America’s Children and the Environment Biomonitoring PCBs March, 2013

Biomonitoring: PCBs

Methods

Indicator

B7. PCBs in women ages 16 to 49 years: Median concentrations in blood serum, by
race/ethnicity and family income, 2001-2004.

Summary

Indicator B7 is the median concentrations of the sum of the four PCBs for women ages 16 to 49
years for 2001-2004, stratified both by race/ethnicity and family income. The median is the
estimated concentration such that 50% of all noninstitutionalized civilian women ages 16 to 49
years during the survey period have a sum of the four PCB concentrations below this level; a
birth rate-adjusted distribution of women's PCB levels is used in calculating this indicator,
meaning that the data are weighted using the age-specific probability of a woman giving birth.
Table B7a presents the 95th percentile concentrations of the sum of the four PCBs for women
ages 16 to 49 years for 2001-2004, stratified both by race/ethnicity and family income. The 95th
percentile is the estimated concentration such that 95% of all noninstitutionalized civilian women
ages 16 to 49 years during the survey period have a sum of the four PCB concentrations below
this level; the population distribution was weighted to account for over-sampling, non-response,
and non-coverage.

Data Summary

Indicator

B7. PCBs in women ages 16 to 49 years: Median concentrations
in blood serum, by race/ethnicity and family income, 2001-2004.

Time Period

2001-2004

Data

Serum

3CB (lipid adjusted) for four PCBs

Years/PCB

2001-
2002/
118

2001-
2002/
138

2001-
2002/
153

2001-
2002/
180

2003-
2004/
118

2003-
2004/
138

2003-
2004/
153

2003-
2004/
180

Limits of Detection (ng/g
lipid)*

10.5

0.6

0.4

1.1

0.4

1 Centers for Disease Control and Prevention. 2009. Fourth National Report onHuinan Exposure to Environmental
Chemicals. Atlanta, GA. Available at: www.cdc.qov/exposurereport.

America's Children and the Environment
March, 2013

Page 1

-------
Biomonitoring: PCBs

Indicator

B7. PCBs in women ages 16 to 49 years: Median concentrations
in blood serum, by race/ethnicity and family income, 2001-2004.

Number of Values

700

700

700

700

627

627

627

627

Number of Non-missing

644

641

644

643

519

520

519

520

Values**

(92%)

(92%)

(92%)

(92%)

(83%)

(83%)

(83%)

(83%)

Number of Missing

56

59

56

57

108

107

108

107

Values**

(8%)

(8%)

(8%)

(8%)

(17%)

(17%)

(17%)

(17%)

Percentage Below Limit of

50

18

11

32

0

0

0

0

Detection***

















*	The Limit of Detection (LOD) is defined as the level at which the measurement has a 95% probability of being greater than
zero. LODs vary among samples. The highest LOD among all the samples is shown.

*	*Non-missing values include those below the analytical limit of detection (LOD), which are reported as LOD/V2. 1,158 sampled
women 16 to 49 years had non-missing values for all four PCB congeners. 6 sampled women 16 to 49 years had missing values
for some but not all of the four PCB congeners. Missing values are the number of sampled women ages 16 to 49 years in the
Mobile Examination Center (MEC) sub-sample that have no value reported for the particular variable used in calculating the
indicator.

*	"This percentage is survey-weighted using the NHANES MEC survey weights for the given period and is weighted by age-
specific birth rates.

Overview of Data Files

The following files are needed to calculate this indicator. The files together with the survey
documentation and SAS programs for reading in the data are available at the NHANES website:
http ://www. cdc. gov/ nchs/ nhanes. htm.

