Health Insurance in the Current Population Survey

Attachment M

Health Insurance in the
Current Population Survey
Redesign and Production
January 31, 2013

Paper presented at FCSM on November 6, 2014
Carla Medalia, Ph.D.
Brett O’Hara, Ph.D.
Joanne Pascale
Jonathan Rodean
Amy Steinweg
U.S. Census Bureau

Disclaimers:
These data are being released on request, despite concerns about their quality. The Census
Bureau’s policy is not to withhold data that are available, unless releasing such data would
violate confidentiality requirements. The Census Bureau recommends using these data only for
research or evaluation purposes, and not to make statements about characteristics of the
population or economy because they do not meet the criteria outlined in the Census Bureau’s
Statistical Quality standard: Releasing Information Product.
The views expressed in this research, including those related to statistical, methodological, technical, or
operational issues, are solely those of the author and do not necessarily reflect the official positions or
policies of the Census Bureau. The author accepts responsibility for all errors. This presentation reports
the research and analysis undertaken by Census Bureau staff. It has undergone more limited review than
official publications.
1

The Annual Social and Economic Supplement to the Current Population Survey (CPS ASEC)
generates widely used estimates on health insurance coverage and the uninsured. 1, 2, 3, 4 However,
research suggests that the calendar year estimate of the uninsured is higher than it should be and
that estimates actually reflect a mixture of current and past year coverage. 5, 6,3 To address this
concern, the Census Bureau substantially redesigned the CPS ASEC health insurance module
over the past ten years. 7 The redesigned instrument was fielded in a large national content test in
March 2013. In addition to the features of the redesigned instrument, the content test also takes
better advantage of automated computer-assisted interviewing and adds important new content to
the instrument. The 2013 content test fielded a complete CPS ASEC interview to previous CPS
respondents who were interviewed by Census Bureau telephone center staff.
This report highlights results of the content test with those from the 2013 CPS ASEC. 8
Specifically, the evaluation includes only production CPS data collected in telephone centers in
March 2013, rather than the full ASEC CPS dataset. In addition, this report compares calendaryear estimates with point-in-time estimates from the content test. The 2013 content test is not
compared to other surveys here because the purpose is to evaluate the change in the questions
from the CPS ASEC. Both the datasets are unedited. Comparing to other surveys would
introduce too many additional sources of variation then is necessary for the evaluation of the
content test itself. From this point, the restricted CPS ASEC will be referred to as the production
instrument.
HIGHLIGHTS
•
•
•

The percentage of people without health insurance was 10.6 percent in the content test
and 13.2 percent in the production instrument.
The content test was not statistically different from the production instrument’s Medicaid
rate.
The percentage of people with private coverage was statistically higher in the content test
than the production instrument.

1

Glied, Sherry, Dahlia K. Remler, and Joshua Graff Zivin. 2002. “Inside the Sausage Factory: Improving
Estimates of the Effects of Health Insurance Expansion Proposals.” Milbank Quarterly: 80(4): 603-35.
2
Holahan, John. 2011. “The 2007-09 Recession and Health Insurance Coverage.” Health Affairs: 30(1): 145-52.
3
Klerman, Jacob A., Michael Davern, Kathleen Thiede Call, Victoria Lynch, and Jeanne D. Ringel. 2009.
“Understanding the Current Population Survey’s Insurance Estimates and the Medicaid ‘Undercount.’” Health
Affairs – Web Exclusive: w991-w1001.
4
Ziegenfuss, Jeanette Y., and Michael E. Davern. 2011. “Twenty Years of Coverage: An Enhanced Current
Population Survey – 1989–2008.” Health Services Research: 46(1): 199-209.
5
DeNavas-Walt, Carmen, Bernadette D. Proctor, and Jessica C. Smith. 2012. Pg. 21 in Income, Poverty, and
Health Insurance Coverage in the United States: 2011. U.S. Bureau of the Census, Current Population Reports,
P60-243. Washington, D.C.: U.S. Government Printing Office.
6
Kenney, Genevieve, and Victoria Lynch. 2010. “Monitoring Children’s Health Insurance Coverage Under
CHIPRA Using Federal Surveys.” Pgs. 65-82 in Databases for Estimating Health Insurance Coverage for
Children: A Workshop Summary, edited by Thomas J. Plewes. Washington, D.C: National Academies Press.
7
Examples of this research will be discussed in the background.
8
"Data are subject to error arising from a variety of sources. For information on confidentiality protection, sampling
error, non-sampling error, and definitions see: http://www.census.gov/prod/techdoc/cps/cpsmar13.pdf and
http://www.reginfo.gov/public/do/PRAViewDocument?ref_nbr=201211-0607-002."

