarpeted floor dust-
lead loading of 50 ug/ft2 are expected to have blood-lead concentrations below 24.1 ug/dL. As
22.4 ug/dL is slightly below 24.1 ug/dL, this implies that a carpet dust-lead loading standard of
50 ug/ft2 would be at least as  protective of children's blood-lead concentrations as the same
standard for uncarpeted floors.

       Figure 15-7 shows that the upper 95% prediction bounds for the two surfaces are very
similar, generally within 2 ug/dL, with the bound for uncarpeted floors exceeding that for
carpeted floors above approximately the "optimal" carpet dust-lead loading  standard of 17 ug/ft2.
Approximately 95% of children exposed to either carpeted or uncarpeted floor dust-lead loadings
of 17 ug/ft2 would have blood-lead concentrations below approximately 20 ng/dL. Note that no
candidate dust-lead loading standards in the ranges considered in Figure 15-7 result in 95% of
children having blood-lead concentrations below 10 fig/dL.
                                           1-49
-------
 0
 o
 c
 O»
 T3
 O
 O
    100000
      10000
       1000
        100
         10
        0.1
       0.01
          0.01
0.1
1        10       100      1000


 Dust—Lead Loading (Carpeted)
10000   100000
Figure 15-6.   Plot of Area-Weighted Average Wipe Dust-Lead Loadings (pg/ft2) for

             Uncarpeted Floors Versus Carpeted Floors Within a Housing Unit in the

             Rochester Study
                                       1-50
-------
 e
 o
 I
 "c
 <0
 o
 c
 o
 o
 •o
 o
 
-------
       Figure 15-8 presents the results of this performance characteristics analysis performed on
the Rochester study data. One plot exists in Figure 15-8 for each of the four performance
characteristics and for the sum of these four characteristics. The vertical axes of these plots
identify the performance characteristic being plotted. Solid-line performance curves correspond
to carpeted floors, and dashed-line performance curves correspond to uncarpeted floors. Like in
Figure 15-7, vertical dashed lines exist in each plot at 50 and 17 ug/ft2.

       The plots within Figure 15-8 indicate the following:

       •     The proposed uncarpeted floor standard of 50 ug/ft2 results in a considerably
             lower value for the sum of the four performance characteristics when the standard
             is assumed to be for carpeted floors rather than for uncarpeted floors. In contrast,
             candidate standards from 15 to 20 ug/ft2 result in considerably higher values for
             this sum when the standard  is assumed to be for carpeted floors.  (Note that this
             result tends to agree with the results hi Section 15.1.3.)

       •     To achieve sensitivity at the level observed for the §403 proposed standard for
             uncarpeted floors  (50 ug/ft2), the carpeted floor dust-lead loading standard must
             be below approximately 33  ug/ft2.

       •     At a standard of 50 ug/ft2, PPV is lower if the standard is for carpeted floors than
             if it is for uncarpeted floors. Among the candidate carpeted floor dust-lead
             loading standards, PPV is maximized at 30 ug/ft2; this maximum is approximately
             equal to the PPV for the §403 proposed standard for uncarpeted floors of 50
             fig/ft2.

       •     The performance curves for NPV differ little, if any, between carpeted and
             uncarpeted surfaces across the range of candidate standards.

The conclusion of this performance characteristics analysis is that, for carpeted floors, a standard
of 30 ug/ft2 may be needed to achieve a level of protection equal to that of the §403 proposed
standard of 50 ug/ft2 for uncarpeted floors. Furthermore, a standard of 17 ug/ft2 continues to be
among the better performers when the total of the four performance characteristics is considered
as a criterion.
                                           1-52
-------
    0    10   20   3>   40   00
                 FtoOUK-UUlu
                               TB   m   go
                       i
                       +
                       I
                           o    n   a   ID   «
                                                  «   m   to   •>
Figure 15-8.   Values of the Four Performance Characteristics Versus Floor Dust-Lead
              Loading Standard By Surface Type, Where No Other Standards Were
              Considered, Based on Analyses Performed on Rochester Study Data

(Note: Vertical dashed lines correspond to the §403 proposed uncarpeted floor dust-lead loading standard of
50 //g/ft* and the "optimal" carpeted floor dust-lead loading standard of 17 ^g/ft2 from Section 15.1.3.)
                                            I-53
-------
15.3   DETERMINING AN APPROPRIATE METHOD FOR
       SAMPLING CARPET DUST

See Section 143 and its subsections for details on the statistical methods associated with the
results presented in this section.

       Besides wipe sampling, the Rochester study employed BRM and DVM vacuum sampling
on carpeted and uncarpeted floors, while the HUD Grantees evaluation included a few
measurements on carpeted floor dust samples collected using the DVM. These vacuum sampling
methods, however, require specialized equipment and more training to use effectively. In
addition, vacuum sampling is more complex and costly relative to any added benefit it may
provide (Section 403 Dialogue Process minutes, December 14,1995). Therefore, in discussions
regarding the §403 risk analysis, wipe sampling was supported as dust collection method in
which the dust-lead standards would be expressed.

       Sections 15.3.1 through 15.3.3 contain the results of analyses to compare dust-lead
loadings between the different dust sampling methods employed in the Rochester study and HUD
Grantees evaluation for carpeted and uncarpeted floors. Also compared in these analyses were
dust-lead concentrations measured within dust samples obtained using vacuum techniques
(BRM, DVM). The results in this section are supported by the additional data summaries found
hi Appendix £2. The main findings of these results were as follows:

       •      Blood-lead concentration correlated more highly with dust-lead loading than with
             dust-lead concentration on both carpeted and uncarpeted surfaces.

       *      Each dust collection method resulted in measured dust-lead loadings that were
             statistically significant predictors of blood-lead concentration.  There was not
             strong evidence to favor any particular method based on predictive ability.

       •      Dust-lead loadings on either surface were significantly positively correlated
             between dust collection methods. Additionally, one may predict wipe loadings
             based on BRM- and DVM-measured loadings using the regression results in
             Section 15.3.3.  Thus, exclusive use of wipe sampling for floor-dust captured some
             of the information that would be available from use of vacuum sampling.

       •      On carpeted floors in these two studies, vacuum sampling methods collected
             samples having significantly different loading measurements compared to wipe
              sampling (see Tables 12-1 and I2-6a of Appendix 12, and Section 15.3.3). As a
             consequence, a standard designed for wipe sampling would not apply to vacuum-
              sampled floor dust-lead, and vice versa.

       •       As the uncarpeted  floor dust-lead loading standard assumes wipe sampling, and
              dust-lead loadings under each of the three dust collection methods have
              significant correlations with blood-lead concentration for both carpeted and

                                          1-54
-------
             uncarpeted floors (as seen in Section 15.3.1), these results imply that it is
             reasonable to develop a carpeted floor dust-lead standard for the wipe sampling
             method. As this standard would not apply to vacuum sampled dust-lead loadings,
             measurements taken with vacuum sampling could not be used in risk assessment
             via the §403 rule.

15.3.1 Investigating the Association Between Floor Dust-Lead
       Levels and Blood-Lead Concentration for Different
       Sampling Methods

       This subsection presents, for both carpeted and uncarpeted floors and for each of the three
dust collection methods, analyses of the Rochester study and HUD Grantees evaluation data to
investigate the bivariate relationships between children's blood-lead concentration and area-
weighted household average floor dust-lead loading.  Furthermore, using the Rochester study
data, this subsection also investigates the relationships between children's blood-lead
concentration and mass-weighted average floor dust-lead concentration, for each of the two
vacuum dust collection methods and for carpeted and uncarpeted floors separately.

Rochester Study

       Figure 15-9 contains six plots, each depicting blood-lead concentration versus household
average floor dust-lead loading for a given combination of dust collection method and floor
surface type (carpeted or uncarpeted), as measured in the Rochester study.  Figure 15-10 contains
four plots, each presenting blood-lead concentration versus household average floor dust-lead
concentration for each combination of the two vacuum collection methods and the two floor
surface types. Each point within the plots in Figures  15-9 and 15-10 represents a single housing
unit surveyed in the Rochester study.

       As all plots in Figure 15-9 cover the same ranges along their vertical and horizontal axes,
it is possible to see, for example, how average dust-lead loadings are generally higher when
samples are collected by the BRM than by the DVM, especially for carpeted surfaces.  The plots
in Figure 15-9 show some positive correlation between dust-lead loadings and blood-lead
concentration, but the level of variability in these relationships is high under all dust collection
methods. Little, if any, correlation is observed between dust-lead concentration and blood-lead
concentration (Figure 15-10) for either vacuum method or floor surface type.

       For each plot in Figures 15-9 and 15-10, a Pearson correlation coefficient was calculated
on the data in the plot to quantify the extent of linear relationship between log-transformed
blood-lead concentration and log-transformed average floor dust-lead level, with each average
weighted by the proportion of total floor sample area in the unit represented by the given surface
type.  The correlation coefficients for a particular surface type (carpet, non-carpet) are presented
in Table 15-12.
                                           1-55
-------
                  BRM and Carp*t«f
BRUondUneaipctad
                                                                                              B
100
i
|
8 10
s
Y
I
£




i


0
a
O 9** (
0° 0
K**<£l

Oo»*|^
« °VV

8o a
%


0.01 0.1 1 10 100


e o «
» 0 °0 0

>tK<£fp «°° *
oSp o
LI o .
0 0
o/o* »

0 °


100'
c
1
o 10
i
"l
I
a




1

« 0
e« °0. ; • *•
^<> *%«
% •> ssfi^^ *
o °{»^a r C °
SI'wye *, o
° O° OnS^ °
0 5* » ,° o
0 ° 0 °


0.01 0.1 1 10 100
Q mumt Unearp«M |




^ o o
o
o





too
i
1
f
S 10
•^
1
!
s



1
> a
0 "ee o *
o « ^ a
8 ** o \,^» *o o °
* * *0 ^^^ J^« ^p A '
,3««f 4***5V *
8 *»o 4^8 * o<"fo „*
Off ° * Qj.O^O° 0
0 • a ,*« o e
* «" °« t
a

1000 10000 looooo 0.01 0.1 i 10 100 1000 loooo looaa
Dnt-Uad Loading Ouj»-Uod Uadlng
WIPE and Canxtte
100
|

^
"c
0 10

s
J
Tl
D




1
0
» ..
°S o
-nSf.'
*jgS?
Q t^nSt tfiO
°o°ooSi0<>*o^sV
qOo 0?S^
". «*«lc

% "e
" 
-------
                   8RU oivd Corprtri
                                                                       BRM ond Uncsipcttd
                                                                                                   B
  100
   10
    0.0)    O.t
                .   .     .

    0.01    0.1     t     10    too   1000   loooo  looooo
                   Dwt-lMd CancMlrrilan

                   ami and CariHttd                   r
1      10    100    1000   10000 100000
   Dint-LMd Concmtrothm
                                                      100
                                    g
                                    &  10
                                    •o
                                    o
                                    Y
                                                       100
                                                        0.01    0.1
                                                                                          -"
                                       0.01     0.1     i      10    100    looo   loooo 100000
                                                       Dutf-Lnd Cwc.ntrotk.ri

                                                      OVM and Uncap«t*d                 n
                                                                                r  ft:"-    •
                                                                          «   ;.    »-% s
                                                                               ° -O „  «
                                                                           10    100    1000   10000  1
                                                                             Cwiowitralion
Figure 15-10. Plots of Blood-Lead Concentration U/g/dL) Versus Weighted Average Floor
               Dust-Lead Concentration (pg/g) in the Rochester Study, by Dust Collection
               Method and Floor Surface
                                                I-57
-------
Table 15-12.  Pearson Correlation Coefficients of Log-Transformed Average Dust-Lead
              Levels with Log-Transformed Blood-Lead Concentration, as Measured in the
              Rochester Study, for Differing Dust Collection Methods and Measurement
              Types
Floor Dust-Lead Variable1
? • '
Area-weighted average dust-
lead loading
Mass-weighted average dust-
lead concentration
1: . Correlation wrilh Blood-Lead Concentration -f
JBRM /''- ?-
Carpeted
Floors :*•
0.339»*
(179)
0.100
<178)
Uncarpeted;
Floors -„-
0.364**
{191)
0.086
(189)
« - DVM
Carpeted
, ROOTS
0.239**
(181)
0.046
(177)
Uncarpeted
f. Floors
0.152*
(194)
-0.037
(177)
i -•>_ , t *•"* -
Wipe -7 ,
Carpeted
Floors «*
0.190*
(179)
Uncarpeted
Floors
0.313**
(193)
'-:-'•,,
1 Correlation coefficients are calculated on unit-wide area-weighted average dust-lead loadings or mass-weighted dust-lead
concentrations, where averages are taken across all samples in a housing unit of the given surface type (carpet or non-
carpet). The average for a given housing unit is weighted by the proportion of total floor sample area in the unit represented
by carpeted (uncarpeted) surfaces in calculating the correlation coefficient for carpeted (Uncarpeted) floors.

* * Significant at the 0.01 level.
* Significant at the 0.05 level.
       The results in Table 15-12 indicate the following:

       •      None of the correlations between blood-lead concentration and average dust-lead
              concentration were significant at the 0.05 level for either the BRM or DVM or for
              either carpeted or uncarpeted surfaces (see die last row of the table).

       •      Significant correlation was observed at the 0.05 level between blood-lead
              concentration and average dust-lead loading for each dust collection method when
              sampling from either carpeted or uncarpeted floors.  Among carpeted floor data,
              the correlation coefficients between dust-lead loading and blood-lead
              concentration ranged from 0.190 under wipe methods to 0.339 under the BRM,
              while for uncarpeted floor data, these correlation coefficients ranged from 0.152
              for the DVM to 0.364 for the BRM. Only for the DVM was the correlation
              coefficient larger for carpeted surfaces than for uncarpeted surfaces.

These results differ slightly from correlation coefficients reported in the Rochester study report
(the Rochester School of Medicine and NCLSH, 1995), primarily due to the form of the dust-lead
parameter (this analysis used a log-transformed weighted arithmetic average of untransformed
data, while the Rochester study report used an untransformed, unweighted average of log-
transformed data). However, the results in Table 15-12 agree with the findings of other studies
(see Section 15.1.2 of USEPA, 1997a) that blood-lead concentration correlates more highly with
                                             1-58
-------
dust-lead loading than dust-lead concentration; this result was observed for both carpeted and
uncarpeted surfaces.

       To further investigate the statistical nature of the bivariate relationships represented in
Table 15-12, the regression model (1) of Section 14.1.1.1 was fitted to Rochester study data for
each of these ten pairs of parameters. Table 15-13 presents the estimated slope and intercept
terms for each model fit, along with the standard errors of each estimate. Significant slope
estimates imply that the predictor variable is significantly associated with blood-lead
concentration.