•	NHANES 2001-2002: Demographic file demo_b.xpt. Laboratory 28POC file 128poc_b.
The demographic file demob.xpt is a SAS transport file that contains the subject
identifier (SEQN), age (RIDAGEYR), sex (RIAGENDR), race/ethnicity (RIDRETH1),
poverty income ratio (INDFMPIR), pseudo-stratum (SDMVSTRA) and the pseudo-PSU
(SDMVPSU). The laboratory file 128poc_b.xpt contains SEQN, the lipid-adjusted PCB
118, 138, 153, and 180 (LBX118LA, LBX138LA, LBX153LA, LBX180LA), the PCB
118, 138, 153, and 180 non-detect comment codes (LBD118LC, LBD138LC,

LBD153LC, LBD180LC), and the PCB sub-sample two year Mobile Examination Center
(MEC) weight (WTSP02YR). The two files are merged using the common variable
SEQN.

•	NHANES 2003-2004: Demographic file democ.xpt. Dioxins, Furans, and Coplanar
PCBs Laboratory file 128dfp_c.xpt. The Non-dioxin-like PCB Laboratory file 128npb_c.
The demographic file demo c.xpt is a SAS transport file that contains the subject
identifier (SEQN), age (RIDAGEYR), sex (RIAGENDR), race/ethnicity (RIDRETH1),
poverty income ratio (INDFMPIR), pseudo-stratum (SDMVSTRA) and the pseudo-PSU
(SDMVPSU). The Dioxins, Furans and Co-planar PCB laboratory file 128dfp_c.xpt
contains SEQN, the lipid-adjusted PCB 118 (LBX118LA), the PCB 118 non-detect
comment code (LBD118LC), and the sub-sample C two year MEC weight (WTSC2YR).
The Non-dioxin-like PCB laboratory file 128npb_c.xpt contains SEQN, the lipid-adjusted
PCB 138, 153, and 180 (LBX138LA, LBX153LA, LBX180LA), the PCB 138, 153, and
180 non-detect comment codes (LBD138LC, LBD153LC, LBD180LC), and the PCB

America's Children and the Environment
March, 2013

Page 2

-------
Biomonitoring: PCBs

sub-sample C two year MEC weight (WTSC2YR). The three files are merged using the
common variable SEQN.

National Health and Nutrition Examination Surveys (NHANES)

Since the 1970s, the National Center for Health Statistics, a division of the Centers for Disease
Control and Prevention, has conducted the National Health and Nutrition Examination Surveys
(NHANES), a series of U.S. national surveys of the health and nutrition status of the
noninstitutionalized civilian population. The National Center for Environmental Health at CDC
measures environmental chemicals in blood and urine samples collected from NHANES
participants. This indicator uses serum PCB measurements of four PCBs from NHANES 2001 -
2002 and 2003-2004 in women ages 16 to 49. The NHANES data were obtained from the
NHANES website: http://www.cdc.gov/nchs/nhanes.htm. Following the CDC recommended
approach, values below the analytical limit of detection (LOD) were replaced by LOD/V2." This
analysis uses the sum of the four PCB congeners 118, 138, 153 and 180 listed in the following
table. If some but not all of the four PCB congeners are missing, then the sum is over the non-
missing PCB congeners. In the rest of this section, we will refer to this sum as the total serum
PCB.

PCB Congener

Full name

SAS name (lipid-
adjusted)

SAS name for non-
detect comment code*

118

2,3',4,4',5-
pentachl orophenyl

LBX118LA

LBD118LC

138

2,2',3,4,4',5-and
2,3,3',4,4',6-
hexachl orophenyl

LBX138LA

LBD138LC

153

2,2',4,4',5,5'-
hexachl orophenyl

LBX153LA

LBD153LC

180

2,2',3,4,4',5,5'-
heptachl orophenyl

LBX180LA

LBD180LC

*The non-detect comment code equals 1 if the measurement is below the analytical limit of detection, and equals 0
if the measurement is at or above the analytical limit of detection.

The NHANES use a complex multi-stage, stratified, clustered sampling design. Certain
demographic groups were deliberately over-sampled, including Mexican-Americans and Blacks,
to increase the reliability and precision of estimates of health status indicators for these
population subgroups. The publicly released data includes survey weights to adjust for the over-
sampling, non-response, and non-coverage. The statistical analyses used the applicable PCB sub-
sample MEC survey weights (WTSP02YR for 2001-2002 and WTSC2YR for 2003-2004) to re-
adjust the serum PCB data to represent the national population.