2

•
•

The percentage of people uninsured in the previous calendar year in the content test was
significantly lower than the percentage of people uninsured at the time of the interview.
The average time to complete the health insurance questions in the content test was 1:32
minutes longer than in the production instrument. However, when excluding questions
that added new material on health insurance exchanges and employer-sponsored
insurance offers and take-up, the content test was 1:20 minutes longer than the production
instrument.

BACKGROUND
For a more complete discussion of the motivation for the redesigned health insurance instrument
and the 2013 content test, please see OMB Supporting Statement A 9 and the paper on the 2013
content test presented at the Joint Statistical Meetings in 2013. 10
Inflated estimate of the uninsured
Although the CPS ASEC is a widely used indicator of the uninsured in the United States, many
researchers have questioned its validity. 11 In particular, researchers have suggested that
estimates of the uninsured in the previous calendar year is too high, and may actually reflect a
mixture of current and past year coverage. 12, 13, 14 Research has also shown that people have
difficulty recalling health insurance coverage in the distant past. For example, short spells of
Medicaid that occurred early on in the previous calendar year may not be salient enough for
people to recall during the interviewed up to a year or so later.14 This measurement error is
partially due to both the reference period and timing of data collection. 15
Redesigned health insurance instrument
The redesigned health insurance instrument differs from the traditional CPS ASEC in three
primary ways: questions about type of coverage, questions about past coverage, and the
9

U.S. Census Bureau. 2013. OMB Supporting Statement A. Available at:
http://www.reginfo.gov/public/do/PRAViewDocument?ref_nbr=201211-0607-002.
10
Medalia, Carla. 2013. “Health Insurance in the Current Population Survey: Now and Later?” Unpublished paper
presented at the Joint Statistical Meetings in Montreal on August 4, 2013. Available from author upon request.
11
The issues with the traditional CPS ASEC health insurance estimates have been well established, as discussed in
the Census Bureau’s annual publication on health insurance. The Census Bureau devotes two-thirds of a page in the
Income, Poverty, and Health Insurance Coverage in the United States: 2011 (DeNavas-Walt et. al.: 21) to flaws in
the estimate. The quality of health insurance data has long been a concern of Health and Human Services.
12
DeNavas-Walt, Carmen, Bernadette D. Proctor, and Jessica C. Smith. 2012. Pg. 21 in Income, Poverty, and
Health Insurance Coverage in the United States: 2011. U.S. Bureau of the Census, Current Population Reports,
P60-243. Washington, D.C.: U.S. Government Printing Office.
13
Kenney, Genevieve, and Victoria Lynch. 2010. “Monitoring Children’s Health Insurance Coverage Under
CHIPRA Using Federal Surveys.” Pgs. 65-82 in Databases for Estimating Health Insurance Coverage for
Children: A Workshop Summary, edited by Thomas J. Plewes. Washington, D.C: National Academies Press.
14
Klerman, Jacob A., Michael Davern, Kathleen Thiede Call, Victoria Lynch, and Jeanne D. Ringel. 2009.
“Understanding the Current Population Survey’s Insurance Estimates and the Medicaid ‘Undercount.’” Health
Affairs – Web Exclusive: w991-w1001.
15
Klerman, Jacob A., Michael Davern, Kathleen Thiede Call, Victoria Lynch, and Jeanne D. Ringel. 2009.
“Understanding the Current Population Survey’s Insurance Estimates and the Medicaid ‘Undercount.’” Health
Affairs – Web Exclusive: w991-w1001.