       Results from Table 15-13 are as follows:

       •      For all but one of the model fits, the slope estimate was positive, implying
              increased blood-lead concentrations associated with increased values of the dust-
              lead predictor variable.  (The negative estimate  associated with the remaining
              model fit was not significantly different from zero.)

       •      At the 0.05 level, dust-lead loadings were statistically significant predictors of
              blood-lead concentration under each dust collection method and for both carpeted
              and uncarpeted floors, while dust-lead concentrations were not significant
              predictors.

       •      All three dust collection methods, when used to measure dust-lead loading, were
              significant predictors of blood-lead concentration. No strong evidence was
              uncovered to favor any one over the others based on predictive ability from this
              analysis.

       •      Dust-lead levels from carpeted floors did not appear to predict blood-lead
              concentration any more or less accurately than did dust-lead levels from
              uncarpeted floors.

HUD Grantees Program Evaluation

       Floor dust samples  were collected by either wipe or DVM vacuum methods in the HUD
Grantees evaluation, with the DVM method used only to collect a few carpet-dust samples.
Figure 15-11 graphically portrays the three sets of relationships between blood-lead concentration
and average floor dust-lead loading (carpet dust-lead loadings under DVM and under wipe, and
uncarpeted floor wipe dust-lead loadings). Each point within the plots represents a single
housing unit. While each plot in Figure 15-11 tends to show a positive relationship between the
two endpoints, considerable variability associated with this relationship is present.

       Pearson correlation coefficients were calculated on the  data within each plot in Figure
15-11 to quantify the extend of linear relationship between the log-transformed blood-lead
                                            1-59
-------
Table 15-13.  Estimates of Intercept and Slope Parameters (and Their Standard Errors)
               Associated With Regression Models Fitted to Rochester Study Data That
               Predict Blood-Lead Concentration Based on  Average Floor Dust-Lead Level,
               for Different Surface Types and Dust Collection Methods
Boor Surface
Type
Carpeted surfaces
Uncarpeted
surfaces
<\.
Dust-Lead Endpoint!
(PbDf
BRM Loading
DVM Loading
Wipe Loading
BRM Concentration
DVM Concentration
BRM Loading
DVM Loading
Wipe Loading
BRM Concentration
DVM Concentration
" , Estimates (Standard Errors) "-_--; "*--•-
"" <*l
intercept \i/i
1.08(0.16)
1 .66 (0.06)
1.53(0.11)
1.59(0.16)
1.71 (0.14)
1.55(0.08)
1.88(0.05)
1.39(0.12)
1.73(0.16)
1.98(0.14)
Baseline ' 3
(tf'JWB/dU T
2.95
5.25
4.61 '
4.92
5.55
4.72
6.56
4.03
5.62
7.24
ia.K$*- :*£i- *
- Slope Igl- _ -' -
0.129" (0.027)
0.094** (0.029)
0.103* (0.040)
0.042(0.031)
0.016 (0.027)
0.1 11** (0.021)
0.054* (0.025)
0.174** (0.038)
0.030 (0.025)
-0.012 (0.024)
The regression model takes the form logtPbB,) = p + adoglPbO,)) + eif or equivalent^ PbB, =exp(//) x(PbDi)°'xexp(ei),
where PbB is the blood-lead concentration for the child in the ith housing unit, e, refers to the random error associated with
the model-based blood-lead concentration for the ith unit, and remaining notation is specified in the column headings. For a
specific surface type, results for the ith unit are weighted by the proportion of total area represented by that surface type. -

* Significantly different from zero at the 0.05 level.
*"  Significantly different from zero at the 0.01 level.
                                                 1-60
-------
              Carp*t*d noor. — «fl|M Method
                                                              Uncarpctol Floor. ~ Wlp, Ifothod
  100.0
          . „   g. »". *   '.
          * «  „?*,«*>  _ ••
  10O.O
                                                                    a  ooo oa
                                                                 « « o OCDOQ e
               10        100        1000
                   Aug. Dint-trod Looalnj
                Carp
-------
Table 15-14.  Pearson Correlation Coefficients of Log-Transformed Blood-Lead
               Concentration and Log-Transformed Average Dust-Lead Loading as
               Measured in the HUD Grantees Program Evaluation, According to Dust
               Collection Method and Floor Surface Type
Dust Collection Method
DVM
Wipe
Pearson Correlation Coefficients1 (Number of Housing Units}
"•.^Carpj^Bd Floors
0.640" (24)
0.308** (226)
:: Uncarpeted Fjoors ,
(Not collected)
0.335* • (390)
' Area-weighted average dust-lead loadings are taken across all samples in a housing unit of the given dust collection
method and surface type (carpeted or uncarpeted). The average for a given housing unit is weighted by the proportion of
total sample area in the unit represented by carpeted (uncarpeted) floors for calculating the correlation coefficient with
carpeted (uncarpeted) floors for each dust collection method.

** Significantly different from zero at the 0.01 level.
Table 15-15.  Estimates of Slope Parameters (and Their Standard Errors) Associated With
               Regression Models Fined to Data from the HUD Grantees Program
               Evaluation That Predict Blood-Lead Concentration Based on Average Floor
               Dust-Lead Loading, For Different Surface Types and Dust Collection
               Methods
Dust Collection Method
Wipe
DVM
: Surface Type
Carpeted Floor
Uncarpeted Floor
Carpeted Floor
# of Units
226
390
24
.;siope ^iSttfc.Errort
0.160** (0.048)
0.117'* (0.030)
0.279** (0.074)
1 The regression model takes the form log(Pb8,l = Yj + a*log(PbDs) + e,, where PbB, represents the blood-lead
concentration for the selected child in the ith housing unit within the jth grantee, PbDs corresponds to the observed average
floor dust-lead loading for the ith housing unit within the jth grantee (for the given dust collection method and surface type),
and a and YJ are parameters representing the slope of the model and the intercept for the jth grantee, respectively. The
residual error left unexplained by the model is denoted by e,. Observations entering into the model are weighted by the
proportion of total sample area in the unit represented by carpeted (or uncarpeted) floors for each dust collection method.

•  Significantly different from zero at the 0.05 level.
• *  Significantly different from zero at the 0.01 level.
               The slope for each model fit was statistically significantly positive (at the 0.01
               level), indicating that average dust-lead loadings were significantly associated
               with blood-lead concentration and that high blood-lead concentrations were
               associated with high dust-lead loadings. (Similar results were observed in Table
               15-13  when the Rochester data were analyzed, but significance was not always at
               the 0.01 level.)
                                                1-62
-------
             As in the Rochester study data analysis, there was not strong evidence to favor
             DVM over wipe sampling based on predictive ability in this analysis. However,
             there were so few DVM measurements taken in the HUD Grantees evaluation that
             it was difficult to make any conclusions from the available DVM measurement
             data.

             As was seen in analysis of the Rochester study data, dust-lead levels from
             carpeted floors were not found to predict blood-lead concentration any more or
             less accurately than do dust-lead levels from uncarpeted floors.
15.3.2 Determining the Relationship of Average Dust-Lead
       Levels Between Sampling Methods

       This analysis of Rochester study data, documented in Section 14.3.2, investigated the
bivariate relationship between the following pairs of dust-lead measurements, with each
comparison done separately for carpeted and uncarpeted floors:
             Average BRM dust-lead loading versus average DVM dust-lead loading
             Average BRM dust-lead concentration versus average DVM dust-lead
             concentration
             Average BRM dust-lead loading versus average wipe dust-lead loading
             Average DVM dust-lead loading versus average wipe dust-lead loading
             Average BRM dust-lead loading versus average BRM dust-lead concentration
             Average DVM dust-lead loading versus average DVM dust-lead concentration
       Data for these six pairs of parameters are plotted within Figures 15-12 through 15-14, with
separate plots generated for data from carpeted floors and from uncarpeted floors. Four plots of
BRM versus DVM dust-lead levels (loadings and concentrations) are found in Figure 15-12, four
plots of wipe versus vacuum dust-lead loadings are found in Figure 15-13, and four plots of dust-
lead concentrations versus loadings for vacuum methods are found in Figure 15-14. Each plotted
point corresponds to average results for a single housing unit in the Rochester study. If dust-lead
levels agreed perfectly among samples of different dust collection methods within a unit, the
plotted points in Figures 15-12 and 15-13 would fall along the solid line representing equality in
these plots.

       The plots in Figures 15-12 through 15-14 indicate the following:

       •      For both uncarpeted and carpeted surfaces in a housing unit, dust-lead loadings
             were generally lower for the DVM than for the BRM (plots A and B of Figure
             15-12) or under the wipe method (plots C and D of Figure 15-13).
                                          1-63
-------
                                                                                                     B
   100000


   10000

    TOOO


     100


     10


      1'


     0,1
    0.01
      0.01    0.1
                   1     10    100    1000  10000 100000
                     DlBt-Uod Loading (DVU)
                    100     1000    10000   100000 1000000
                    Duft-Uad Concentration (BVU)
10000]


 1000


 100


  10

   1


 0.1
                                                           0.01    0.1
               1     10    100    1000   10000  100000
                  Diat-tMO Uoding (DVy)
                                                         10000


                                                          1000


                                                          100
           10     100     1000    1OQQQ   1OOOOO 100000O
                 Duri-LMd CgtiMOtraHon (DVU)
Figure 15-12.  Plots of Weighted Average Dust-Lead Loadings (//g/ft2) and Concentrations
                Ovg/g) for BRM Dust Samples Versus DVWI  Dust Samples in the Rochester
                Study, by Floor Surface Type
                                                 I-64
-------
                      Cerp*M
                                                                                                     B
   10000

    1000

    100

     10

      1

    O.I
    0.01
      0.01    0.1
   10000
§  10OO

f   '°»
j
I    10

I     '
    o.t
                   1     10    100    1000  10000 100000
                     Dwt-Uod Uodng (BUM)
                                                        10000


                                                     ji   1000


                                                     I   100
                                                     a

                                                          10


                                                           I


                                                         0.1
                                                        100000


                                                         10000


                                                         1000


                                                          100


                                                           10
                                                           0.01    0.1    1     10     100    1000  10000 100000
                                                                          Dui»-Uod Uedlng (BSM)
      0.01    O.I     I     to .    100    1000  10000 100000
                                                           0.01   0.1     1      10    100   1000   10000 100000
                                                                          Owl-LMd LMdln« (DVU)
Figure 15-13.  Plots of Weighted Average Dust-Lead Loadings (//g/ft2) for Wipe Dust
                Samples Versus BRM and DVM Dust Samples in the Rochester Study, by
                Floor Surface Type
                                                 I-65
-------
                                                                         BUI mi UncapiM
                                                                                                      B
  1000000
   100000
    10000
    1000
     100
      10
                                       *> "
1000000
100000
10000
• 1000
100
10
1
o
Q
o e
•' «a°
o • 08 a, °>9
Q 9*99 V 0 ° °
.. v* V: • °° '
o°0*o „. V o
• o
0 0 0
      0.01    0.1     1     10    100   1000   10000 100000
                       Dust-Uod Loading

                   OVM and CorptM                   f>
   1 00000
    1000
      10
0.01    0.1    1     10    100    1000   10000  100000
                0«jt-l*Dd Loading

            DVU and Uneafp«*d                   Q
               "V
                                                        1000000
                                                         100000
                                                          10000
                                                          1000
                                                                 8 . o  o
       0.01    0.1
                         10    100   1000   10000 100000
                       D<£rt-Uod loceHns
0.01    0.1     1     10    100    1000   10000 100000
                OuU-Uod locdlng
Figure 15-14.  Plots of Weighted Average Dust-Lead Concentrations (pg/g) Versus Average
                Dust-Lead Loadings (jtig/ft2) for BRM and DVM Dust Samples in the
                Rochester Study, by Floor Surface Type
                                                  I-66
-------
       •      Iii Figure 15-13, larger dust-lead loadings for the BRM were observed relative to
              the wipe method for carpeted surfaces (plot A) but not for uncarpeted surfaces
              (plotB).

       •      In general, wipe results were less variable than were the BRM and DVM results
              for both carpeted and uncarpeted surfaces (Figure 15-13).

The plots in Figure 15-14 show generally positive relationships between dust-lead concentrations
and dust-lead loadings among the (vacuum) dust collection methods and surface types.

       For carpeted and uncarpeted surfaces separately in the Rochester study, Pearson
correlation coefficients were calculated to observe the extent of a linear relationship in the log-
transformed area-weighted average dust-lead loadings (and mass-weighted average dust-lead
concentrations) between different dust collection methods, as well as the extent of a linear
relationship between log-transformed dust-lead loadings and log-transformed dust-lead
concentrations for each dust collection method. These correlation coefficients are presented in
Table 15-16. Note that in calculating a correlation coefficient on data associated with carpeted
floors, each data point was weighted by the proportion of floor sample area in the housing unit
represented by carpeted surfaces for the dust collection method(s) being considered, while data
associated with uncarpeted floors were weighted by the proportion of floor sample area
represented by uncarpeted surfaces.

Table 15-16.   Pearson Correlation  Coefficients of Log-Transformed Dust-Lead Levels
               Measured in the Rochester Study, for Differing Dust Collection Methods or
               Measurement Types
Pair of Parameters
Considered in the Correlation
p(BRM, DVM}
p(BRM, Wipe)
p(DVM, Wipe)
pfdust-lead loading,
dust-lead concentration)
Type of Data Considered
, in the Correlation1 :;e
Dust-Lead Loading
Dust-Lead Concentration
Dust-Lead Loading
Dust-Lead Loading
BRM
DVM
Pearson Correlation Coefficients1
. .,., (Number of Housing Units)
Carpeted Surfaces
0.545" * (179)
0.549" (175)
0.520** (177)
0.456** (179)
0.510** (178)
0.601** (177)
Uncarpeted Surfaces
0.493** (191)
0.389** (173)
0.523** (191)
0.463** (193)
0.551** (189)
0.623** (177)
' Correlation coefficients are calculated on unit-wide area-weighted average dust-lead loadings or mass-weighted dust-lead
concentrations, where averages are taken across all samples in a housing unit of the given surface type (carpet or non-
carpet). In these calculations, the average for a given housing unit is weighted by the proportion of total sample area in the
unit represented by carpeted (uncarpeted) surfaces in calculating the correlation coefficient for carpeted (uncarpeted)
surfaces.