" See Hornung RW, Reed LD. 1990. Estimation of average concentration in the presence of nondetectable values.

Applied Occupational and Environmental Hygiene 5:46-51.

America's Children and the Environment
March, 2013

Page 3

-------
Biomonitoring: PCBs

Age-Specific Birth Rates

In addition to the NHANES survey weights, the data for women of child-bearing age (ages 16 to
49) were also weighted by the birth rate for women of the given age and race/ethnicity to
estimate prenatal exposures. Thus the overall weight in each two year period is the product of the
NHANES MEC survey weight and the total number of births in the two calendar years for the
given age and race/ethnicity, divided by twice the corresponding population of women at the
midpoint of the two year period:111

Adjusted Survey Weight =

MEC survey weight x U.S. Births (NHANES cycle, age, race/ethnicity) /

{Number of years in NHANES cycle x U.S. Women (NHANES cycle midpoint, age,
race/ethnicity)}.

Race/Ethnicity and Family Income

For this indicator, the percentiles were calculated for demographic strata defined by the
combination of race/ethnicity and family income.

The family income was characterized based on the INDFMPIR variable, which is the ratio of the
family income to the poverty level. The National Center for Health Statistics used the U.S.
Census Bureau Current Population Survey definition of a "family" as "a group of two people or
more (one of whom is the householder) related by birth, marriage, or adoption and residing
together" to group household members into family units, and the corresponding family income
for the respondent was obtained during the interview. The U.S. Census Bureau defines annual
poverty level money thresholds varying by family size and composition. The poverty income
ratio (PIR) is the family income divided by the poverty level for that family. Family income was
stratified into the following groups:

• Below Poverty Level: PIR < 1

• Above Poverty Level: PIR > 1

• Unknown Income: PIR is missing

For the four year period 2001-2004, the weighted percentage of women ages 16 to 49 years with
unknown income was 5%.

Race/ethnicity was characterized using the RIDRETH1 variable. The possible values of this
variable are:

• 1. Mexican American

• 2. Other Hispanic

• 3. Non-Hispanic White

• 4. Non-Hispanic Black

m Axelrad, D.A., Cohen, J. 2011. Calculating summary statistics for population chemical biomonitoring in women
of child-bearing age with adjustment for age-specific natality. Environmental Research 111 (1): 149-155..

America's Children and the Environment
March, 2013

Page 4

-------
Biomonitoring: PCBs

• 5. Other Race - Including Multi-racial

• "."Missing

Category 5 includes: all Non-Hispanic single race responses other than White or Black; and
multi-racial responses.

For this indicator, the RIDRETH1 categories 2, 5, and missing were combined into a single "All
Other Races/Ethnicities" category. This produced the following categories:

• White Non-Hispanic: RIDRETH1 = 3

• Black Non-Hispanic: RIDRETH1 =4

• Mexican-American: RIDRETH1 = 1

• All Other Races/Ethnicities: RIDRETH1 = 2 or 5 or missing

The "All Other Races/Ethnicities" category includes multiracial persons and individuals whose
racial or ethnic identity is not White non-Hispanic, Black non-Hispanic, or Mexican-American.
Persons of "All Other Races/Ethnicities" are selected into the survey with a probability that is
very much lower than White non-Hispanic, Black non-Hispanic and Mexican-American
individuals, and as a group they are not representative of all other race and ethnicities in the
United States.

Calculation of Indicator

Indicator B7 is the median for total serum PCB in women of ages 16 to 49 years, stratified by
race/ethnicity and family income. The median is the estimated concentration such that 50% of all
noninstitutionalized civilian women ages 16 to 49 years during the survey period have total
serum PCB concentrations below this level. Table B7a presents the 95th percentile for total serum
PCB in women of ages 16 to 49 years, stratified by race/ethnicity and family income. The 95th
percentile is the estimated concentration such that 95% of all noninstitutionalized civilian women
ages 16 to 49 years during the survey period have total serum PCB concentrations below this
level. To adjust the NHANES data to represent prenatal exposures, the data for each woman
surveyed was multiplied by the estimated number of births per woman of the given age and
race/ethnicity.