3

household-level design. The complete redesigned instrument was previously tested in 2010, in a
survey that replicated certain elements of the CPS ASEC interview. 16, 17 While this 2010 survey
provided the proof of concept for the redesigned instrument,18 it did not include the full battery
of CPS Basic and supplemental income questions that would be asked during an actual CPS
ASEC interview. Therefore, in order to evaluate the redesigned health insurance instrument
operationally in the CPS environment, the Census Bureau fielded a nation-wide content test in
2013, which follows the method of redesigned health insurance instrument closely. 19
One of the main differences between the traditional and redesigned health insurance instruments
is the reference period for data collection. The traditional instrument asks about coverage at any
time in the past calendar year. The redesigned instrument also captures this information, but
does so in a different way: it starts by asking about current coverage and then uses follow-up
questions to find out (1) when that coverage began and (2) which months the individual had the
coverage. If the individual does not have current coverage, the instrument asks about coverage
during the previous calendar year through the present. The change in the reference period should
make it easier for respondents to report their coverage. It also means that the redesigned
instrument can capture information about all months between January of the previous year and
the interview month. In this regard, the redesign captures similar information as the redesigned
Survey of Income and Program Participation, another Census Bureau survey, which measures
monthly coverage of health insurance during the previous calendar year.
Other differences between the traditional and redesigned instruments include the identification of
plan type and which household members are covered by each plan. First, the traditional
instrument asks about each plan type. By contrast, the redesigned instrument asks a series of
questions, beginning with the general source of coverage and then asking more specific questions
to identify the plan type. Second, the traditional instrument asks if anyone in the household was
covered by a particular plan type, and if yes, who was covered. On the other hand, the
redesigned instrument asks about an individual’s own coverage, and then asks if the same plan
type covered any other household members. Additional questions ask each household member
about any additional plans that they may have.
New content in the CPS ASEC
In addition to redesigning the CPS ASEC health insurance section, the Census Bureau added
new questions to address two areas. First, new questions allow for the measurement of health
insurance exchanges, and second, new questions allow the Census Bureau to determine rates of
employer-sponsored insurance offers and take-up.

16

Pascale, Joanne. 2012. “Findings from a Split-Ballot Experiment on a New Approach to
Measuring Health Insurance in the Current Population Survey.” Unpublished Census Bureau report. Presented at
American Society of Health Economists Conference, Minneapolis, MN, June 2012.
17
Pascale, Joanne, Jonathan Rodean, Jennifer Leeman, Carol Cosenza, Alisu schoua-Glusberg. Forthcoming.
“Preparing to Measure Health Coverage in Federal Surveys Post-Reform: Lessons from Massachusetts.”
18
Boudreaux, Michael, Brett Fried, Joanna Turner, and Kathleen Thiede Call. 2013. “SHADAC Analysis of the
Survey of Health Insurance and Program Participation.” State Health Access and Data Assistance Center.
Retrieved September 26, 2013. (http://www.shadac.org/files/shadac/publications/SHIPP_final_report.pdf).
19
U.S. Census Bureau. 2013. OMB Supporting Statement A. Available at:
http://www.reginfo.gov/public/do/PRAViewDocument?ref_nbr=201211-0607-002.

4

Exchanges. One of the key features of the Patient Protection and Affordable Care Act (ACA) is
the implementation of state-specific “Health Insurance Exchanges” in 2014. An exchange is a
state-level marketplace of private health insurance options for individuals and small businesses.
While the redesigned health insurance instrument measures the types of coverage included in the
traditional instrument, it also is able to measure health insurance exchange coverage and
premium subsidization through new questions. This new content was cognitively tested in 2012
in Massachusetts. 20
Employer-sponsored insurance offers and take-up. The ACA may lead to changes in the rates of
employer-sponsored health insurance offers and take-up, or whether or not someone who is
offered insurance through their employer enrolls in that plan. 21, 22 As a result, the Council for
Economic Advisors asked the Census Bureau to add questions to the ASEC to measure these
changes. The redesigned instrument includes a short series of questions asked of people who are
currently employed but who do not have health insurance through their own employer. These
questions were originally asked in the CPS Contingent Worker Supplement, fielded in February
of 1995, 1997, 1999, 2001, and 2005. This particular set of questions only works with a current
coverage question.
DATA AND METHODS
Data
The sample for the 2013 content test was selected from retired sample for the CPS. None of
these households had been given the CPS ASEC before, but they had received the CPS Basic and
other CPS supplements, such as the Displaced Workers supplement, Child Support, or Food
Security. The retired sample was between one and two years old, meaning that their last CPS
interview was one to two years prior. The sample was selected to exclude households that had
been selected for another survey, the American Time Use Survey, as well as other
characteristics. 23 To be interviewed, households needed to be reachable by the same phone
number that they had when they were last interviewed one to two years before. Furthermore,
households needed to live at the same addresses as their last interview. The final sample size for
the content test was 22,508 households with 1,168 households ineligible for the CPS ASEC
interview, which reduced the sample to 21,340 households. The data presented have been
weighted by the Census Bureau to reflect the total civilian non-institutionalized population.
Trained CPS ASEC interviewers at three Census Bureau telephone centers, in Hagerstown
Maryland, Jeffersonville Indiana, and Tucson Arizona, conducted 9,195 household interviews
20

