“Can Survey Design Reduce the Undercount of Public Health Insurance Coverage?”

Brian Robertson PhD, Mark Noyes, MPH

Poster Session, AAPOR Annual Conference, May 2018


The undercount in public coverage observed in survey results has been an issue for decades. Due to factors including confusion about coverage type and a strong social desirability effect, Medicaid enrollment is significantly underestimated in most surveys. The recent CHIME study comparing ACS and CPS survey responses to the known insurance status of a study group reported that only 83% of public enrollees in the study group under age 65 accurately reported their coverage in response to these two surveys (Call, Pascale, Fertig, and Oellerich, 2017).

Surveys such as the American Community Survey and the Behavioral Risk Factor Surveillance System assess insurance coverage using a simple series of yes/no questions about each type of coverage. This is known as the traditional design.

Survey design can help to reduce this undercount and improve data quality. Our experience is that successful survey design can minimize one primary source for undercounts; confusion about coverage type. This includes undercounts due to the state specific program structure, method of enrollment, and program administration. For example, respondents may confuse Medicaid and Medicare. Others may assume they have private insurance because they enrolled through a health exchange, despite being determined eligible for public coverage. Appropriate survey design incorporates information and questions to help resolve confusion.

In this research we compare results of surveys from four states (Oregon, Rhode Island, Vermont, and South Dakota) to determine which survey elements are most effective in reducing the public coverage undercount. We evaluate sequentially the reporting of public coverage as respondents proceed through questions identifying health insurance coverage.
In this research, we show that one cannot rely on the traditional method of asking about health insurance coverage; a more considered design is required. Further, the survey must be tailored to individual states. Finally, we make specific recommendations for future survey design.