To simply demonstrate the calculations, we will use the NHANES 2001-2004 total serum PCB
values for women ages 16 to 49 years of all race/ethnicities and all incomes as an example. We
have rounded all the numbers to make the calculations easier:

We begin with all the non-missing NHANES 2001-2004 total serum PCB values for women ages
16 to 49 years. Assume for the sake of simplicity that valid data on total serum PCB were
available for every sampled woman. Each sampled woman has an associated annual survey
weight that estimates the annual number of U.S. women represented by that sampled woman.
Since two 2-year periods are combined for these analyses, the associated annual survey weight
for each woman is defined as WTSP02YR/2 for 2001-2002 and WTSC2YR/2 for 2003-2004, so
that the combined 2001-2004 sample represents the annual population. Each sampled woman
also has an associated birth rate giving the numbers of annual births per woman of the given age,

America's Children and the Environment
March, 2013

Page 5

-------
Biomonitoring: PCBs

race, and ethnicity. The product of the annual survey weight and the birth rate estimates the
annual number of U.S. births represented by that sampled woman, which we will refer to as the
adjusted survey weight. For example, the lowest total serum PCB measurement for a woman
between 16 and 49 years of age is 4.6 ng/g lipid with an annual survey weight of 10,000, a birth
rate of 0.03, and thus an adjusted survey weight of 300, and so represents 300 births. The total of
the adjusted survey weights for the sampled women equals 4 million, the total number of annual
U.S. births to women ages 16 to 49 years. The second lowest measurement is also 4.6 ng/g lipid
with an adjusted survey weight of 800, and so represents another 800 U.S. births. The highest
measurement was 715.5 ng/g lipid, with an adjusted survey weight of only 4, because it was for a

48 year old woman, and so represents another 4 U.S. births.

To calculate the median, we can use the adjusted survey weights to expand the data to the entire
U.S. population of births to women ages 16 to 49. We have 300 values of 4.6 ng/g lipid from the
lowest measurement, 800 values of 4.6 ng/g lipid from the second lowest measurement, and so
on, up to 4 values of 715.5 ng/g lipid from the highest measurement. Arranging these 4 million
values in increasing order, the 2 millionth value is 30.1 ng/g lipid. Since half of the values are
below 30.1 and half of the values are above 30.1, the median equals 30.1 ng/g lipid. To calculate
the 95th percentile, note that 95% of 4 million equals 3.8 million. The 3.8 millionth value is 106.2
ng/g lipid. Since 95% of the values are below 106.2, the 95th percentile equals 106.2 ng/g lipid.

In reality, the calculations need to take into account that total serum PCB measurements were not
available for every respondent, and to use exact rather than rounded numbers. There were total
serum PCB measurements for only 1,164 of the 1,327 sampled women ages 16 to 49 years.

These 1,164 sampled women included 1,158 women with measured concentrations for all four
congeners and 6 women with measured concentrations for only one, two, or three of the four
congeners; for those six women the total serum PCB is defined as the sum of the non-missing
congeners. The adjusted survey weights for all 1,327 sampled women add up to 4.2 million, the
U.S. population of births to women ages 16 to 49. The adjusted survey weights for the 1,164
sampled women with total serum PCB data add up to 3.7 million. Thus the available data
represent 3.7 million values and so represent only 90 % of the U.S. population of births. The
median and 95th percentiles are given by the 1.85 millionth (50 % of 3.7 million) and 3.52
millionth (95 % of 3.7 million) U.S. birth's value. These calculations assume that the sampled
women with valid total serum PCB data are representative of women giving birth without valid
total serum PCB data. The calculations also assume that the sampled women are representative
of women that actually gave birth in 2001-2004, since NHANES information on pregnancy and
births was not incorporated into the analysis.