Pascale, Joanne, Jonathan Rodean, Jennifer Leeman, Carol Cosenza, Alisu schoua-Glusberg. Forthcoming.
“Preparing to Measure Health Coverage in Federal Surveys Post-Reform: Lessons from Massachusetts.”
21
Maxwell, Nan L. 2013. “The ACA, Health Care Costs, and Disparities in Employer-Sponsored Health
Insurance.” Mathematica Policy Research Reports, Mathematica Policy Research. Retrieved August 1, 2013
(http://EconPapers.repec.org/RePEc:mpr:mprres:7683).
22
Thurm, Scott. 2013. “Will Companies Stop Offering Health Insurance Because of the Affordable Care Act?”
The Wall Street Journal, June 16. Retrieved August 1, 2013
(http://online.wsj.com/article/SB10001424127887323582904578488781195872870.html).
23
For a complete description of the sampling method used for the 2013 content test, please see the following paper:
Brault, Matthew. "Non-Response Bias in the 2013 CPS ASEC Content Test." Paper presented at the Federal
Committee on Statistical Methodology 2013 Annual Meeting. Available from author upon request.

5

(including complete and partial interviews) for the CPS Basic part of the interview. The
analytical sample for the content test, which included only those individuals that were still in the
survey by the health status question, was 16,401 individuals.
The data from the content test have been partially recoded to replicate the data produced by the
traditional ASEC. Census Bureau staff edited the demographic characteristics, such as those
collected during the CPS Basic part of the interview, using the same programs used in the
traditional CPS. Furthermore, the Census Bureau recoded data from the content test so that they
contain the same information as the traditional ASEC variables.
The data that serve as the comparison group is the 2013 CPS ASEC, which was conducted in
February, March and April which was conducted with a combination of in-person and phone
center interviews. The original sample size for the full ASEC sample was 98,095 households.
After removing ineligible addresses, the sample was reduced to 83,225 households, which
includes both interviews and Type A noninterviews. These households contained 202,634
individuals. The analysis focuses on interviews conducted in March 2013 in ASEC call centers,
which included 6,410 households. These households contain the final analytical sample, which
was 13,228 individuals.
Nonresponse and Section Dropout
Due to survey dropout, not all of the households interviewed for the CPS Basic were included in
the sample for the health insurance section. The overall response rate for the CPS Basic
production instrument was 90.7 percent in 2013, which is substantially higher than for the
redesigned instrument, which was 43.1 percent. As shown in Table 1, this corresponds to an
initial nonresponse rate of 9.3 percent in the traditional instrument and a 56.9 percent
nonresponse rate in the content test.
In addition to the CPS Basic nonresponse rate, respondents sometimes drop out of the survey
once it has begun. The dropout rate among those households that were considered complete or
partial CPS Basic interviews but dropped out before the end of the health insurance section was
15.9 percent in the content test, compared to 14.5 percent in the traditional instrument.
Combining the section dropout with the CPS Basic nonresponse, 63.8 percent of eligible
households were dropped from the analytical sample in the content test, compared with 22.5
percent in the traditional instrument. This means that the analytical sample for the content test
included 36.2 percent of the eligible households, compared with 77.5 percent for the traditional
instrument.