**  Significant at the 0.01 level.
                                             I-67
-------
       All correlation coefficients in Table 15-16 were significant at the 0.01 level, regardless of
whether data for carpeted or uncarpeted floors were being considered. Thus, the extent that
linear relationships are present among the log-transformed dust-lead levels of differing dust
collection methods or between dust-lead loadings and dust-lead concentrations under a specific
vacuum method was consistent for both carpeted surfaces and uncarpeted floors.  In particular,
for carpeted floors, all three methods were significantly positively correlated.

15.3.3 Investigating the Relationship in Lead Loadings of Side-by-Side
       Dust Samples Collected by Different Methods

       To determine how the dust-lead loading measurement at a given sampling area differs
between dust collection methods, regression model (6) of Section 14.3.3 was fitted to the
measured dust-lead loadings for individual samples collected in Rochester study housing units,
with samples taken from the same room assumed to be from adjacent, side-by-side areas. The
regression model predicted the dust-lead loading for a sample taken by a specified dust collection
method (method A) as a function of the dust-lead loading for the adjacent sample taken by
another collection method (method B), with separate model fits for carpeted floor data and
uncarpeted floor data.

       Table 15-17 contains the estimated intercept and slope parameters and their standard
errors associated with predicting dust-lead loadings under method A given the dust-lead loadings
under method B. This table indicates that, for both carpeted and uncarpeted floors and at the
0.05 level, the intercepts were significantly different from zero in all but two instances, and the
slope estimates were always significantly different from one. Thus, based on analysis of data
from the Rochester study, different dust collection methods tended to provide dust samples with
quantitatively different lead loadings, regardless of floor surface type, even when the dust
samples were collected from adjacent locations. The extent of these differences was a function
of the magnitude of the measurements.
                                           1-68
-------
Table 15-17.  Estimates of Intercept and Slope Parameters (and Their Standard Errors)
                When Fitting Regression Models to Rochester Study Data That Predict Roor
                Dust-Lead Loadings Under Dust Collection Method A From Loadings for an
                Adjacent Roor Area Collected Using Method B
Roor Surface Type
*• ,
* i
Carpeted surfaces
Uncarpeted
surfaces
- Dust4Lead Level '
£ **n-&!p&ta**«< -
•^ nu>jOe
DVM Loading
Wipe Loading
Wipe Loading
BRM Loading
BRM Loading
DVM Loading
DVM Loading
Wipe Loading
Wipe Loading
BRM Loading
BRM Loading
DVM Loading
Estimate (Standard Brorj
» • f —^ftr " * .,<..
Intercept (p) , ; '
4.81* (0.10)
4.52* (0.25)
0.164(0.244)
-0.585 (0.337)
1.70* (0.237)
2.16* (0.068)
2.46* (0.091)
0.870* (0.373)
-1.03* (0.338)
-0.965* (0.162)
2.39* (0.099)
2.78* (0.052)
."V- Slope (/,
v -" j- &* • ;--
0.347t (0.064)
0.303t (0.100)
0.444T (0.098)
0.343 f (0.063)
0.1 33t (0.044)
0. 1 91 1 (0.042)
0.454t (0.073)
0.557 1 (0.131)
0.335 1 (0.1 18)
0.359T (0.058)
0.1 52t (0.036)
0.119t (0.042)
The regression model takes the form loglPbDA,) = fi + a(iog(PbDB,)( + H, + e,, where subscript i corresponds to the ith
housing unit, subscript j corresponds to the jth room within a housing unit, H, refers to the random effect associated with
the ith housing unit, e, refers to the random effect representing withirvunit variability and other-random error, and remaining
notation is specified in the column headings.

*  Significantly different from zero at the 0.05 level (indicating results for one method are consistently higher or lower than
results for the other method).

t  Significantly different from one at the 0.05 level (indicating the magnitude of differences between the two methods is a
function of the value of the predictor variable).
                                                 I-69
-------
REFERENCES FOR APPENDIX I

Adgate, J.L., Weisel, C, Rhoads, G.G., and Lioy, PJ.  (1995) "Lead in House Dust:
       Relationships Between Exposure Metrics and Sampling Techniques." Environmental
       Research. 70:134-147.

ASTM(1996) "Standard Practice for Collection of Roor Dust for Chemical Analysis."  1996
       Annual Book of ASTM Standards, ASTM D5438-94.11.03:521-527.

CDC (1997) "Update: Blood-Lead Levels - United States 1991-1994." Morbidity and Mortality
       Weekly Report. U.S. Department of Health and Human Services, Public Health Service,
       Centers for Disease Control and Prevention. 21 February 1997,46(7): 141-146.

Clark, S., Bomschein, R.L., Pan, W., Menrath, W<, Roda, S., and Grote, J. (1996) "The
       Relationship Between Surface Dust Lead Loadings on Carpets and the Blood Lead of
       Young Children." Environmental Geochemistry and Health. 18:143-146.

Emond, M.J., Lanphear, B.P., Watts, A., Eberly, S., and Members of the Rochester Lead-in-Dust
       Study Group. (1997) "Measurement Error and Its Impact on the Estimated Relationship
       Between Dust Lead and Children's Blood Lead.", Environmental Research 72:82-92

Lanphear, B.P., Weitzman, M., Winter, N.L., Eberly, S., Yakir, B., Tanner, M., Emond, M., and
       Matte, T.D.  (1996a) "Lead-Contaminated House Dust and Urban Children's Blood Lead
       Levels." American Journal of Public Health 86(10):1416-1421.

Lanphear, B.P., Weitzman, M., and Eberly, S. (1996b) "Racial Differences m Urban Children's
       Environmental Exposures to Lead." American Journal of Public Health 86(10): 1460-
       1463.

Lanphear, B.P., Emond, M., Jacobs, D.E., Weitzman, M., Tanner, M., Winter, N.L., Yakir, B.,
       and Eberly, S. (1995) "A Side-by-Side Comparison of Dust Collection Methods for
       Sampling Lead-Contaminated House Dust." Environmental Research 68:114-123.

NCLSH and UCDEH (1998) "Evaluation of the HUD Lead-Based Paint Hazard Control Grant
       Program" Fifth Interim Report to the U.S. Department of Housing and Urban
       Development by the National Center for Lead-Safe Housing and the University of
       Cincinnati Department of Environmental Health. March 1998.

Que Hee, S.S., Peace, B., Clark, C.S., Boyle, J.R., Bomschein, R.L., and Hammond, P.B. (1985)
       "Evolution of Efficient Methods to Sample Lead Sources, Such as House Dust and Hand
       Dust, in the Homes of Children." Environmental Research. 38:77-95.

Roberts, J.W., Budd, W.T., Ruby, M.G., Stamper, V.R.,  Camann, D.E., Fortmann, R.C.,
       Sheldon, L.S., and Lewis, R.G. (1991) "A Small High Volume Surface Sampler (HVS3)

                                        1-70
-------
      for Pesticides, and Other Toxic Substances in House Dust." In: Proceedings, Annual
      Meeting - Air and Waste Management Association.  Publication No. 91-150.2.

The Rochester School of Medicine, and NCLSH.  (1995) "The Relation of Lxsad-Contaminated
      House Dust and Blood Lead Levels Among Urban Children: Volumes I and H."
      Departments of Pediatrics, Biostatistics, and Environmental Medicine, The Rochester
      School of Medicine, Rochester, New York, and The National Center for Lead-Safe
      Housing, Columbia Maryland, June, 1995.

USEPA (1997a) "Summary and Assessment of Published Information on Determining Lead
      Exposures and Mitigating Lead Hazards Associated with Dust and Soil in Residential
      Carpets, Furniture, and Forced Air Ducts" Office of Pollution Prevention and Toxics,
      U.S. Environmental Protection Agency. EPA 747-S-97-001, December 1997.

USEPA (1997b) "Risk Analysis to Support Standards for Lead hi Paint, Dust, and Soil",
      Volumes I and DL Office of Pollution Prevention and Toxics, U.S. Environmental
      Protection Agency. EPA 747-R-97-006, December 1997.

USEPA (1997c) "Conversion Equations for Use in Section 403 Rulemaking" Office of Pollution
      Prevention and Toxics, U.S. Environmental Protection Agency. EPA 747-R-96-012,
      December 1997.

USHUD (1995) "Guidelines for the Evaluation and Control of Lead-Based Paint Hazards in
      Housing." Office of Lead-Based Paint Abatement and Poisoning Prevention, U.S.
      Department of Housing and Urban Development.

Wang, E., Rhoads, G.G., Wainman, T., and Lioy, P.J.  (1995) "Effects of Environmental and
      Carpet Variables on Vacuum Sampler Collection Efficiency." Applied Occupational
      Environmental Hygiene. 10(2): 111-119.
                                        1-71
-------
(This page left blank intentionally.)
               1-72
-------
                         APPENDIX 12

      DESCRIPTIVE SUMMARIES OF DATA ENDPOINT VALUES
UTILIZED IN THE CARPET DUST-LEAD DATA ANALYSIS OF APPENDIX
                             1-73
U.8. EPA Headquarters Library
      Mail code 3201
1200 Pennsylvania Avenue NW
   Washington DC 20460
-------
                                    APPENDIX 12
             DESCRIPTIVE SUMMARIES OF DATA ENDPOINT VALUES
      UTILIZED IN THE CARPET DUST-LEAD DATA ANALYSIS OF APPENDIX I

       In this appendix, data values for variables considered in the statistical analyses of
Appendix I are summarized across housing units to provide important information when
interpreting results of these analyses. Descriptive statistics such as the sample size (i.e., numbers
of housing units), arithmetic and geometric means, standard deviation, geometric standard
deviation, minimum, maximum, and selected percentiles were calculated for selected endpoints
from each study. Descriptive statistics on dust-lead variables were calculated within the data
categories noted in Table 13-1 of Section £3 (Appendix I).  The percentage of floor-dust samples
collected from carpeted floors within a housing unit was summarized across units to determine
the extent to which dust-lead data from carpeted surfaces were available for these units. When
summarizing blood-lead concentration data, the percentage of children with blood-lead
concentrations at or above a specified threshold (10, IS, or 20 ug/dL) was also summarized.

       Note that the summaries presented in this appendix may differ from similar summaries
presented in previously-published documents on these studies. This is due to  differences in the
subsets of data included in the analysis and in any transformations and summary calculations
performed on the data prior to analysis.

       While the descriptive statistics were calculated across all surveyed housing units in each
study, they were also calculated by grantee and by categories denoting the year in which the
housing units were built (pre-1940,1940-1959,1960-1977, post-1977) for the HUD Grantees
evaluation. As the specified year in which a housing unit was built may be unreliable in the
Rochester study, summaries of Rochester study data (and any subsequent analyses of these data)
did not consider age of housing unit.

ROCHESTER LEAD-IN-DUST STUDY

       Area-weighted average floor dust-lead loadings and mass-weighted average floor dust-
lead concentrations for the 205 housing units in the Rochester study are summarized in Tables
12-1 and 12-2, respectively, according to surface type (carpeted and uncarpeted floors) and dust
collection method. As seen in these tables, not all units had dust-lead data available for a given
dust collection method. The following conclusions can be made from these two tables:

       •     While carpeted floors had a substantially higher geometric mean average dust-lead
             loading relative to uncarpeted floors under the BRM (255 ng/ft2 versus  17.5
             ug/ft2), this disparity was considerably less for the DVM (4.51  ug/ft2 versus 1.28
             ug/ft2).  In contrast, little, if any, difference between carpeted and uncarpeted
             floors was seen in the geometric mean under the wipe (12.5 Hg/ft2 for carpeted
             floors versus  18.0 ug/ft2 for uncarpeted floors).
                                         1-74
-------
Table 12-1.   Summary Statistics of Area-Weighted Average Floor Dust-Lead Loadings
             (jig/ft2) Across Housing Units in the Rochester Study, According to Type of
             Surface and Oust Collection Method
Method
# Units
Arithmetic
' Mean.-
fSta.Dev.j
:• Geometric
f - " r>^-*"
\- Mean /-
} (Geometric
: Sfdrbev:)
Minimum
, 25th '
Percentile
Median
Carpeted Floors
BRM
DVM
Wipe
179
181
179
1210(4470)
33.2 (212)
141 (1340)
255 (4.95)
4.51 (4.81)
12.5 (3.09)
8.27
0.0500
0.810
82.7
1.90
8.35
266
4.18
13.
££'„ •SHUMP*.^. .—"•
^/otn^-i*
PercGntQe:
v*"* >~ "l*Et ™
"1>J, - * ~
Maximum'
.« - ."-si" -
r JJK'^jr*

627
9.18
19.1
47300
2680
17300
Uncarpeted Floors
BRM
DVM
Wipe
191
194
193
530 (5370)
10.6 (55.7)
134(1310)
17.5(7.91)
1.28(6.45)
18.0(3.12)
0.0800
0.0500
0.640
5.00
0.250
10.1
13.1
1.90
17.0
45.3
4.34
28.1
74100
690
18100
Table 12-2.   Summary Statistics of Mass-Weighted Average Floor Dust-Lead
             Concentrations (pglg) Across Housing Units in the Rochester Study,
             According to Type of Surface and Dust Collection Method
Method

BRM
DVM
f Units

Arithmetic
Mean
(Std. Dev.)
Geometric;
- , Mean .
" '(Geometric '-
Std. Dev.)
Minimum
25th
Percentile
Median
75th
Ba«M»m*«lA
f&fCvliuIe
Maximum
Carpeted Floors
178
177
500. (3040)
1290(9320)
131 (4.81)
148 (5.73)
1.00
1.00
72.0
78.2
Uncarpeted Floors
BRM
DVM
189
177
2310(8800)
1240(3890)
394 (6.44)
208(7.61)
1.76
1.00
157
49.6
163
164
353
381

406
318
1200
747
40600
119000

92000
35800
                                        I-75
-------
       •      For carpeted floors, the geometric mean dust-lead loading for samples collected
              by the BRM was an order of magnitude higher than under the DYM and wipe
              methods. This result was not observed for uncarpeted floors. The geometric
              mean dust-lead loading using the DVM was slightly lower than for the wipe
              method for both surface types.

       *      Little difference was observed in geometric mean floor dust-lead concentrations
              between the BRM and DVM samplers.

       •      For both dust-lead loadings and concentrations, the arithmetic mean is
              considerably larger than the geometric mean and the 75th percentile, indicating
              skewness in the data distribution. This is evidence of the need to take a
              transformation of the data, such as a logarithmic transformation, prior to analysis.

Higher dust-lead loadings associated with the BRM on carpeted surfaces is primarily due to its
high sampling velocity which removes a greater amount of the total dust (and lead) in the carpet
relative to the DVM and the wipe, which tend to remove only surface dust.