Equations

These percentile calculations can also be given as the following mathematical equations, which
are based on the default percentile calculation formulas from Statistical Analysis System (SAS)
software. Exclude all missing total serum PCB values. Suppose there are n women of ages 16 to

49 years with valid total serum PCB values. Arrange the total serum PCB concentrations in
increasing order (including tied values) so that the lowest concentration is x(l) with an adjusted
survey weight of w(l), the second lowest concentration is x(2) with an adjusted survey weight of
w(2), ..., and the highest concentration is x(n) with an adjusted survey weight of w(n).

America's Children and the Environment
March, 2013

Page 6

-------
Biomonitoring: PCBs

1.	Sum all the adjusted survey weights to get the total weight W:

W = E[1 < i < n] w(i)

2.	Find the largest number i so that the total of the weights for the i lowest values is less than or
equal to W/2.

E[j < i] w(j) < W/2 < E[j < i + 1] w(j)

3.	Calculate the median using the results of the second step. We either have

E[j < i] w(j) = W/2 < E[j < i + 1] w(j)

or

E[j < i] w(j) < W/2 < E[j < i + 1] w(j)

In the first case we define the median as the average of the i'th and i + l'th values:

Median = [x(i) + x(i + l)]/2 if E[j < i] w(j) = W/2

In the second case we define the median as the i + l'th value:

Median = x(i + 1) if E[j < i] w(j) < W/2

(The estimated median does not depend upon how the tied values of x(j) are ordered).

A similar calculation applies to the 95th percentile. The first step to calculate the sum of the
weights, W, is the same. In the second step, find the largest number i so that the total of the
weights for the i lowest values is less than or equal to 0.95W.

E[j < i] w(j) < 0.95W < E[j < i + 1] w(j)

In the third step we calculate the 95th percentile using the results of the second step. We either
have

E[j < i] w(j) = 0.95W < E[j < i + 1] w(j)

or

E[j < i] w(j) < 0.95W < E[j < i + 1] w(j)

In the first case we define the 95th percentile as the average of the i'th and i + l'th values:
95th Percentile = [x(i) + x(i + l)]/2 if E[j < i] w(j) = 0.95W

America's Children and the Environment
March, 2013

Page 7

-------
Biomonitoring: PCBs

In the second case we define the 95th percentile as the i + l'th value:
95th Percentile = x(i + 1) if E[j < i] w(j) < 0.95W

Relative Standard Error

The uncertainties of the median and 95th percentile values were calculated using a revised
version of the CDC method given in CDC 2005,1V Appendix C, and the SAS® program provided
by CDC. The method uses the Clopper-Pearson binomial confidence intervals adapted for
complex surveys by Korn and Graubard (see Korn and Graubard, 1999,v p. 65). The following
text is a revised version of the Appendix C. For the birth rate adjusted calculations for women
ages 16 to 49, the sample weight is adjusted by multiplying by the age-specific birth rate.

Step 1: Use SAS® Proc Univariate to obtain a point estimate PSAs of the percentile value. Use the Weight
option to assign the exact correct sample weight for each chemical result.

Step 2: Use SUDAAN® Proc Descript with Taylor Linearization DESIGN = WR (i.e.,
sampling with replacement) and the proper sampling weight to estimate the proportion (p) of subjects with
results less than and not equal to the percentile estimate PSAs obtained in Step 1 and to obtain the standard
error (sep) associated with this proportion estimate. Compute the degrees-of-freedom adjusted effective
sample size

^df — ( tnum/ tdenom) P ( 1 — P ) / ( SCp )

where tnum and tdenom are 0.975 critical values of the Student's t distribution with degrees of freedom
equal to the sample size minus 1 and the number of PSUs minus the number of strata, respectively. Note:
the degrees of freedom for tdenom can vary with the demographic sub-group of interest.