6

Methods
Differences between the traditional CPS ASEC and the content test include the different survey
designs, as described above, as well as differences in mode, both of which could have an effect
on results. All content test interviews were conducted in call centers in the month of March,
compared with a combination of call center and in-person interviews in February, March and
April for the traditional CPS ASEC. An example of the mode effect can be seen by looking at
just the traditional ASEC. Table 2 shows that the distribution of demographic characteristics,
such as age, differs between the total sample and the call center, March only sample. In order to
focus the analysis on the differences between survey design, this analysis controls for survey
mode by comparing the content test to the traditional ASEC for March call centers only.
The content test, which was only in call centers, was weighted to reflect the national population
in March of 2013. The CPS ASEC is also weighted to the nation in March of 2013. However,
the call centers for the CPS ASEC are a subset of the interviews and represent a subset of the
national population estimate. For instance, there were proportionately too many full-time, fullyear workers in the ASEC call centers than in the national population. 24 This might bias the
estimates because most of this group gets their insurance through an employer. To address this
problem, CPS ASEC call center estimates were reweighted to the national population that is
represented by the weighted content test estimates. The final weight adjustment for the CPS
ASEC call centers was as follows:
𝑎𝑑𝑗𝑤𝑔𝑡𝑖 = 𝑟𝑘 𝑎𝑔𝑒,𝑠𝑒𝑥,𝑟_𝑒,𝑓𝑡𝑓𝑦 ∗ 𝑤𝑔𝑡𝑖

Where 𝑎𝑑𝑗𝑤𝑔𝑡 is the adjusted weight that equals the Content Test national estimate

Rk is the raking factor. The raking factor is controlled to age (0-17, 18-34/35-64, 65+),
sex, race and ethnicity (nonHispanic Whites, nonHispanic W=Black, nonHispanic
Other, and Hispanics), and full-time full-year workers aged 18-64
Wgt is the ASEC weight.
Again, this adjusted weight makes the data from the call centers that were used to collect CPS
ASEC data similar to the content test data in terms of demographic and work characteristic; i.e.,
we wanted to control for population effects within the mode of call centers.
All statistical tests were performed at the 90 percent confidence level.
The types of health insurance coverage are private insurance plans and government coverage.
Private insurance includes direct purchase and employer-sponsored insurance. Government
insurance includes Medicare, Medicaid, and military coverage. In this paper, we do not show
military coverage here due to the small sample size. Indian Health Services does not count as
health insurance coverage because it is clinic-based health care.

24

In the production CPS ASEC final sample, there were 86.1 percent under age 65. In the March call centers, 75.2
percent of population was under age 65. After adjusting the call center estimate, 85.3 percent of the sample was
under age 65 in both the content test and the March call centers.

7

FINDINGS
We analyze the health insurance content test data in two ways. The analysis compares estimates
for the 2012 calendar year from the redesigned instrument to the production CPS ASEC March
call centers. Our expectation is that the redesigned instrument will have an uninsured rate that is
lower than the traditional instrument. Second, estimates for the 2012 calendar year from the
content test are compared to the current coverage estimates from the content test. We expect that
the uninsured rate for the calendar year estimate will be lower than the current coverage estimate.
Comparing the content test to production ASEC
Table 3 shows results for the comparison of the content test to the production ASEC. Unless
otherwise noted, the estimates that are reported are statistcally different at the 90 percent
confidence level. The discussion of the results will focus on the comparison of the production
ASEC March call center unedited and adjusted weighted results (hereafter referred to as the
ASEC call centers) to the unedited weighted redesign results (hereafter referred to as the content
test). 25
The percentage of people without health insurance was 10.6 percent in the content test and 13.2
percent in the ASEC call centers. In other words, the content test had a 2.6 percentage point
lower uninsured rate than the ASEC call centers. This pattern, where the percentage of those
uninsured was lower in the content test than in the production call centers, was consistent by age
group. Those aged 18 to 64, under age 19, and aged 35 to 44 had lower levels of uninsured in
the content test than in the production call centers. The difference for those in all other age
groups were in the same direction but not statistically different between the two surveys. For
this last group, aged 65 and above, we would not have expected a difference between the content
test and production ASEC because the population is almost entirely covered by Medicare.
The percentage of people who were uninsured in the content test were generally lower than in the
ASEC call centers by race and Hispanic origin. The differences for White alone, non-Hispanic
White, Black alone, and Asian alone were all lower in the content test than in the ASEC call
centers, while the difference for Hispanic was in the same direction but not statistically
significant.
In addition to the percentage of people without coverage for the previous calendar year, rates by
type of coverage are also shown. For all ages, the percentage of people with private coverage
was higher in the content test than in the production ASEC, while the percentage of people with
government coverage was lower in the content test than in the production ASEC. This pattern is
also present by age group, and was consistenet for those under age 18 and those aged 18 to 64.
Comparing the content test calendar year to current coverage estimate
As previously described, the CPS ASEC is often criticized for producing calendar year estimates
that may actually reflect a combination of past year and current coverage. To address this, the
redesign explicitly asks about both current coverage as well as calendar year coverage.
25