       Measured dust-lead loadings on carpeted floors can be affected by the height of the carpet
pile, as dust can be more difficult to sample from high-piled carpet.  Therefore, it would be of
interest to summarize carpet dust-lead loadings according to high-piled carpet versus low-piled
carpet within a housing unit.  However, only 9% of the 1,263 carpet-dust samples collected in the
Rochester study were from high-piled carpet  Of the 181 housing units in the Rochester study
with carpet-dust sample results, 20 units had at least one dust sample taken from high-piled
carpet and at least one from low-piled carpet.  Of these units, only two units had more than one
dust sample taken from high-piled carpet (both had two such samples collected). Therefore, a
lack of data precluded a summary of carpet dust-lead measurements by carpet height.

       Most of the carpet-dust samples in the Rochester study were collected from carpets rated
as being in average or good condition. Only 33 of the 181 housing units with carpet-dust sample
results had at least one such sample collected from a carpet in poor condition, with 15 of these
units having all carpet-dust samples (up to three such samples per unit) taken from carpets in
poor condition.

       Area-weighted average dust-lead loadings on window sills were used as predictor
variables for blood-lead concentration in the regression modeling analyses.  Table 12-3 presents
summaries of these endpoints by dust collection method. Although not used in the statistical
analyses, area-weighted average dust-lead loadings on window wells and mass-weighted average
dust-lead concentrations on window sills and window wells are also summarized in this table.
These summaries indicate the following:

       •      Lead levels on window components tend to be very high in both studies
              (especially for window wells and when using BRM or wipe collection techniques)
                                          1-76
-------
Table 12-3.   Summary Statistics of Weighted Average Dust-Lead Levels for Window Sills
              and Window Wells Across Housing Units in the Rochester Study, According
              to Dust Collection Method1
Motfcwi
s- **
•t.
*-
Units/;
Arithmetic Mean
, > (Std: Dev.)» J

I _ Mean _ tj%
(Geometric
" Std. Dev.)
Miftinttifii '
"*"» v> *-T Stu
i, * s~- "K, **\
=fi- S- it ^ *
-f f
, *25th~~
Per centfle ,
Median;
frf
2
- 75th >
i D__-L_«J|_
rUfUvfllllo
Vt'^ '
Maximum-
?" *'C,
"^ i fc •
Window Sill Dust-Lead Loadings U/g/ft2)
BRM
DVM
Wipe
196
198
196
4750(14100)
255(1510)
586 (1460)
362 (10.4)
27.1 -(7.1 6)
202 (3.97)
0.680
0.266
2.83
60.9
9.06
82.3
266
32.5
189
1610
80.5
434
11800
20000
14900
Window Sni Dust-Lead Concentrations (pg/g)
BRM
DVM
193
192
16800(43500)
3490 (9840)
2960 (8.70)
722 (7.23)
3.15
0.750
1030
222
3200
941
13600
2810
448000
97800
Window Well Dust-Lead Loadings (fig/ft2)
BRM
DVM
Wipe
188
190
189
243000 (456000)
6110(24600)
39200 (93000)
22700(21.7)
612(11.9)
4520(10.7)
6.86
0.210
28.5
1820
128
739
49800
676
4810
285000
4450
25500
3030000
303000
641000
Window Well Dust-Lead Concentrations (pg/g)
BRM
DVM
186
189
35000 (43600)
10500(32300)
8710(10.8)
2230 (8.36)
5.15
0.00
2140
550
19600
3010
50400
9860
207000
41300
1 In calculating weighted averages for each housing unit, loadings are weighted by area of sample, and concentrations are
weighted by mass of sample.
       •      A logarithmic transformation should be applied to these data prior to their
              inclusion in any statistical analyses.

Table 12-4 presents data summaries for other continuous endpoints used in statistical analyses,
such as average soil-lead concentration and the percentage of floor-dust sample area consisting of
carpet.  Although not used in the statistical analysis presented in Section 15, data on the 75th
percentile of XRF measurements in a housing unit are also summarized in Table 12-4.  Table 12-5
provides additional information on the percentage of floor-dust samples in a unit taken from
carpet  These two tables indicate the following:
                                            1-77
-------
Table 12-4.    Summary Statistics for Continuous Endpoints Other Than Dust-Lead
                Measurements, Across Housing Units in the Rochester Study
Endpoint
% of Floor
Sample Area
from Carpet1
% of Carpeted
Floor Sample
Area from
High-Pile
Carpet1
Soil-lead
concentration
(fine fraction)
75th percentile
of interior XRF
measurements
(mg/cm2)1
75th percentile
of exterior XRF
measurements
(mg/cm2)2
Blood-lead
concentration
U/g/dU
Age of Child
(years)
Cleaning
Frequency
Mouthing
Behavior3
#
Units

204



181



190

204


204


204

204
204
202
Arithmetic
1 MeaiV -
(Std. Dev.)

51.1 (26.8)



9.6 (24.8)



1120(1360)

1.88(5.10)


4.74 (8.04)


7.70 (5.14)

1.74(0.44)
0.73 (0.16)
0.19(0.14)
.Geometric
„ Mean .-
[Geometric *
Std. Dev.)

—



-



622 (3.36)

-


-


6.37(1.85)

-
-
-
*$* "*!
Minimum

0



0



12.3

0


0


1.40

1.01
0.25
0
* -f f
\ 2^th.
Percentile

33.3



0



380.

0


0


4.20

1.35
0.625
0.0625
. Median<

50



0



751

0


0


6.10

1.69
0.75
0.1875
: J75th "
:Perceatite

75



0



1330

1.35


8.50


9.70

2.13
0.8125
0.25
-rt *""* i>s "
: Maximum

100



100



10700

28.4


35.0


31.7

2.62
1
0.75
1 Calculated without regard to dust collection method.

1  XRF measurements less than 1.0 mg/cm1 or corresponding to surfaces with intact paint were set to zero prior to
determining this value. For this reason, geometric means were not calculated for this endpoint. The value of the interior
measurement endpoint was zero for 72% of the units, while the value of the exterior measurement endpoint was zero for
61% of the units.

3 One-sixteenth of the sum of the values assigned to the four variables denoting a child's frequency of putting mouth on
window sill, pacifier in mouth, soil in mouth, or thumb in mouth.  Each of these four variables have possible values of 0
(never), 1 (rarely), 2 (sometimes), 3 (often), or 4 (always).
                                                  I-78
-------
Table 12-5.   Numbers (and Percentages) of Housing Units in the Rochester Study With
              Specified Values for the Percentage of Total Sampled Floor Area from
              Carpet and the Percentage of Total Sampled Carpeted Floor Area from High-
              Pile Carpet
.\ ""••'
#of^-_
Unite1-
204
' #of s
Units1',
181
-- _;••„ ^n^l^Totd^Simni^FI^AieaTak^iFromCanMrt^'. i?*" *•&¥;'„
'-,
0%
23
(11.3%)
Between JOSfef
"~rianel'* * ""
(Induding)
. '25%'--;.
27
(13.2%)
r Between
- 25%
rand 50%
9
(4.4%)
•«* "«<•
" S0*
62
(30.4%)
Between 50%
" ~and * '
(Indudmg)
75% /
61
(29.9%)
,V*- ,%• -,3t
;f'Betweens3
1 75%»
and 100%.
12
(5.9%)
? -":" j_f> if'
= •"/;x«^-':<
100%Z
•" *«• i.
10
(4.9%)
Percent , of Tqtal , Carpeted Floor Area Taken from High-PHe Carpet- -^ ,- - , -
0%
153(84.5%)
, Between 0"% ,
ami" .
(including}
"X25%;
2
(1.1%)
rf-&V - "
; Between
" ,25%
and 50%
2
(1.1%)
*_
50%
14
(7.7%)
Between 50%
and
{Including}
75%
2
(1.1%)
Between
-75%
and 100%
0
(0%)
100%
8
(4.4%)
 ' Numbers of housing units having data for the given sample type.
       •      The observed distribution of average soil-lead concentration indicates that this
              variable should be log-transformed prior to inclusion in any statistical analyses.

       •      For both interior and exterior painted surfaces, over half of the housing units had
              at least 75 percent of its XRF paint measurements either 1) below 1.0 mg/cm2 or
              2) taken from a surface with intact paint.

       *      Housing units, on average, had 51 % of its floor-dust samples taken from carpet
              (without regard to dust collection method), with the majority of housing units
              having from 50-75% of floor-dust samples taken from carpeted surfaces.

       •      As approximately 84% of the 181 units with carpet-dust sampling had no samples
              taken from high-pile carpets, carpet height provides little discerning information
              for statistical analysis and was therefore not considered in further analyses.

       Lead-based paint hazard score, defined in Table 13-2 of Section D, was used in the
statistical analyses to indicate the extent to which deteriorated lead-based paint is present in a
housing unit and that the monitored child in the unit exhibits pica tendencies. For the Rochester
study, 188 housing units (92%) had a lead-based paint hazard score of 0, indicating that no
deteriorated lead-based paint was present, or that the resident child exhibits no pica tendencies.
Of the remaining 16 housing units, only five achieved the highest score of 2, indicating the
                                           1-79
-------
presence of deteriorated lead-based paint and the resident child exhibits pica tendencies at least
sometimes. Therefore, this score would not provide much predictive power in determining
blood-lead concentration in a child.

       The geometric mean blood-lead concentration data for 204 children in the Rochester
study was 6.37 jig/dL (Table 12-4). Further investigation shows that 48 (23.5%) of the children
had a blood-lead concentration at or above 10 ug/dL, while 16 (7.8%) were at or above IS ug/dL,
and 6 (2.9%) were at or above 20 Ug/dL.

HUP GRANTEES PROGRAM EVALUATION

       A total of 395 housing units across 13 grantees had data for both blood-lead concentration
and floor dust-lead loading in the September 1997 database. All but three of these units were
built prior to 1960, with 353 (89%) built prior to 1940 and 39 (10%) built from 1940-1959.  Only
one housing unit was built after 1977.  The large number of older housing units reduces the
usefulness of the year built categorization in predicting blood-lead concentration.

       Table I2-6a summarizes area-weighted arithmetic average of (untransformed) floor dust-
lead loadings according to surface type (carpeted and uncarpeted floors) and dust collection
method (wipe, DVM).  Tables I2-6b and I2-6c contain the same summary statistics as Table
I2-6a, but presented by year in which the housing unit was built and grantee, respectively.
Results from these three tables are as follows:

       •      The geometric mean wipe dust-lead loading across units was somewhat higher for
             uncarpeted floors (32.4 ug/ft2 across 390 units) than for carpeted floors (17.1
             pg/ft2 across 226 units). For carpeted floors, the geometric mean DVM dust-lead
             loading in 24 units (9.43 ug/ft2) averaged lower than the average wipe dust-lead
             loading in 226 units (17.1 ug/ft2). These trends were similar to those seen in the
             Rochester data summary in Table 12-1.

       •      The grantees differ in the percentage of housing units having all floor dust-lead
             loading measurements reported at a constant value, suspected to be the detection
             limit divided by the square root of two. This percentage is as high as 85% for 20
             Baltimore samples. This constant value also differs among the grantees.

       •      Arithmetic means are larger than the geometric means and medians, indicating
             right skewness in the data distribution.  This finding, along with additional data
             investigation, led to the conclusion that a logarithmic transformation would be
             made to these data prior to each statistical analysis.  The same conclusion was
             made for the Rochester study based on results in Table 12-1.
                                          1-80
-------
Table !2-6a.  Summary Statistics of Area-Weighted Average Floor Dust-Lead Loadings
             (fig/ft2) Across Housing Units in the HUD Grantees Program Evaluation,
             According to Type of Surface and Dust Collection Method
*** ^ isw^-s.
CoDeclfon
' Metftod"
*
" *r, *
" ft, of
Units
^ ' ^. ^ /^Wtoyfodj&Hnrf jii
Arithmetic
-Mean
(Std. Dev.l
Geometric
* Mean
(Geometric
Std. Dev;)
* £** , v ^"
Minimum
x>r Dust-Lead Loadings

•«. *
Median
A **s?fifft ej f „,, , _.»-
SJW^^I^-^e
•«• j^-*-^ ^ -••r
, 75th "
Percentile ,
Maximum;"
Carpeted Roors
Wipe
DVM

Wipe
226
24

390
62.7(341.7)
40.3 (77.9)
17.1 (3.2)
9.43 (6.18)
1.06
0.707
Uncarpeted Floors
93.1 (249.1)
32.4 (3.6)
0.511
10.0
1.94
15.9
10.2
25.0
31.0
4764.
350.
t
14.1
25.7
66.5
2600.
1  Only wipe dust samples were collected from uncarpeted floors.
Table !2-6b.  Summary Statistics of Area-Weighted Average Floor Dust-Lead Loadings
             (pg/ft2) Across Housing Units in the HUD Grantees Program Evaluation,
             According to Type of Surface, Dust Collection Method, and Age of Housing
             Unit
Year that the
Unit was Built
#of
Units
Area-Weight)
Arithmetic
Mean
(Std. Oev.)
Geometric
Mean,'"
(Geometric
Std. Dev.J


Minimum
; 25th- .
Percentife
Lead Loadings U/g/ft2) r
Median
75tfi
PercentBe
Maximum
Carpeted Wipe
Prior to 1940
1940-1959
1960- 1979
216
9
1
65.2 (349.4)
9.91 (5.34)
6.77 (-)
17.7(3.3)
8.61 (1.78)
6.77 (-)
1.06
3.54
6.77
11.8
5.01
6.77
17.6
9.00
6.77
26.5
13.6
6.77
. 4764.
17.7
6.77
Carpeted DVM
Prior to 1940
1940- 1959
15
9
62.1 (92.7)
4.09 (4.89)
20.9 (5.7)
2.50 (2.75)
0.707
0.707
9.49
1.41
24.0
2.28
Uncarpeted Wipe
Prior to 1940
1940-1959
1960- 1979
After 1977
349
38
2
1
98.5(261.3)
38.7 (57.9)
16.9(8.6)
440. (-)
34.0 (3.6)
20.4 (2.9)
15.7(1.7)
440. (-)
0.511
3.54
10.8
440.
16.0
11.3
10.8
440.
26.7
17.7
16.9
440.
98.0
5.00
350.
16.0

72.0
34.0
22.9
440.
2600.
293.
22.9
440.
                                         1-81
-------
Table !2-6c. Summary Statistics of Area-Weighted Average Floor Dust-Lead
            Loadings U/g/ft2) Across Housing Units in the HUD Grantees Program
            Evaluation, According to Type of Surface, Dust Collection Method,
            and Grantee
Grantee
#of
Units
„ _"" " * Area-Weightejd Average Floor D
Arithmetic
Mean
, tstd.
5 Devi)
Geometric
Mean
(Geometric
Std.Dev.)
Mini- f
mum
f '25*'"
DkftMtfKMtSlA"
Wt iwi lUIB
it$t*ttf£cid[ L03ffinQ5*(ijfj