Step 3: After obtaining an estimate of p (i.e., the proportion obtained in Step 2), compute the Clopper-
Pearson 95% confidence interval (PL(x,ndf), Pu(x,ndf)) as follows:

PL(x,ndf) = ViFv1>v2 (0.025)/(v2 + ViFv1>v2(0.025))

Pu(x,ndf) = v3Fv3)V4 (0.975)/(v4 + v3Fv3>v4(0.975))

where x is equal to p times ndf, vi = 2x, v2 = 2(ndf - x + 1), v3 = 2(x + 1), v4 = 2(ndf - x), and Fdijd2(P) is
the (3 quantile of an F distribution with dl and d2 degrees of freedom. (Note: If ndf is greater than the
actual sample size or if p is equal to zero, then the actual sample size should be used.) This step will
produce a lower and an upper limit for the estimated proportion obtained in Step 2.

Step 4: Use SAS Proc Univariate (again using the Weight option to assign weights) to determine the
chemical percentile values PCdc, Lcdc and UCdc that correspond to the proportion p obtained in Step 2 and
its lower and upper limits obtained in Step 3. Do not round the values of p and the lower and upper limits.
For example, if p = 0.4832, then PCdc is the 48.32'th percentile value of the chemical. The alternative
percentile estimates PCdc and Psas are not necessarily equal.

1V CDC Third National Report on Human Exposure to Environmental Chemicals. 2005
vKornE. L., Graubard B. I. 1999. Analysis of Health Surveys. Wiley.

America's Children and the Environment
March, 2013

Page 8

-------
Biomonitoring: PCBs

Step 5: Use the confidence interval from Step 4 to estimate the standard error of the estimated percentile

PcDC*

Standard Error (PCDc) = (UCDc ~ Lcdc) / (2t

denom)

Step 6: Use the estimated percentile PCdc and the standard error from Step 4 to estimate the relative
standard error of the estimated percentile PCdc:

Relative Standard Error (%) = [Standard Error (PCdc) / Pcdc] x 100 %

The tabulated estimated percentile is the value of PSAs given in Step 1. The relative standard error is given
in Step 6, using PCdc and its standard error.

The relative standard error depends upon the survey design. For this purpose, the public release
version of NHANES includes the variables SDMVSTRA and SDMVPSU, which are the Masked
Variance Unit pseudo-stratum and pseudo-primary sampling unit (pseudo-PSU). For
approximate variance estimation, the survey design can be approximated as being a stratified
random sample with replacement of the pseudo-PSUs from each pseudo-stratum; the true stratum
and PSU variables are not provided in the public release version to protect confidentiality. If the
relative standard error is too high, then the estimated percentile will not be accurately estimated.
Furthermore, if the degrees of freedom (from Step 2) is too low, then the relative standard error
will be less accurately estimated and thus may be underestimated. For these reasons, percentiles
with high relative standard errors or with low degrees of freedom are unstable or unreliable.

Percentiles with a relative standard error less than 30% and with 12 or more degrees of freedom
were treated as being reliable and were tabulated. Percentiles with a relative standard error that is
30% or greater but less than 40% and with 12 or more degrees of freedom were treated as being
unstable; these values were tabulated but were flagged to be interpreted with caution. Percentiles
with a relative standard error less than 40% and with between 7 and 11 degrees of freedom were
also treated as being unstable; these values were tabulated but were flagged to be interpreted
with caution. Percentiles with a relative standard error that is 40% or greater, or without an
estimated relative standard error, or with 6 or less degrees of freedom, were treated as being
unreliable; these values were not tabulated and were flagged as having a large uncertainty.

Questions and Comments

Questions regarding these methods, and suggestions to improve the description of the methods,
are welcome. Please use the "Contact Us" link at the bottom of any page in the America's
Children and the Environment website.