The table also includes additional production ASEC estimates for the original weighted ASEC (final estimate) and
the ASEC March call center unedited but not adjusted weighted estimate.

8

Therefore, in addition to comparing the 2012 calendar year content test estimate to the
production ASEC, it can also be compared to the current coverage estimate from the content test.
Note that the calendar year estimates are subtracted from current coverage estimates.
In the sample, the percentage of people who were uninsured in 2012 was 10.6 percent, compared
to 12.0 percent at the time of the interview. Therefore, more people report being uninsured
currently than are uninsured for the entire previous calendar year, which is consistent with the
expectation that people are generally more likely to be currently uninsured than they were for the
entire previous calendar year.
In addition to examining the difference between calendar year and current coverage for the
uninsured rate, we examined the difference by coverage type. While the differences by type
were in the expected direction, none of them were statistically significant. Unlike the uninsured
status, a negative relationship is expected for private coverage, since someone is more likely to
have had private insurance at any time in the previous calendar year than they are to currently
have private insurance. This was found to be the true in the content test: the percentage of
people with private health insurance coverage was 72.3 percent in 2012 and 70.7 percent at the
time of the interview, a non-significant difference. In 2012, the percentage of people covered by
government insurance was 28.2 percent, which was not statistically different from the 27.9
percent for the current coverage estimate. The percentage of people with Medicare was 15.7
percent in the calendar year and 16.1 percent at the time of the interview, a non-significant
difference. Finally, the percentage of people on Medicaid in 2012 was 11.2 percent, which was
not statistically different from the 10.5 percent at the time of the interview. Given the churning
on and off Medicaid,26 it would make sense if the spread between these two estimates were
greater than 0.6 percentage points.
Comparing the content test to production instrument: mean length of interview
The mean time for a total health insurance interview increased for the content test from the
production instrument. 27 The total mean household time for a health insurance interview was
2:39 for the production instrument, compared to 4:12 for the content test, for an average increase
of 1:32. This figure includes the complete health insurance instrument, including the self-rated
health questions for both instruments as well as the new content added to the content test health
insurance instrument (employer-sponsored insurance take-up and offers and exchange-related
health insurance questions). Excluding the new content in the content test, the average
household time for an interview in the production instrument remains steady at 2:39 but is 3:59
for the content test, for an average increase of 1:20. In other words, the new content added 0:13
on average to the length of the interview.
DISCUSSION

26

Czajka, John L. 2012. “Medicaid Enrollment Gaps, 2005 to 2007: Final Report.” Washington, D.C.:
Mathematica Policy Research.
27
For a thourough analysis of the time it took to complete the content test and CPS ASEC interviews, please see:
Bee, Adam and Aaron Cantu. 2013. "Evaluating Respondent Burden of the CPS ASEC Content Test with Timer
Data.” Paper to be presented at the 2013 FCSM meeting in Washington, D.C.

9

One of the main hypotheses for this comparison is that the redesigned calendar year estimate of
the uninsured will be lower than it is for the production instrument. This hypothesis is based on
research which shows that the production CPS uninsured estimate is too high, and that it reflects
a mixture of current and past year coverage. 28, 29, 30 Because the redesigned instrument explicitly
addresses this concern by asking about both current and past coverage, the instrument allows for
estimates of current coverage and calendar year coverage that have discrete reference periods.
Making comparable measures to the production CPS proves difficult when the sample’s
respondents are not reflective of the larger population. The content test’s low response rate
raised concerns about what role nonresponse bias would have on key estimates. A separate
analysis of nonresponse showed that non-response exerted downward bias on estimates of
uninsurance and upward bias on Medicare coverage. 31 However, the weights used in this
analysis may account for and correct this kind of bias. More research is being conducted to
examine these possible sources of nonresponse bias and the effect that it could have on the
results presented in this paper.