Median
f z &
tt "E
%'fi1i»wtvw^w -i '"V*"*
Perc&riule
:W •#?
'Kfcte^ -;% .
>, x^se
;»mum
fiv~I *i- !-
s ' V
Mode1
' (%fof
Units)
* -s-^j. ft
Carpeted Wipe
Baltimore
Boston
California
Cleveland
Massachusetts
Minnesota
Rhode Island
Wisconsin
Milwaukee
Chicago
New York City
Vermont
20
14
10
40
25
70
15
5
2
7
12
6
21.1
(9.9)
18.7
(20.0)
11.8
(13.8)
192.
(758.)
45.6
(97.2)
21.3
(20.5)
57.9
(88.5)
8.51
(5.71)
6.63
(2.31)
16.7
(18.8)
22.7
(35.0)
295.
(669.)
19.9
(1.4)
12.5
(2.5)
7.70
(2.44)
26.8
(5.2)
14.8
(4.5)
18.2
(1.6)
22.4
(4.0)
6.99
(2.04)
6.43
(1.43)
11.3
(2.4)
7.61
(4.73)
45.5
(5.9)
17.7
4.51
3.54
3.54
1.06
14.1
5.08
3.54
5.00
5.30
1.50
20.5
17.7
5.00
3.54
10.5
6.30
14.1
5.66
3.54
5.00
5.30
2.25
20.5
17.7
10.8
5.00
18.6
12.5
14.1
17.0
6.20
6.63
8.50
3.39
21.2
17.7
21.2
13.6
67.2
40.0
17.7
47.5
14.4
8.27
22.2
37.4
28.1
58.0
78.0
46.8
4764.
481.
153.
291.
14.9
8.27
57.0
118.
1660.
17.7
(85.0%)
5.00
(28.6%)
3.54
(30.0%)
14.1
(20.0%)
1.06
(8.0%)
14.1
(51.4%)
5.66
(20.0%)
—
—
5.30
(28.6%)
1.50
(8.3%)
20.5
(33.3%)
Carpeted DVM
Alameda
County
California
Cleveland
15
3
2
45.7
(96.5)
9.22
(5.93)
56.1
(59.3)
5.92
(7.89)
8.11
(1.83)
37.2
(3.9)
0.707
5.00
14.1
1.41
5.00
14.1
2.28
6.67
56.1
24.0
16.0
98.0
350.
16.0
98.0
0.707
(13.3%)
-
-
                                     I-82
-------
                                            Table !2-6c.  (cont.)
"• Grantee ,'
A «»"-! *
•V M. S\
* it * iw "*
%ft£
r

**•>•*•< - xye&x? a.* 3i.
Aritnimric

m
:^fg|4f
'. l^'jV-^ijyjjj.:-.!' ,"*..•* I
'•H";'.y-i-"!V!Si""^'£i-';'!!
Median
SSi-p ';v,S:|;j
H
•
P
Carpeted DVM (cont.)
Minnesota
New York City
3
1
35.1
(28.1)
38.0 (-)
28.4
(2.2)
38.0 (-)
14.1
38.0
14.1
38.0
24.1
38.0
67.0
38.0
67.0
38.0
—
-
Uncarpeted Wipe
Alameda
County
Baltimore
Boston
California
Cleveland
Massachusetts
Minnesota
Rhode Island
Wisconsin
Milwaukee
Chicago
New York City
Vermont
31
48
30
17
46
32
94
29
5
2
19
27
10
50.6
(123.0)
58.5
(100.2)
137.
(376.)
16.9
(20.1)
200.
(372.)
166.
(470.)
74.8
(210.5)
72.7
(101.6)
103.
(104.)
11.4
(6.5)
37.1
(55.6)
32.4
(36.0)
178.
(168.)
15.9
(3.7)
32.6
(2.4)
44.1
(3.5)
10.8
(2.5)
70.4
(4.4)
37.6
(4.9)
33.0
(2.7)
37.0
(3.2)
40.6
(6.7)
10.4
(1.8)
22.1
(2.5)
16.8
(3.7)
100.
(4.)
3.54
17.7
5.83
3.54
3.54
4.50
14.1
5.66
3.54
6.77
6.29
0.511
20.5
7.07
17.7
20.0
5.00
26.1
10.6
16.1
16.4
7.90
6.77
10.9
6.46
21.2
10.3
19.6
29.4
10.2
64.7
33.6
24.1
40.3
116.
11.4
19.0
25.1
133.
28.5
41.0
90.4
20.2
165.
103.
53.5
72.2
134.
16.0
39.8
45.1
300.
640.
545.
2045.
84.4
1864.
2600.
1831.
440.
255.
16.0
252.
158.
448.
7.07
(19.4%)
17.7
(50.0%)
17.7
(13.3%)
5.00
(23.5%)
14.1
(6.52%)
4.50
(3.1%)
14.1
(24.5%)
5.66
(10.3%)
—
—
6.28
(5.3%)
0.511
(3.7%)
21.2
(20.0%)
1   The mode is the most frequently reported value for area-weighted average floor dust-lead loadings (the lowest value if
more than one mode exists) and is specified for grantees having data for more than five housing units.  It likely represents
the detection limit for the given grantee, divided by the square root of two.
                                                      I-83
-------
       The HUD Grantees program evaluation did not record information on the type of carpet
(e.g., high-piled versus low-piled) but did report on the condition of sampled surfaces. Of the
585 dust samples that were collected from carpets by wipe methods and that had lead loading
data, only 34 came from carpets reported to be in poor condition.

       Table 12-7 presents data summaries for other environmental and demographic variables,
some of which were included in the statistical analyses due to their likelihood of being associated
with blood-lead concentration. These variables include area-weighted average window sill and
window well dust-lead loadings, average soil-lead concentration (over dripline and play areas in
the yard), 75th percentile of XRF paint-lead measurements (Section A.1), age of child at blood
collection, household annual income, and child's mouthing behavior. Results in this table are the
following:

       •      The geometric means (across housing units) of average dust-lead loadings on
              window sills and window wells and average soil-lead concentration were similar
              to or slightly higher than those in the Rochester study (Tables 12-3 and 12-4).

       •      As soil sampling was optional in this program, only 77 of the 395 housing units
              had soil-lead concentration data reported at both the dripline and play areas.
              Thus, attempting to control for effects of soil-lead concentration in the statistical
              analyses results in a substantial reduction in the available numbers of housing
              units with sufficient data.

       •      Age of the children at blood collection ranged from 7 months to 8 years, with an
              average (and median) of approximately three years. Thus, approximately half of
              the blood-lead concentration data are for children older than 1-2 years, which was
              the population of interest in the §403 risk analysis.

       Lead-based paint hazard score, as defined in Table 13-2 of Section D, indicates the extent
to which deteriorated lead-based paint was present in a housing unit and that the monitored child
placed non-food objects in his/her mouth. In the HUD Grantees program evaluation, nearly 60%
of the housing units had the highest possible score of 2, indicating that deteriorated lead-based
paint was  present in the unit, and the monitored child put non-food objects in his/her mouth
several times per day or more.  In contrast, only 25% of the housing units had the lowest score of
zero, indicating that either no deteriorated lead-based paint was present or the monitored child
did not place non-food objects in his/her mouth. This is in contrast to the Rochester study, where
92% of housing units had a score of zero. As in the Rochester study, the lead-based paint hazard
score was used in the analyses rather than a direct measure of lead levels in paint.

       Blood-lead concentration data are summarized in Table 12-8 according to year in which
the housing unit was built, grantee, and ownership status, as well as across all units. Among
grantees, geometric mean blood-lead concentration was highest for Cleveland (13.9 Mg/dL),  and
lowest for California (3.14 ng/dL). This disparity is primarily due to the different criteria that
each grantee used to  select housing units. To further illustrate differences in blood-lead

                                           1-84
-------
Table 12-7.    Summary Statistics of Area-Weighted Average Window Sill and Window
               Well Dust-Lead Loadings (pg/ft2). Average Soil-lead Concentration (//g/g),
               75th Percentile of XRF Paint Measurements (mg/cm2). Age of Child, Annual
               Household  Income, and Mouthing Behavior for Housing Units and Children in
               the HUD Grantees Program Evaluation
Endpoint
Window Sill
Dust-Lead
Loading (pg/ft2)
Window Well
Dust-Lead
Loading fag/ft2)
Soil-Lead
Concentration
U/g/g)1
75th Percentile
of Interior

Paint XRF
Measurements
{mg/cm2)2
75th Percentile
of Exterior

Paint XRF
Measurements
!mg/cm2)2
Age of Child at
Blood
Collection
(years)
Annual
Household
Income ($}
• Mouthing
Behavior3
#of
Units
394
354
77



379




202


395


393
395
Arithmetic
Mean
(Std:Dev;)
2160.
(7050.)
26100.
(49000.)
1690.
(2000.)



(4.91)




(9.44)


3.14
(1-51)

18800
(14400.)
0.58
(0.39)
Geomelnc.
Mean .
{Geometric
Std. Dev.):
374.
(6.)
4690. (10.)
979.
(3.)



«




*~


-


"
-
•Minimum
7.85
4.95
39.5



0.0




0.0


0.61


0.0
0.0
|^ 25th

93.2
805.
534.



0.0




2.60


1.81


8814.
0.25
Median
352.
6300.
1085.

.

0.0




8.13


2.89


16000.
0.50
*s,"t^ '•£'**t?t* ~ix£$i>1"
€& t •& "* "*"Xssf
sPercentfli:
1168.
31950.
1930.



3.60




10.8


4.40


24000.
1
•Maximum'
j««.
78400.
621000.
12648.



26.0




56.9


8.41


112500.
1
1 Average of dripline and play area soil-lead concentration.

2 75th percentile of XRF paint-lead measurements in each unit, with XRF measurement for a given surface reset to zero
when the measurement is less than 1.0 mg/cm2, or the measurement is greater than or equal to 1.0 mg/cm2 but the paint on
the surface was considered intact.

3 One-fourth of the sum of values assigned to the two variables denoting the frequency of the child putting fingers in mouth
and toys/other objects in mouth.  Both variables have possible values of 0 (never or less than once per week), 1 (several
times per week), or 2 (several times a day or more).
                                              1-85
-------
Table 12-8.   Summary Statistics of Blood-Lead Concentration (pg/dU Across Housing
             Units in the HUD Grantees Program Evaluation, by Age of Housing Unit,
             Grantee, and Ownership Status1
. ,-
All Units
4rof
Units
395
Blood-Lead Coi
Arithmetic
* Mean
IStd. Dev.)
10.3 (7.8)
Geometric
Mean
(Geometric
Std.Dev.) ,
-, *. «•
7.76 (2.23)
*"t* -
~p; : ;4^;
Minimum
iff
0.707

[tcentraaon (|
• 41 1 * \ _
" ' -**fi *. " -
r.^tKV.
; Percehtfle
f-""
y ^X**r>,"
4.00
ig/du --> ;; J.^*-""":
*•=. —
Median-
? ~ r
8.00
*>!» *
: ,75^1
»DaMAn*Sitfk
rercenuie
- *>T —
"'*"
15.0
v iSj.af"-V* -»
" J^-*'.
Maxirrtum-
*^-
53.0
By Year in Which the Unit Was Built
Prior to 1940
1940-1959
1960-1977
After 1977

Alameda
County
Baltimore
Boston
California
Cleveland
Mass-
achusetts
Minnesota
Rhode Island
Wisconsin
Milwaukee
Chicago
New York
City
Vermont
353
39
2
1
10.6(7.90)
7.91 (6.00)
15.0(12.7)
11. OH

31
48
30
18
47
33
94
30
5
2
19
27
11
5.97 (5.50)
9.65 (6.26)
9.99 (5.72)
4.09 (3.29)
16.7 (9.99)
9.96 (6.22)
11.0(8.7)
11.4(7.2)
8.68 (4.97)
6.50 (0.71)
12.0(6.4)
5.37 (3.39)
12.8(4.4)
7.97 (2.21)
5.87 (2.27)
12.0(2.67)
11. OH
0.707
1.41
6.00
11.0
By Grantee
4.35 (2.20)
7.88(1.94)
8.48(1.81)
3.14 (2.08)
13.9(1.9)
8.17(1.92)
7.72 (2.52)
9.21 (2.04)
7.72(1.70)
6.48(1.12)
10.5(1.7)
4.77 (1.57)
12.1 (1.5)
1.41
2.00
3.00
1.41
3.00
3.00
0.707
2.00
4.00
6.00
3.00
2.00
6.00
4.50
3.54
6.00
11.0

3.00
5.50
6.00
1.41
10.0
4.00
4.00
6.00
6.00
6.00
8.00
4.00
10.0
8.00
6.00
15.0
11.0
15.0
12.0
24.0
11.0
53.0
26.0
24.0
11.0

4.50
7.00
8.50
3.25
14.0
9.00
8.00
10.0
8.00
6.50
11.0
5.00
13.0
5.90
14.0
14.0
6.00
23.0
16.0
15.0
17.0
8.40
7.00
14.0
5.00
16.0
24.8
29.0
24.0
12.8
53.0
27.0
37.0
29.0
17.0
7.00
28.0
19.0
20.0
By Ownership Status
Rent
Own
193
202
10.7 (7.5)
10.0(8.0)
8.30 (2.09)
7.27 (2.35)
1.00
0.707
4.90
4.00
9.00
8.00
15.2
14.0
37.0
53.0
1 Blood-lead data for only one child per housing unit were selected (see Section 3.2).
                                           1-86
-------
concentrations across grantees, Table 12-9 summarizes the frequency counts of children with
blood-lead concentration at or above 10,15, and 20 ug/dL according to grantee. For example,
79% of the 47 sampled children in Cleveland had blood-lead concentrations at or above 10
ug/dL, compared to a program-wide percentage of 44%.

       Table 12-10 summarizes the percentage of total sampled floor area from carpeted surfaces
under wipe collection methods by presenting numbers of units within specified ranges of
percentages. Table 12-11 contains additional descriptive statistics on the percentage of total
sampled floor area from carpeted samples.  Information obtained from these two tables includes
the following:

       •       A total of 169 of the 395 units did not sample from carpeted floors, while only 5
              units sampled from exclusively carpeted floors.

       •       Carpet sampling was more prevalent for units built prior to 1940 (compared to
              units built from 1940 -1959) and for the Cleveland grantee.

       •       On average, about 29% of floor areas sampled using wipes were carpeted across
              the 395 housing units.