America's Children and the Environment
March, 2013

Page 9

-------
Biomonitoring: PCBs

Statistical Comparisons

Statistical analyses of the percentiles were used to determine whether the differences between
percentiles for different demographic groups were statistically significant. For these analyses, the
percentiles and their standard errors were calculated for each combination of age group, income
group (below poverty, at or above poverty, unknown income), and race/ethnicity group using the
method described in the "Relative Standard Error" section. In the notation of that section, the
percentile and standard error are the values of Pcdc and Standard Error (Pcdc), respectively.
These calculated standard errors account for the survey weighting and design and, for women,
for the age-specific birth rate.

Using a weighted linear regression model, the percentile was assumed to be the sum of
explanatory terms for age, income and/or race/ethnicity and a random error term; the error terms
were assumed to be approximately independent and normally distributed with a mean of zero and
a variance equal to the square of the standard error. In this model, the weight is the inverse of the
variance, so that percentiles with larger standard errors are given less of a statistical weight in the
fitted regression model. Using this model, the difference in the value of a percentile between
different demographic groups is statistically significant if the difference between the
corresponding sums of explanatory terms is statistically significantly different from zero. A p-
value at or below 0.05 implies that the difference is statistically significant at the 5% significance
level. No adjustment is made for multiple comparisons.

For each type of comparison, we present unadjusted and adjusted analyses. The unadjusted
analyses directly compare a percentile between different demographic groups. The adjusted
analyses add other demographic explanatory variables to the statistical model and use the
statistical model to account for the possible confounding effects of these other demographic
variables. For example, the unadjusted race/ethnicity comparisons use and compare the
percentiles between different race/ethnicity pairs. The adjusted race/ethnicity comparisons use
the percentiles for each age/ income/race/ethnicity combination. The adjusted analyses add age,
and income terms to the statistical model and compare the percentiles between different
race/ethnicity pairs after accounting for the effects of the other demographic variables. For
example, if White non-Hispanics tend to have higher family incomes than Black non-Hispanics,
and if the serum PCB level strongly depends on family income only, then the unadjusted
differences between these two race/ethnicity groups would be significant but the adjusted
difference (taking into account income) would not be significant.

Comparisons between pairs of race/ethnicity groups and between income groups are shown in
Tables 1 and 2, respectively, for women ages 16 to 49 years. In Table 1, for the unadjusted "All
incomes" comparisons, the only explanatory variables are terms for each race/ethnicity group.
For these unadjusted comparisons, the statistical tests compare the percentiles for each pair of
race/ethnicity groups. For the adjusted "All incomes (adjusted for age, income)" comparisons,
the explanatory variables are terms for each race/ethnicity group together with terms for each
age, and income group. For these adjusted comparisons, the statistical test compares the pair of
race/ethnicity groups after accounting for any differences in the age and income distributions
between the race/ethnicity groups.

America's Children and the Environment
March, 2013

Page 10

-------
Biomonitoring: PCBs

In Table 1, for the unadjusted "Below Poverty Level" and "At or Above Poverty Level"
comparisons, the only explanatory variables are terms for each of the twelve
race/ethnicity/income combinations (combinations of four race/ethnicity groups and three
income groups). For example, in row 1, the p-value for "Below Poverty Level" compares White
non-Hispanics below the poverty level with Black non-Hispanics below the poverty level. The
same set of explanatory variables are used in Table 2 for the unadjusted comparisons between
one race/ethni city group below the poverty level and the same race/ethni city group at or above
the poverty level. The corresponding adjusted analyses include extra explanatory variables for
age, so that race/ethni city/income groups are compared after accounting for any differences due
to age. Although these comparisons only involve the two income groups with known incomes,
these statistical models were fitted to all three income groups (including those with unknown
income) to make a more general, better fitting model; this approach has no impact on the
unadjusted p-values but has a small impact on the adjusted p-values. Also in Table 2, the
unadjusted p-value for the population "All" compares the percentiles for women ages 16 to 49
years below poverty level with those at or above poverty level, using the explanatory variables
for the two income groups (below poverty, at or above poverty), excluding those with unknown
income. The adjusted p-value includes adjustment terms for age and race/ethni city in the model.

For women, the age groups used were 16-19, 20-24, 25-29, 30-39, and 40-49.

