By Jennifer Cheeseman Day
In an ever-changing world, we must constantly work to enhance our ways of measurement. This year, the U.S. Census Bureau implemented a new set of questions in the Current Population Survey Annual Social and Economic Supplement about health insurance coverage, and next week, we will begin releasing the results. We expect these new questions will better reflect our changing health insurance environment.
Like most data collections, we have a long history of research and survey improvement. Research dating back to the 1990s showed that estimates of the uninsured seemed higher than those of other major surveys, indicating that underreporting of health insurance coverage might be a larger problem for the Current Population Survey (see infographic).
In response, beginning with the 1999 estimates, we added a verification question at the end of the supplement (see infographic). This asked respondents who had not reported any health insurance coverage whether they were, in fact, uninsured during the previous year. This resulted in an 8 percent decline in the uninsured rate between the old and new estimate for 1999, thus moving the Current Population Survey closer to other published estimates.
For the 2000 estimates, we added a new question about the State Children’s Health Insurance Program, and in order to provide new, reliable state level estimates of health insurance coverage, we added 28,000 more households to the sample.
In 2011, we enhanced our data processing methods and released revised estimates on health insurance coverage back to 1999.
Yet, research continued to suggest the Current Population Survey needed further improvement, as the health insurance estimates still were not in line with other sources.
We researched reporting problems in the survey itself, considering, for example, whether question order matters. We held focus groups and conducted cognitive testing and field tests. We also conducted expert reviews, interviewer debriefings, record-check studies, and other kinds of research, working to find any potential sources of measurement error.
From this research, we developed a new set of survey questions. The aim was to capture coverage in a more intuitive way and make it easier for the respondents to correctly identify their coverage.
The questions were shorter and simpler. They drilled down as needed and were designed to clarify areas of ambiguity. The survey instrument used previous answers about age, income and other coverage within the household to present appropriate follow-up questions that could capture underreported plans.
And finally, the new design used a hybrid flow of both household- and person-level questions. It begins by asking the first person in the household about his or her coverage, but then fills the information where appropriate for other household members so the questions don’t have to repeat for each person.
In addition, we ran a final field test in 2013 to demonstrate that the questions worked well. With positive results in hand, the Census Bureau moved forward with the redesigned questions earlier this year.
Our goal was to produce the best health insurance coverage estimates with the Current Population Survey. The improvements this year will better measure health insurance coverage for the prior calendar year, thus providing a strong 2013 baseline to measure future changes in health insurance coverage caused by the Affordable Care Act.