Therefore, in general, the HUD Grantees program evaluation had fewer occurrences of floor-dust
samples taken from carpeted surfaces compared to the Rochester study (Tables 12-4 and 12-5). In
this analysis, percentage of floor-dust sampling from carpeted surfaces was used as a surrogate
for the percentage of carpeting in a housing unit.
                                          1-87
-------
Table 12-9.   Frequency Counts of Children in the HUD Grantees Program Evaluation with
             Blood-Lead Concentration Greater than or Equal to 10, 15 and 20//g/dL, by
             Grantee and Across All Grantees1
Grantee
* t ..* _^.t
Alameda County
Baltimore
Boston
California
Cleveland
Massachusetts
Minnesota
Rhode Island
Wisconsin
Chicago
New York City
Vermont
All Grantees
^Number 61 CWldren ':
* 10 //g/dL
6
19
13
2
37
15
42
17
1
11
2
9
174
**15pg78k >
4
10
7
0
23
10
27
8
1
3
1
5
99
vi 20//g/dL~
1
3
2
0
15
2
19
5
0
3
0
1
51
%of Chadren _ : * -/' :
2 10/jg/dL
19.4%
39.6%
43.3%
11.1%
78.7%
45.5%
44.7%
56.7%
20.0%
57.9%
7.4%
81.8%
44.1%
* 15/ig/di.
12.9%
20.8%
23.3%
0.0%
48.9%
30.3%
28.7%
26.7%
20.0%
15.8%
3.7%
45.5%
25.1%
tZOjtgldl^
3.2%
6.3%
6.7%
0.0%
31.9%
6.1%
20.2%
16.7%
0.0%
15.8%
0.0%
9.1%
12.9%
1  The frequency counts were based on 395 housing units (one child per housing unit). Total numbers of housing units
within each grantee are found in Table 12-6.
Table 12-10. Percentage of Total Sampled Floor Area from Carpeted Surfaces under Wipe
             Collection Techniques for Housing Units in the HUD Grantees Program
             Evaluation, by Age of Housing Unit and by Grantee
.•-';;:' .'.
All Units

Prior to 1940
1940-1959
#of
Units
395

353
39
: f Frequency Count of Percentage of Total Wipe Sampled Floor Area
i From Carpeted Surfaces (% of Total Units)
0%
169
(42.8%)

137
(38.8%)
30
(76.9%)
Between 0%
and
(Including)
25%
68
(17.2%)
Between
25%
and 50%
36
19.1%)
50%
32
(8.1%)
Between
50% and
(Including)
75%
57
(14.4%)
Between
75%
and 100%
28
(7.1%)
100%
5
(1.3%)
By Year in Which the Unit Was Built
66
(18.7%)
2
(5.1%)
34
(9.6%)
2
(5.1%)
30
(8.5%)
1
(2.6%)
54
(15.3%)
3
(7.7%)
28
(7.9%)
0
<0%)
4
(1.1%)
1
(2.6%)
                                          I-88
-------
Table 12-10.  (cont.)
i
5. ^-i. « *
-*• h-
y
r A
#ot
"Units
-a* •*-*•>*'• ^ _^ ~*r^* -•%?'
&M. v/saf-'FiwpwncyCotm
t*\v<*- yj?** *• «**.& i*' jj v»{
* ."vjf.?- "& 1 ' _VA^S9!
*-•»_ ,. _i»
';_ 'N'*-
?* ~ -V
0%V
Between 0%
™. f * *Sr
, :-and i'.
(Including)
"25% ~
*«fa?z$ HSWaNjfK^vr ^_ _ ~ _ ___ ^ ** ^^.w^S'JSi*^^ -£"*<-"
l,v»lsr'tSrCeiiU*ytt Onimsu'WIpe Admuiea MOOff jHleiarr
jf*i«S^*SMi«i«'^'S^ -,-A , .1s •_. ic.«*SlfiAi ''SSPW&W
i:Cjrfefedj:Sttffac**; (% of Total.JUnHs^.v^^^g
' JBW^? ;-
-Between
.
> arid 50%-
* *t _»* •
^f*^
HSJWV
hoo%.
SfeJ^
* » - «?-J»
By Year in Which the Unit Was Built (cont.)
1960-1977
After 1977
2
1
1
(50.0%)
1
(100%)
0
(0%)
0
(0%)
0
(0%)
0
(0%)
1
(50.0%)
0
(0%)
0
(0%)
0
(0%)
0
(0%)
0
(0%)
0
(0%)
0
(0%)
By Grantee
Alameda
County
Baltimore
Boston
California
Cleveland
Mass-
achusetts
Minnesota
Rhode Island
Wisconsin
Milwaukee
Chicago
New York
City
Vermont
31
48
30
18
47
33
94
30
5
2
19
27
11
31
(100%)
28
(58.3%)
16
(53.3%)
8
(44.4%)
7
(14.9%)
8
(24.2%)
24
(25.5%)
15
(50.0%)
0
(0%)
0
(0%)
12
(63.2%)
15
(55.6%)
5
(45.5%)
0
(0%)
3
(6.3%)
7
(23.3%)
4
(22.2%)
7
(14.9%)
11
(33.3%)
25
(26.6%)
4
(13.3%)
2
(40.0%)
0
(0%)
1
(5.3%)
2
(7.4%)
2
(18.2%)
0
(0%)
2
(4.2%)
1
(3.3%)
2
(11.1%)
3
(6.4%)
9
(27.3%)
11
(11.7%)
4
(13.3%)
0
(0%)
0
(0%)
2
(10.5%)
1
(3.7%)
1
(9.1%)
0
(0%)
7
(14.6%)
2
(6.7%)
1
(5.6%)
6
(12.8%)
0
(0%)
9
(9.6%)
1
(3.3%)
0
(0%)
2
(100%)
1
(5.3%)
2
(7.4%)
1
(9.1%)
0
(0%)
7
(14.6%)
4
(13.3%)
2
(11.1%)
6
(12.8%)
4
(12.1%)
20
(21.3%)
4
(13.3%)
3
(60.0%)
0
(0%)
2
(10.5%)
4
(14.8%)
1
(9.1%)
0
(0%)
1
(2.1%)
0
(0%)
0
(0%)
17
(36.2%)
0
(0%)
5
(5.3%)
1
(3.3%)
0
(0%)
0
(0%)
1
(5.3%)
3
(11.1%)
0
(0%)
0
(0%)
0
(0%)
0
(0%)
1
(5.6%)
1
(2.1%)
1
(3.0%)
0
(0%)
1
(3.3%)
0
(0%)
0
(0%)
0
(0%)
0
(0%)
1
(9.1%)
        I-89
-------
Table 12-11.  Summary Statistics of the Percentages of Total Sampled Floor Area from
             Carpeted Floors Across Housing Units in the HUD Grantees Program
             Evaluation, by Age of Housing Unit and by Grantee
1-
All Units,
All Samples
All Units, Wipe
Samples Only
#of
Units
395
395
1 "~ «~* - " lr *- --J l-r *4 * ~ -A-" "~ ^'sK.*-?^ '*fl
Percentage of TotaTSampfed Rooir, Area from CarpeteBrlioorsi(%}4^.^v*';
.. 1»& **' -*.~^n* *& ,£Z „* *»^Jt *•" M* h* *< T-*s jS^**^*-** r ifc ^S^V'^kr *«?-*&
'Arithmetic
,. Mean^ -
(Std. Dev.)
31.6(30.6}
28.6 (30.6)
- r~f-s_* _\.J
_~— f '"to •*
s r mi*1i_* 	 i_±1»«i[Lii-
MinimLnTk
* ^r j * n«*f
'**' * j
0.0
0.0
r
. i,^25th^c^,^
:Peroehtae J
0.0
0.0
fr. V
Median
25.0
24.7
i -t*'-,*."^
.'5^€P
Perceritileii
60.0
50.0
"• *• 3* •*leij«?'a «s-^?"i- ~~
» ' "> "*- ££
100
100
By Year in Which the Unit Was Built (wipe samples only)
Prior to 1940
1940-1959
1960-1979
After 1977
353
39
2
1
30.5 (30.6)
12.2(25.5)
25.0 (35.4)
0.0 (-)
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
25.0
0.0
25.0
0.0
50.0
0.0
50.0
0.0
100
100
50.0
0.0
By Grantee (wipe samples only)
Alameda County
Baltimore
Boston
California
Cleveland
Massachusetts
Minnesota
Rhode Island
Wisconsin
Milwaukee
Chicago
New York City
Vermont
31
48
30
18
47
33
94
30
5'
2
19
27
11
0.0 (0)
22.2 (28.9)
19.8 (25.7)
26.1 (31.4)
54.1 (32.7)
27.8 (24.8)
34.9 (28.0)
22.7 (28.3)
51.0(26.8)
50.0 (0)
19.2(28.6)
26.1 (33.2)
26.8 (32.9)
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
20.0
50.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
25.0
9.9
0.0
0.0
25.0
50.0
0.0
0.0
0.0
0.0
0.0
0.0
20.0
60.0
24.7
25.0
4.4
60.0
50.0
0.0
0.0
20.0
0.0
50.0
25.0
40.0
85.7
40.0
60.0
41.7
75.0
50.0
40.0
60.0
50.0
0.0
80.0
75.0
100
100
100
80.0
100
75.0
50.0
80.0
80.0
100
                                       1-90
-------
                     APPENDIX J

  ADDITIONAL PERFORMANCE CHARACTERISTICS ANALYSES,
WHERE CANDIDATE STANDARDS FOR LEAD IN PLAY-AREA SOILS
                  ARE CONSIDERED
-------
-------
                  Additional Performance Characteristics Analyses,
               Where Candidate Standards for Lead in Play-Area Soils
                                    Are Considered

Note:  This appendix was not included in the version of this report that EPA distributed for
       external peer review.

       This appendix is an extension to the performance characteristics analyses presented in
Section 6.1. As discussed in Section 6.1, EPA employed performance characteristics analysis as
a non-modeling approach to evaluating candidate §403 standards relative to their ability to detect
lead hazards in homes containing children with elevated blood-lead concentrations.

       The performance characteristics analyses in Section 6.1 evaluated candidate standards for
dust-lead loadings on uncarpeted floors and window sills, yard-wide average soil-lead
concentration, and the extent of deteriorated lead-based paint. After these analyses were
completed and documented in this report,  EPA wished to evaluate candidate soil-lead
concentration standards that distinguished between areas of the yard where children played and
other areas of the yard.  Therefore, the performance analysis approach in Section 6.1 was
repeated on data from the Rochester Lead-in-Dust study, where separate standards were
considered for soil in play areas and soil in other areas of the yard. The results of these analyses
are presented in this appendix.

       According to the soil sampling protocol developed for the Rochester Lead-in-Dust study,
a composite soil sample was to be collected within at least two feet of the foundation at each
housing unit participating in the study. Then, a second composite  soil sample was to be collected
from bare areas of the yard where it could be determined that a resident child frequently played.
Therefore, the Rochester study database distinguished between soil-lead concentrations collected
from play areas and along the foundation.

       As seen in Table 3-36 of the  §403  risk analysis report, play area soil-lead concentration
was specified for only 77 of the 205  housing units in the Rochester study. However, the soil
sampling protocol for this study implied that play area soil samples were collected only when
such areas contained bare soil. Therefore, the performance characteristics analyses presented in
this appendix assumed that homes containing no data for play area soil-lead concentration had no
bare soil in play areas, and therefore, the play area soil-lead concentration for these homes was
assumed to be 0 ppm.

       In this analysis, for a given housing unit in the Rochester study, the soil-lead
concentration in areas of the yard other than play areas was equivalent to the yard-wide average
soil-lead concentration  calculated for the analyses presented within Section 6.1. For a given
housing unit, this value was equal to the following:
                                           J-1
-------
       •      the average of play aiea and foundation soil-lead concentrations, if both were
              reported

       •      the play area soil-lead concentration, if it was reported but the foundation soil-lead
              concentration was not (assumes that no bare soil existed along the foundation)

       •      the foundation soil-lead concentration, if it was reported but the play area soil-lead
              concentration was not (assumes that no bare soil existed in play areas)

       *      0 ppm, if neither the play area nor the foundation soil-lead concentrations were
              reported (assumes that no bare soil was available anywhere in the yard).

       The performance characteristics analyses presented hi this appendix consist of calculating
estimates of sensitivity, specificity, positive predictive value, and negative predictive value, as
they were defined within Section 6.1. In particular, sensitivity represents the number of homes
with children having elevated blood-lead concentrations (i.e., blood-lead concentrations at or
above 10 ug/dL) that exceed at least one candidate standard. Also, 100% minus the negative
predictive value represents the percentage of homes at or above at least one standard that contain
children with elevated blood-lead concentrations.

       Table J-l contains results of performance characteristics analyses performed under the
following candidate standards:

       •      Uncarpeted floor dust-lead loading: 10,20,25,40,50,100 ug/ft2
       •      Window sill dust-lead loading: 250 ug/ft2
       *      Plav area soil-lead concentration: 250,400,1200,2000 ppm
       •      Soil-lead concentration in non-plav areas (see above): 400,1200,2000,3000,
              4000,5000 ppm.

In addition, Table J-l  considers candidate play area soil-lead concentration standards of 100,250,
400, 800,1000,1200,2000,  and 5000 ppm in situations where a play area soil-lead standard is
the only standard being considered (i.e., no other dust or soil standards are considered). Note that
the analyses within Table J-l do not consider whether deteriorated lead-based paint is present in
the housing units.

       Results in Table J-l show that when the only standard being considered is for play area
soil-lead, the likelihood of having homes with elevated blood-lead levels that are at or above die
candidate standard is quite low, even when the candidate standard is low.  In turn, the likelihood
of having elevated blood-lead children in homes that do not exceed the candidate standard is
quite high.  This is evidence that the other dust-lead and/or soil-lead standards should be
considered simultaneously with the play area soil-lead standard in order to achieve desired goals
for detecting homes with children having elevated blood-lead concentration.
                                           J-2
-------
Table J-1.    Results of Performance Characteristics Analysis Performed on Data for
             Housing Units in the Rochester Lead-in-Dust Study, for Specified Sets of
             Candidate Standards for Lead in Floor Dust, Window Sill Dust, Soil in Play
             Areas, and Soil in Non-Play Areas

Note: Houses in the Rochester study with no play area soil-lead concentration specified
were assumed to have no bare soil present in play areas, and therefore, their play area
soil-lead concentration values were set to 0 ppm in this analysis.  Non-play area soil-lead
is specified if either dripline or play area soil-lead is nonzero or if no visible soil is present
(when it is set to O ppm).
EBL  = elevated blood-lead {i 10//g/dU. LBP = Lead-based paint.
CA*
set
Hay
Area
son
.{ppm)
5000
2000
1200
1000
800
400
250
100
2000
2000
2000
2000
2000
2000
1200
1200
1200
1200
1200
1200

for Lea
Non-
Play
.Area
Soil
(ppm)








5000
5000
5000
5000
5000
5000
5000
5000
5000
5000
5000
5000
BV*A Gfwwli
din.:.
Win-
dow SB
Dust








250
250
250
250
250
250
250
250
250
250
250
250
a - •*
Mvto'* • **

Uncar-'
peted








100
50
40
25
20
10
100
50
40
25
20
10
? * ' .•"" ?