For more details on these statistical analyses, see the memorandum by Cohen (2010).V1

Table 1. Statistical significance tests comparing the percentiles of PCBs in women ages 16 to 49
years, between pairs of race/ethni city groups, for 2001-2004.



P-VALUES

Variable

Percentile

First
race/ethnicity
group

Second
race/ethnicity
group*

All

incomes

All incomes
(adjusted
for age,
income)

Below
Poverty
Level

Below
Poverty
Level
(adjusted
for age)

At or
Above
Poverty
Level

At or
Above
Poverty
Level
(adjusted
for age)

PCB

50

White non-
Hispanic

Black non-
Hispanic

0.622

0.211

0.893

0.033

0.759

0.363

PCB

50

White non-
Hispanic

Mexican-
American

<0.001

<0.001

0.002

0.009

0.001

<0.001

PCB

50

White non-
Hispanic

Other

0.691

<0.001

0.588

0.477

0.702

<0.001

PCB

50

Black non-
Hispanic

Mexican-
American

0.001

<0.001

<0.001

<0.001

0.010

<0.001

PCB

50

Black non-
Hispanic

Other

0.957

<0.001

0.572

0.307

0.991

<0.001

PCB

50

Mexican-
American

Other

0.002

0.134

0.326

0.020

0.001

0.299

PCB

95

White non-
Hispanic

Black non-
Hispanic

0.601

<0.001

0.402

<0.001

0.947

0.121

PCB

95

White non-
Hispanic

Mexican-
American

<0.001

0.002

0.041

<0.001

0.007

<0.001

PCB

95

White non-
Hispanic

Other

0.003

0.034

<0.001

<0.001

0.096

<0.001

PCB

95

Black non-
Hispanic

Mexican-
American

<0.001

<0.001

0.085

<0.001

0.020

<0.001

V1 Cohen, J. 2010. Selected statistical methods for testing for trends and comparing years or demographic groups in
ACE NHIS and NHANES indicators. Memorandum submitted to Dan Axelrad, EPA, 21 March, 2010.

America's Children and the Environment
March, 2013

Page 11

-------
Biomonitoring: PCBs



P-VALUES

Variable

Percentile

First
race/ethnicity
group

Second
race/ethnicity
group*

All

incomes

All incomes
(adjusted
for age,
income)

Below
Poverty
Level

Below
Poverty
Level
(adjusted
for age)

At or
Above
Poverty
Level

At or
Above
Poverty
Level
(adjusted
for age)

PCB

95

Black non-
Hispanic

Other

0.001

<0.001

<0.001

0.550

0.125

<0.001

PCB

95

Mexican-
American

Other

<0.001

<0.001

<0.001

<0.001

0.002

0.009

* "Other" represents the "All Other Races/Ethnicities" category, which includes all other races and ethnicities not specified, together with those

individuals who report more than one race.

Table 2. Statistical significance tests comparing the percentiles of PCBs in women ages 16 to 49
years, between those below poverty level and those at or above poverty level, for 2001-2004.







P-Values for difference between income levels

Variable

Percentile

Population*

Unadjusted

Adjusted (for age)**

PCB

50

All

0.047

0.274

PCB

50

White non-Hispanic

0.307

0.004

PCB

50

Black non-Hispanic

0.214

0.684

PCB

50

Mexican-American

0.272

0.488

PCB

50

Other

0.798

0.103

PCB

95

All

0.193

<0.001

PCB

95

White non-Hispanic

0.216

<0.001

PCB

95

Black non-Hispanic

0.044

0.035

PCB

95

Mexican-American

0.432

0.002

PCB

95

Other

0.089

<0.001

* "Other" represents the "All Other Races/Ethnicities" category, which includes all other races and ethnicities not specified, together with those
individuals who report more than one race.

** Comparison for "All" is adjusted for age and race/ethnicity; comparisons for race/ethnicity categories are adjusted for age.

America's Children and the Environment
March, 2013

Page 12

-------