28

DeNavas-Walt, Carmen, Bernadette D. Proctor, and Jessica C. Smith. 2012. Pg. 21 in Income, Poverty, and
Health Insurance Coverage in the United States: 2011. U.S. Bureau of the Census, Current Population Reports,
P60-243. Washington, D.C.: U.S. Government Printing Office.
29
Kenney, Genevieve, John Holahan, and Len Nichols. 2006. “Toward a More Reliable Federal Survey for
Tracking Health Insurance Coverage and Access.” Health Services Research: 41(3): 918-45.
30
Klerman, Jacob A., Michael Davern, Kathleen Thiede Call, Victoria Lynch, and Jeanne D. Ringel. 2009.
“Understanding the Current Population Survey’s Insurance Estimates and the Medicaid ‘Undercount.’” Health
Affairs – Web Exclusive: w991-w1001.
31
Brault, Matthew. "Non-Response Bias in the 2013 CPS ASEC Content Test." Paper presented at the Federal
Committee on Statistical Methodology 2013 Annual Meeting. Available from author upon request.

10

TABLES
Table 1. Nonresponse and dropout in the 2013 CPS ASEC and 2013 Content Test

CPS Basic nonresponse
Section dropout (through end
health insurance)
Total nonresponse and dropout
Analytical sample (as percent of
eligible households)

Content
Test
56.9%

Production CPS ASEC
March Call
Total
Centers
Sample
0.0%
9.3%

15.9%
63.8%

11.4%
11.4%

14.5%
22.5%

36.2%

88.6%

77.5%

Source: CPS ASEC health insurance production instrument and content test (2013)
Notes: “CPS Basic nonresponse” refers to nonresponse in the CPS Basic interview. “Section dropout” refers to the
all dropout from before the supplement began through the end of the health insurance section. “Total nonresponse
and dropout” refers to all nonresponse and dropout from the beginning of the CPS Basic interview through the end
of the health insurance section.

11

Table 2. Demographic characteristics of the 2013 CPS ASEC and 2013 Content Test
Production instrume nt
Final e stimate s
N (in
SE
1000s)
% (%)

Conte nt Te st

Call ce nters, March
N (in
SE
1000s)
%
(%)

Call ce nte rs, March,
adjuste d
N (in
SE
1000s)
%
(%)

Call ce nte rs, March
N (in
SE
1000s)
%
(%)

Age
Under 65

268,008

86.1

0.0

16,400

78.6

0.6

207,615

85.3

0.4

207,615

85.3

0.4

0 to 17

74,425

23.9

0.0

3,459

16.6

0.4

57,760

23.7

0.5

57,760

23.7

0.4

18 to 34

71,777

23.1

0.0

4,253

20.4

0.5

51,480

21.1

0.5

51,480

21.1

1.2

35 to 64

121,806

39.2

0.0

8,687

41.6

0.5

98,376

40.4

0.6

98,376

40.4

0.7

65 and over

43,108

13.9

0.0

4,472

21.4

0.6

35,860

14.7

0.4

35,860

14.7

0.4

White non-Hispanic

195,330

62.8

0.0

16,856

80.8

0.6

159,950

65.7

1.0

159,950

65.7

0.6

Black non-Hispanic

37,619

12.1

0.0

1,307

6.3

0.4

26,566

10.9

0.7

26,566

10.9

0.5

Other non-Hispanic

24,937

8.0

0.0

994

4.8

0.4

19,608

8.1

0.6

19,608

8.1

0.3

Hispanic

53,230

17.1

0.0

1,714

8.2

0.5

37,351

15.3

0.8

37,351

15.3

0.5

Full time, full year

98,762

51.0

0.2

98,762

51.0

0.2

75,509

50.4

0.8

75,509

50.4

0.8

Less than full time, full year

47,070

24.3

0.2

47,070

24.3

0.2

39,022

26.0

0.6

39,022

26.0

0.9

Out of labor force

47,753

24.7

0.2

47,753

24.7

0.2

35,325

23.6

0.7

35,325

23.6

0.6

Male

152,335

49.0

0.0

10,258

49.2

0.4

118,284

48.6

0.4

118,284

48.6

0.6

Female

158,781

51.0

0.0

10,613

50.9

0.4

125,191

51.4

0.4

125,191

51.4

0.6

Race and e thnicity

Work Status for pe rsons
age d 18-64

Se x

N (in thousands)