?•"
2-tr OllO jj,
Standsrdj
/Total #'
Unftsl-
q
-------
Table J-1. (cont.)
--.'Set
' '.Soap
*• •* nt t
400
400
400
400
400
400
250
250
250
250
250
250
2000
2000
2000
2000
2000
2000
1200
1200
1200
1200
1200
1200
400
400
400
400
400
400
nf f^muRlf irtii etnndnntu "*
^ «br,tie«t In ^ ^£. ,^fe?
NOJV
Area"-
Soi
(ppn»)
5000
5000
5000
5000
5000
5000
5000
5000
5000
5000
5000
5000
4000
4000
4000
4000
4000
4000
4000
4000
4000
4000
4000
4000
4000
4000
4000
4000
4000
4000
Dust
to/ftf
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
'Uncar-
""Dust,
100
50
40
25
20
10
100
50
40
25
20
10
100
50
40
25
20
10
100
50
40
25
20
10
100
50
40
25
20
10
ipk
Above"
At Least
T/Totailr
Unite'
88/185
92/185
97/185
106/185
119/185
155/185
94/185
98/185
103/185
110/185
123/185
156/185
72/185
76/185
81/185
93/185
106/185
148/185
74/185
78/185
83/185
94/185
107/185
148/185
88/185
92/185
97/185
106/185
119/185
155/185
.- J^l- 1 ~-
J'>sl»ri«w&v--"
$$£%$$•[•
28/44 (63.6%)
30/44 (68.2%)
33/44 (75.0%)
35/44 (79.5%)
37/44(84.1%)
43/44 (97.7%)
28/44 (63.6%)
30/44 (68.2%)
33/44 (75.0%)
35/44 (79.5%)
37/44(84.1%)
43/44 (97.7%)
26/44 (59.1 %)
28/44 (63.6%)
31/44 (70.5%)
34/44 (77.3%)
36/44(81.8%)
43/44 (97.7%)
28/44 (63.6%)
30/44 (68.2%)
33/44 (75.0%)
35/44 (79.5%)
37/44(84.1%)
43/44 (97.7%)
28/44 (63.6%)
30/44 (68.2%)
33/44 (75.0%)
35/44 (79.5%)
37/44(84.1%)
43/44 (97.7%)
',«. * - .-~~~~ -. --/
^TV^' ' ^^I^fSl
C Soecfficitv
» At-or Above No
T ' StaiMuiuTi
81/141 (57.4%)
79/141 (56.0%)
77/141 (54.6%)
70/141 (49.6%)
59/141 (41.8%)
29/141 (20.6%)
75/141 (53.2%)
73/141 (51.8%)
71/141 (50.4%)
66/141 (46.8%)
55/141 (39.0%)
28/141 (19.9%)
95/141 (67.4%)
93/141 (66.0%)
91/141 (64.5%)
82/141 (58.2%)
71/141 (50.4%)
36/141 (25.5%)
95/141 (67.4%)
93/141 (66.0%)
91/141 (64.5%)
82/141 (58.2%)
71/141 (50.4%)
36/141 (25.5%)
81/141 (57.4%)
79/141 (56.0%)
77/141 (54.6%)
70/141 (49.6%)
59/141 (41.8%)
29/141 (20.6%)
ffiSt „
> * (%);of Unte-i
Standard That'
Have'EBL "'i
Chadren* —
28/88(31.8%)
30/92 (32.6%)
33/97 (34.0%)
35/106 (33.0%)
37/119(31.1%)
43/155(27.7%)
28/94 (29.8%)
30/98 (30.6%)
33/103432.0%)
35/110(31.8%)
37/123(30.1%)
43/156(27.6%)
26/72(36.1%)
28/76 (36.8%)
31/81 (38.3%)
34/93 (36.6%)
36/106 (34.0%)
43/148(29,1%)
28/74 (37.8%)
30/78 (38.5%)
33/83 (39.8%)
35/94(37.2%)
37/107 (34.6%)
43/148(29.1%)
28/88(31.8%)
30/92 (32.6%)
33/97 (34.0%)
35/106 (33.0%)
37/119(31.1%)
43/155(27.7%)

IS
81/97 (83.5%)
79/93 (84.9%)
77/88 (87.5%)
70/79 (88.6%)
59/66 (89.4%)
29/30 (96.7%)
75/91 (82.4%)
73/87 (83.9%)
71/82 (86.6%)
66/75 (88.0%)
55/62 (88.7%)
28/29 (96.6%)
95/113(84.1%)
93/109 (85.3%)
91/104 (87.5%)
82/92 (89.1 %)
71/79 (89.9%)
36/37 (97.3%)
95/111 (85.6%)
93/107 (86.9%)
91/102 (89.2%)
82/91 (90.1 %)
71/78(91.0%)
36/37 (97.3%)
81/97 (83.5%)
79/93 (84.9%)
77/88 (87.5%)
70/79 (88.6%)
59/66 (89.4%)
29/30 (96.7%)
        J-4
-------
Table J-1. (cont.)
' ^Sfi

Play *
' Area*.
.- SOB"
(Hum)
250
250
250
250
250
250
20OO
2000
2000
2000
2000
2000
1200
1200
1200
1200
1200
1200
400
400
400
400
400
400
250
250
25O
250
250
250
Zrt*Z~KA,

Non- -
SoB
* K"
4000
4000
4000
4000
4000
4000
3000
3000
3000
3000
3000
3000
3000
3000
3000
3OOO
3000
3000
3000
3000
3000
3000
3000
3000
3000
3000
3000
3000
3000
3000

cMn ££ '
Winr
dowSfl
, Dust
\fftjff\f
s.
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
>nt_ »t

linear--
Boor
Dust
100
50
40
25
20
10
100
50 ,
40
25
20
1O
100
50
40
25
20
10
100
50
40
25
20
10
100
50
40
25
20
10
.« - t
•at InM.
* _"^ --
Above
'AfJeast
Stftmtorti
V
/Total #
Units1
94/185
98/185
103/185
110/185
123/185
156/185
75/185
79/185
84/185
95/185
108/185
149/185
77/185
81/185
86/185
96/1 85
109/185
149/185
90/185
94/185
99/185
108/185
121/185
156/185
96/1 85
100/185
105/185
112/185
125/185
157/185
\ '- " Afjr j.
*r; < :*£$$i*
~} SensHi Vity '
#<%) of Units
Children JThat
Are At or-
Above At
^ Least One
28/44 (63.6%)
30/44 (68.2%)
33/44 (75.0%)
35/44 (79.5%)
37/44(84.1%)
43/44 (97.7%)
26/44(59.1%)
28/44 (63.6%)
31/44 (70.5%)
34/44 (77.3%)
36/44 (81 .8%)
43/44 (97.7%)
28/44 (63.6%)
30/44 (68.2%)
33/44 (75.0%)
35/44 (79.5%)
37/44(84.1%)
43/44 (97.7%)
28/44 (63.6%)
30/44 (68.2%)
33/44 (75.0%)
35/44 (79.5%)
37/44(84.1%)
43/44 (97.7%)
28/44 (63.6%)
30/44 (68.2%)
33/44 (76.0%)
35/44 (79.5%)
37/44(84.1%)
43/44 (97.7%)
' :<^i«illLnanc«
'
fffSpecffiStv.
j ijowVP**8
fCn9drmi5tiBtAro
At or Above No
1 ^ $tflnd8fcP
75/141 (53.2%)
73/141 (51.8%)
71/141 (50-4%)
66/141 (46.8%)
55/141 (39.0%)
28/141 (19.9%)
92/141 (65.2%)
90/141 (63.8%)
88/141 (62.4%)
80/141 (56.7%)
69/141 (48.9%)
35/141 (24.8%)
92/141 (65.2%)
90/141 (63.8%)
88/141 (62.4%)
80/141 (56-7%)
69/141 (48.9%)
35/141 (24.8%)
79/141 (56.0%)
77/141 (54.6%)
75/141 (53.2%)
68/141 (48.2%)
57/141 (40.4%)
28/141 (19.9%)
73/141 (51.8%)
71/141 (50.4%)
69/141 (48.9%)
64/141 (45.4%)
53/141 (37.6%)
27/141 (19.1%)
Charact ., ., ^
*** -'-**z.j£$
PPV- Y?
» (%) of Unite
At or Above Ate'
Least One -^
.Standard That
HaveEBL '
' Children4 ,
28/94 (29.8%)
30/98 (30.6%)
33/103 (32.0%)
35/110(31.8%)
37/123(30.1%)
43/156(27.6%)
26/75 (34.7%)
28/79 (35.4%)
31/84(36.9%)
34/95 (35.8%)
36/108(33.3%)
43/149 (28.9%)
28/77 (36.4%)
30/81 (37.0%)
33/86 (38.4%)
35/96 (36.5%)
37/109 (33.9%)
43/149 (28.9%)
28/90(31.1%)
30/94(31.9%)
33/99 (33.3%)
35/108(32.4%)
37/121 (30.6%)
43/1 56 (27.6%)
28/96 (29.2%)
30/100 (30.0%)
33/105(31.4%)
35/112(31.3%)
37/125(29.6%)
43/157(27.4%)
H1*** Jfe^fr"-^
F^SIS^w
; ijjpvk "J?--
# (%jpJUrirK At
r~> NrtHave'St:
75/91 (82.4%)
73/87 (83.9%)
71/82 (86.6%)
66/75 (88.0%)
55/62 (88.7%)
28/29 (96.6%)
92/110(83.6%)
90/106(84.9%)
88/101(87.1%)
80/90 (88.9%)
69/77 (89.6%)
35/36 (97.2%)
92/108 (85.2%)
90/1 04 (86.5%)
88/99 (88.9%)
80/89 (89.9%)
69/76 (90.8%)
35/36 (97.2%)
79/95 (83.2%)
77/91 (84.6%)
75/86 (87.2%)
68/77 (88.3%)
57/64(89.1%)
28/29 (96.6%)
73/89 (82.0%)
71/85 (83.5%)
69/80 (86.3%)
64/73 (87.7%)
53/60 488.3%)
27/28 (96.4%)
        J-5
-------
                                        Table J-1.  (cont.)
iY v  i T*nrm I,-**'*. . • ••• •<•••» mm*r*ipJ.—•• •• nm •»• %^nr,^;.-.». .: ^
'•- " • (ii • """.ix=i;'••-'•!* '""•'is!! '•'" -i t^s*^**^-'"^"™'- ;£••;•• SwiiB:^1-*;"^?"1!^


y^S:]iS^^^ffi:?Ss3iagf|£
        H8
        SsB
        ffi'&Slffii
        fipif
       sivip
       iwiiii'
       ,. nriSsAs:iix^a. =i«


      ?U«ciiS?lJ
      tjjeJtglSxii™!'^1
                                         rja<;< /is;"-'-. ,.!.),••**¥«*•?.•'!*•*•'•••;•. '— "<•&«>•••"•£•&• £L,f,tjpr.i.•'•••>, m "w.:-<:.-i?-iwi.~^"';."iy==3*??*.'.-' •- ••*«as5£:«a
                                         ilS;?PfiS|tilii1H"fi^s;I^Bwn^^
                                                          "Standard?  < 5:;

                                                      •sS!:^4s;;;;,;6-Siii:iJLniiis!i5
                                                      |Kisyi=i^p«3Sg;|fii!|gj
                                                      j.r-Si-.:'S''-i-if:i,--sa:>ir<:^>-C"-E-;>»!a£K'

 2000
2000
250
100
79/185
27/44(61.4%»
89/141 (63.1%)
 27/79 (34.2%)
89/106 (84.0%)
 2000
2000
250
so
83/185
29/44 (65.9%)
87/141 (61.7%)
 29/83 (34.9%)
87/102 (85.3%)
 2000
2000
250
40
88/185
32/44 (72.7%)
85/141 (60.3%)
 32/88 (36.4%)
 85/97 (87.6%)
 2000
2000
250
25
99/185
35/44 (79.5%)
77/141 (54.6%)
 35/99 (35.4%)
 77/86 (89.5%)
 2000
2000
250
20
112/185
37/44(84.1%)
66/141 (46.8%)
37/112(33.0%)
 66/73 (90.4%)
 2000
2000
250
 10
152/185
43/44 (97.7%)
32/141 (22.7%)
43/152 (28.3%)
 32/33 (97.0%)
 1200
2000
250
100
81/185
29/44 (65.9%)
89/141 (63.1%)
 29/81 (35.8%)
89/104 (85.6%)
 1200
2000
250
50
85/185
31/44(70.5%)
87/141 (61.7%)
 31/85 (36.5%)
87/100 (87.0%)
 1200
2000
250
40
90/185
34/44(77.3%)
85/141 (60.3%)
 34/90 (37.8%)
 85/95 (89.5%)
 1200
2000
250
 25
100/185
36/44(81.8%)
77/141 (54.6%)
36/100 136.0%)
 77/85 (90.6%)
 1200
2000
250
20
113/185
38/44 (86.4%)
66/141 (46.8%)
38/113(33.6%)
 66/72(91.7%)
 1200
2000
250
 10
152/185
43/44 (97.7%)
32/141 (22.7%)
43/152(28.3%)
 32/33 (97.0%)
  400
2000
250
100
92/185
29/44 (65.9%)
78/141 (55.3%)
 29/92(31.5%)
 78/93 (83.9%)
  400
2000
250
50
96/185
31/44(70.5%)
76/141 (53.9%)
 31/96 (32.3%)
 76/89 (85.4%)
  400
2000
250
40
101/185
34/44 (77.3%)
74/141 (52.5%)
34/101 (33.7%)
 74/84(88.1%)
  400
2000
250
 25
110/185
36/44(81.8%)
67/141 (47.5%)
36/110(32.7%)
 67/75 (89.3%)
  400
2000
250
 20
123/185
38/44 (86.4%)
56/141 (39.7%)
38/123(30.9%)
 56/62 (90.3%)
  400
2000
250
 10
157/185
43/44 (97.7%)
27/141 (19.1%)
43/157(27.4%)
 27/28 (96.4%)
  250
2000
250
100
98/185
29/44 (65.9%)
72/141 (51.1%)
 29/98 (29.6%)
 72/87 (82.8%)
  250
2000
250
 50
102/185
31/44(70.5%)
70/141 (49.6%)
31/102(30.4%)
 70/83 (84.3%)
  250
2000
250
 40
107/185
34/44 (77.3%)
68/141 (48.2%)
34/107(31.8%)
 68/78 (87.2%)
  250
2000
250
 25
114/185
36/44 (81 .8%)
63/141 (44.7%)
36/114(31.6%)
 63/71 (88.7%)
  250
2000
250
 20
127/185
38/44 (86.4%)
52/141 (36.9%)
38/127(29.9%)
 52/58 (89.7%)
  250
2000
250
 10
158/185
43/44 (97.7%)
26/141 (18.4%)
43/1 58 (27.2%)
 26/27 (96.3%)
  2000
1200
250
100
 91/185
33/44 (75.0%)
83/141 (58.9%)
 33/91 (36.3%)
 83/94 (88.3%)
 2000
1200
250
 50
 95/185
35/44 (79.5%)
81/141 (57.4%)
 35/95 (36.8%)
 81/90 (90.0%)
 2000
1200
250
 40
100/185
38/44 (86.4%)
79/141 (56.0%)
38/100(38.0%)
 79/85 (92.9%)
  2000
1200
250
 25
107/185
38/44 (86.4%)
72/141 (51.1%)
38/107 (35.5%)
 72/78 (92.3%)
 2000
1200
250
 20
118/185
39/44 (88.6%)
62/141 (44.0%)
39/118(33.1%)
 62/67 (92.5%)
 2000
1200
250
 10
155/185
43/44 (97.7%)
29/141 (20.6%)
43/155(27.7%)
 29/30 (96.7%)
                                                 J-6
-------
Table J-1. (cont.)
m
-, ,, Set r
R»v*
Area«";
- Sta'l
{ppm>
1200
1200
1200
1200
1200
1200
400
400
400
400
400
400
250
250
250
250
250
250
2000
2000
2000
2000
2000
2000
1200
1200
1200
1200
1200
1200

3T oanoicu
Mott¥

- SoT,
(ppni)
1200
1200
1200
1200
1200
1200
1200
1200
1200
1200
1200
120O
1200
1200
1200
1200
1200
1200
400
400
400
400
400
400
400
400
400
400
400
400
. „ ^3.
rte Stand;
?3ttj|;
powSBl
* 1. '
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
250
,« c£--~

;0pc»-t
'rHook*
: Du%-
1OO
50
40
25
20
10
100
50
40
25
20
10
100
50
40
25
20
10
100
50
40
25
20
10
100
50
40
25
20
10
it JBfcMulsiw'
t # units '
r* Above
'At'Least"
One*.
'Standard',
I/fotBfr#
Units1'
91/185
95/185
100/185
107/185
118/185
155/185
102/185
106/185
111/185
117/185
128/185
160/185
108/185
112/185
117/185
121/185
132/185
161/185
145/185
145/185
146/185
147/185
154/185
170/185
145/185
145/185
146/185
147/185
154/185
170/185
-V -V* "tiV -
J/f ;-}.*4':
;-^Senshwitvrf
5#(%J of Units
'.ChildrefiT;itat
' Above At-
' CoastOne -
Standard^ -
33/44 (75.0%)
35/44 (79.5%)
38/44 (86.4%)
38/44 (86.4%)
39/44 (88.6%)
43/44 (97.7%)
33/44 (75.0%)
35/44 (79.5%)
38/44 (86.4%)
38/44 (86.4%)
39/44 (88.6%)
43/44 (97.7%)
33/44 (75.0%)
35/44 (79.5%)
38/44 (86.4%)
38/44 (86.4%)
39/44 (88.6%)
43/44 (97.7%)
41/44(93.2%)
41/44 (93.2%)
42/44 (95.5%)
42/44 (95.5%)
42/44 (95.5%)
43/44 (97.7%)
41/44(93.2%)
41/44(93.2%)
42/44 (95.5%)
42/44 (95.5%)
42/44 (95.5%)
43/44 (97.7%)

,3$ Performance^
- Soecfficftv
T-?- f x, •, ~
#<%) of Units
ChOdren-That Are
" At or Above Ho
83/141 (58.9%)
81/141 (57.4%)
79/141 (56.0%)
72/141 (51.1%)
62/141 (44.0%)
29/141 (20.6%)
72/141 (51.1%)
70/141 (49.6%)
68/141 (48.2%)
62/141 (44.0%)
52/141 (36.9%)
24/141 (17.0%)
66/141 (46.8%)
64/141 (45.4%)
62/141 (44.0%)
58/141 (41.1%)
48/141 (34.0%)
23/141 (16.3%)
37/141 (26.2%)
37/141 (26.2%)
37/141 (26.2%)
36/141 (25.5%)
29/141 (20.6%)
14/141 (9.9%)
37/141 (26.2%)
37/141 (26.2%)
37/141 (26.2%)
36/141 (25.5%)
29/141 (20.6%)
14/141 (9.9%)

^iiatacUHisucsv;? E
. wviaV^
# (%)'of tfnte'i
At ori Above At;
•™K^I "*
Chadrent
-. ^x ^ ia
33/91 (36.3%)
35/95 (36.8%)
38/100(38.0%)
38/107(35.5%)
39/118(33.1%)
43/155 (27.7%)
33/102 (32.4%)
35/106(33.0%)
38/1 1 1 (34.2%)
38/117(32.5%)
39/128(30.5%)
43/160(26.9%)
33/108(30.6%)
35/112(31.3%)
38/117(32.5%)
38/121 (31.4%)
39/132(29.5%)
43/161 (26.7%)
41/145 (28.3%)
41/145(28.3%)
42/146(28.8%)
42/147 (28.6%)
42/154(27.3%)
43/170(25.3%)
41/145 (28.3%)
41/145 (28.3%)
42/146(28.8%)
42/147 (28.6%)
42/154(27.3%)
43/170(25.3%)
f< „_?
« ^T*
^ !
?:r-~vi®!vV"--'r "
^ ^;T£«»I
^"NotHaveBl
^^ ChfldrWf
83/94 (88.3%)
81/90(90.0%)
79/85 (92.9%)
72/78 (92.3%)
62/67 (92.5%)
29/30 (96.7%)
72/83 (86.7%)
70/79 (88.6%)
68/74(91.9%)
62/68(91.2%)
52/57(91.2%)
24/25 (96.0%)
66/77 (85.7%)
64/73 (87.7%)
62/68(91.2%)
58/64 (90.6%)
48/53 (90.6%)
23/24 (95.8%)
37/40 (92.5%)
37/40 (92.5%)
37/39 (94.9%)
36/38 (94.7%)
29/31 (93.5%)
14/16(93.3%)
37/40 (92.5%)
37/40 (92.5%)
37/39 (94.9%)
36/38 (94.7%)
29/31 (93.5%)
14/15(93.3%)'
        J-7
-------
                                           Table J-1.  (cont.)
Set of Candidate Standards^
           for Lewi ta
  Area

  to*1*
Non-
Play
Area
              Wm-
             dowsa
              Dust
                           linear-
•F&
 Dust,
team2)
                                   # Units
-Above-
AtiUast
" One
Standard*

•/Total #
  Units1
                                      Petfofinance.Chj
                                          "\ ." *!
                                                                       teristics
                                                                            "
i»SensitivitY  *

lfc(%) of Units
                                            i* -'«-"* •J«-*'4;:
                                            '   Are-Ator
                                               Above Ar
                                                                  At oriAbove.At?
                                                                     Least One;
                                                                   Standwl That
                                                                     Have^.
                                                                     ChBdren*  ,
  4OO
400
              250
  100
 147/185
41/44(93.2%)
35/141 (24.8%)
41/147 (27.9%)
35/38 (92.1 %)
  400
400
              250
  50
147/185
41/44(93.2%)
35/141 (24.8%)
41/147 (27.9%)
35/38(92.1%)
  400
400
               250
  40
148/185
42/44 (95.5%)
35/141 (24.8%)
42/148 (28.4%)
35/37 (94.6%)
  400
400
              250
  25
 149/185
42/44 (95.5%)
34/141 (24.1%)
42/149(28.2%)
34/36 (94.4%)
  400
400
              250
  20
156/185
42/44 (95.5%)
27/141 (19.1%)
42/156 (26.9%)
27/29(93.1%)
  400
400
               250
  10
 171/185
43/44 (97.7%)
 13/141 (9.2%)
43/171 (25.1%)
13/14(92.9%)
  250
400
              250
  100
 147/185
41/44 (93.2%)
35/141 (24.8%)
41/147 (27.9%)
35/38(92.1%)
  250
400
              250
  50
 147/185
41/44(93.2%)
35/141 (24.8%)
41/147 (27.9%)
35/38(92.1%)
   250
400
               250
  40
 148/185
42/44 (95.5%)
35/141 (24.8%)
42/148 (28.4%)
35/37 (94.6%)
  250
400
              250
  25
 149/185
42/44 (95.5%)
34/141 (24.1%)
42/149 (28.2%)
34/36 (94.4%)
  250
400
              250
  20
 156/185
42/44 (95.5%)
27/141 (19.1%)
42/156(26.9%)
27/29(93.1%)
   250
400
               250
  10
 171/185
43/44(97.7%)
 13/141 (9.2%)
43/171 (25.1%)
13/14 (92.9%)
1 Total number of units having available data that could be compared to all specified candidate standards.
1 Cell entries arefnumber of homes at or above at least one standard that have EBL children)/ number of homes containing EBL children),
followed by the corresponding percentage (in parentheses).
9 Cell entries are (number of homes not at or above at least one standard that do not have EBL childrenl/ttotal number of homes not
containing EBL children!, followed by the corresponding percentage (in parentheses).
* Cell entries are (number of homes at or above at least one standard that have EBL children)/(total number of homes at or above at least
one standard), followed by the corresponding percentage (in parentheses).
1 Cell entries are (number of homes not at or above at least one standard that do not have EBL children)/(totel number of homes not at or
above any standard), followed by the corresponding percentage Cm parentheses).
                                                     J-8
-------
       Table J-2 contains results of performance characteristics analyses performed under the
following candidate standards:
       *      Uncarpeted floor dust-lead loading: 40,50 ug/ft2
       •      Window sill dust-lead loading: 250 ug/ft2
       •      Play area soil-lead concentration! 400 ppm
       •      Soil-lead concentration in non-plav areas: 400,800,1200,1600,2000,3000 ppm.

Unlike Table J-l, Table J-2 (like Table 6-8 in Section 6.1 of the report) documents the extent of
deteriorated lead-based paint that is present in housing units that contain an elevated blood-lead
child but are not at or above at least one of the candidate dust or soil standards. This information
suggests which of these housing units would possibly exceed a standard on the amount of
deteriorated lead-based paint and which would not. (Recall that the information in the Rochester
study database on amount of deteriorated lead-based paint was not in a format that allowed direct
comparisons to candidate standards on deteriorated lead-based paint that were considered for the
§403 rule, and as a result, deteriorated lead-based paint needed to be handled in this manner in
the analysis.)

       Note that Table J-2 differs from Table 6-8 of Section 6.1 in that candidate soil-lead
standards exclusively for play areas has been added to the set of standards. For example, at a
yardwide average soil-lead concentration standard of 1200 ppm, a window sill dust-lead standard
of 250 ug/ft2, and a floor dust-lead standard of 40 ug/ft2, only 100 of 184 homes exceeded at least
one of these standards (Table 6-8), compared to 111 homes when a play area soil standard of 400
ppm is added to  these three standards (Table J-2; where the yardwide average soil-lead standard
is interpreted as  a non-play area soil-lead standard). However, among the 11 additional homes
triggered when a play area soil standard of 400 ppm was added to these standards, none had
elevated blood-lead children. That is, sensitivity was not affected in this instance when adding
the play area standard, and negative predictive value decreased slightly (from 92.9% to 91.8%).
If the yardwide average soil-lead  standard is increased to 2000 ppm, an additional 13 homes are
triggered when a play area soil-lead standard of 400 ppm is added (from 88 to 101 homes; Tables
6-8 and J-2).  Of these 13 homes, 2 contain elevated blood-lead children.
                                           J-9
-------
J-10
-------

            »* I -4 .V'

           «*
#
wi
^

1



cj

-b

_»

a
          -e<^
S
          -
      i s
        '
I



I'
tn
a
0
i
           IIIl
                   •*

                   s
                   tM
                   g*

                   2 
                   to S
                   * 2
                   ^ CO
                   •* 21
II
£ 10
n SJ
                    00
                   o
                   10
                    s
8
    O)
    <- en
    CO J2
I?
li
                        CO
                        cS
    o
    Id
    o
    la
    CN
    i
                            (O
                            «N
        6
        ID
        CM
        ss
    si
    5 ri
    m <2
                            CD 2
                            (D
                            0
    O
    ID
    O
    ID
    CM
    o
    o
        CM
        IO
        CM
              *
34/100

(34.0%)
            o
            o
                                    o
                                    10
                                    s
                                    CM
                                    CM
                                    CM
            ID uj
            1*. jo
                                     to
                                    iS co
                                    f»
                5
            CD
            O
S
o
IO
CM
                                        8
                                        CM
                                        00
                                        CM
                                            CD
                                            CO
                                            CM
                                            ? CO
                                            s *
                                        CO
                                        CD
S
o
IO
CM
                O

                8
                CO
    1
                             J-11
                                            1
                          •° o
                          2*
                          II
                         - ^ 5
                         IS |
                         8^5

                         HI
                         is:
                              I
                          •s
                          .S

                          i

                          I
                          a,
                      „ «  ?
                     ?g*  S
                     • 8 S  -o
                     «J -o —  '
                     |S8  J
                     S=iss.§
                     g " S « 5
                     g-g-3^

                     8r-2'
                                                    s,
                                                    o
                                                    i
                                                    i
                                                    I
                                                    I
                         |!!s! i
                         pill 1
                         Mils I
        -£-s°i  -
        *> "2 a 2 "Q  c

        llpl  I
        s s-i -a o  •«=
      C i T3 O


      illl
      S ~ 3 =
                O


                ^
        iiifiii
        3-lsl-i^
        llol|||
        JocSl^S
                                               -
                                              -» 8 2 I
                                             s S"g s-S£ o

                                             5Si*il1

                                             lIHlii
                                             C 	 > n O £ »
                                                           I  'I
                                        II
                                        w

                                        1
                         « 8 2 «2*t

                         f o 1 o£"c'
-------
-------