311,116

20,871

243,475

243,475

n (sample size)

172,662

13,228

13,228

16,401

Source: CPS ASEC health insurance production instrument and content test (2013)

12

Table 3. Health insurance coverage in the 2013 CPS ASEC and 2013 Content Test
Production CPS ASEC
Productio
n (final
estimate)

Universe

Estimate

Content T est

Production Call
Centers, March
(unedited)

Call Centers,
March (unedited)

Original
weight

Original
weight

Adjusted
weight

Calend
ar
Year

Difference
Production
and Content
Test

Content Test
only

Current
coverag
e

Prod.
(adj) –
Content
Test

S
i
g

Calendar
year Current
coverage

S
i
g

T otal by Coverage T ype
T otal

Uninsured

15.4

11.1

13.2

10.6

12.0

-2.6

*

-1.4

*

T otal

Private

63.9

72.3

69.0

72.0

70.4

3.0

*

1.6

*

T otal

Gov.

32.6

33.6

30.6

28.6

28.3

-2.1

*

0.3

*

T otal

... Medicare

15.7

22.4

16.2

15.7

16.2

-0.5

-0.4

*

T otal

... Medicaid

16.4

9.1

12.4

11.4

10.8

-0.9

0.6

*

8.9

6.9

7.6

6.0

7.0

-1.6

-1.0

*

Under 18 by Coverage T ype
Aged 0 to 17

Uninsured

Aged 0 to 17

Private

60.1

65.8

62.0

65.9

64.1

3.9

*

1.8

*

Aged 0 to 17

Gov.

39.2

33.5

36.6

32.2

30.6

-4.4

*

1.5

*

Aged 0 to 17

... Medicare

1.0

0.9

1.1

0.8

0.8

-0.3

-0.1

Aged 0 to 17

... Medicaid

35.9

29.2

32.6

29.1

27.5

-3.5

1.6

*

Aged 18 to 64

Uninsured

21.0

15.6

18.1

14.6

16.6

-3.6

*

-2.0

*

Aged 18 to 64

Private

67.2

75.8

72.6

75.3

73.7

2.8

*

1.7

*

Aged 18 to 64

Gov.

16.6

13.4

13.8

12.2

12.0

-1.6

*

0.2

Aged 18 to 64

... Medicare

4.2

4.5

4.2

3.8

4.1

-0.4

-0.3

*

Aged 18 to 64

... Medicaid

10.6

5.3

6.2

6.3

5.9

0.1

0.4

*

White Alone

Uninsured

14.7

10.5

12.6

9.9

10.9

-2.6

*

-0.9

*

... Non-Hispanic White

Uninsured

11.1

8.7

8.8

6.5

7.3

-2.3

*

-0.9

*
*

18 to 64 by Coverage T ype

Race and Hispanic Origin

Black Alone

Uninsured

19.0

16.7

16.5

12.4

17.4

-4.1

*

-5.0

Asian Alone

Uninsured

15.1

13.6

14.4

10.3

11.3

-4.0

*

-1.0

Hispanic

Uninsured

29.1

29.9

30.0

26.1

27.5

-3.9

-1.4

Age
0 to 18

Uninsured

9.2

7.2

8.4

6.1

7.3

-2.3

19 to 25

Uninsured

25.5

22.1

24.8

18.3

22.8

-6.5

-4.5

26 to 34

Uninsured

27.1

21.4

25.0

20.1

21.0

-4.9

-0.9

35 to 44

Uninsured

21.2

19.0

22.2

14.1

16.0

-8.0

45 to 64

Uninsured

16.1

10.9

12.2

11.4

13.1

65 and above

Uninsured

1.5

1.4

1.7

1.5

1.0

N (in thousands)

*

*

-1.2

*

-1.9

*

-0.8

-1.7

*

-0.2

0.5

*

--

--

Source: CPS ASEC health insurance production instrument and content test (2013)
Notes: * indicates that the difference is statistically significant at the 90% confidence level.

13