Widely cited estimates from the U.S. Census Bureau likely overstate the number of uninsured people and understate the number of people with Medicaid coverage because of an inability of people to recall their insurance status accurately from the previous year, according to a study published in Health Affairs.
The Census Bureau's Current Population Survey, or CPS, administered in February, March, and April each year, asks respondents whether they had health insurance coverage (including Medicaid) at any point in the previous calendar year.
However, the CPS' long recall period (14-16 months) can lead to inaccurate responses, reports Jacob Klerman, a principal associate at Abt Associates in Cambridge, Mass., and his co-authors in "Understanding The Current Population Survey's Insurance Estimates And The Medicaid 'Undercount.'"
According to Klerman and his colleagues, CPS' responses are most inaccurate for those whose periods of Medicaid coverage in the previous year were relatively short and relatively early in the year, because those are hardest to recall.
They say similar recall problems may affect CPS' estimates of those covered by other types of health insurance in addition to Medicaid, and they call for research into whether other surveys, such as Census' American Community Survey, could provide better estimates of the uninsured.
Klerman and his co-authors reject two competing interpretations of the CPS estimates that have dominated the literature. The first standard interpretation holds that people answer the CPS questions as they are instructed, so that the survey results indeed yield the number of people who did not have Medicaid or other health insurance for even one day in the previous calendar year.
However, the researchers say the CPS estimates are much higher than other surveys' all-year uninsurance estimates and are closer to point-in-time estimates of uninsurance produced by other surveys. For this reason, a second, point-in-time interpretation of the CPS data has developed, which holds that answers to the CPS reflect the insurance status of respondents on the date the questions are asked.
Klerman and his colleagues put forward evidence that both of these interpretations of the CPS data are inadequate. For example, among those who had Medicaid coverage at the time of the survey but who lacked Medicaid for all of the previous year, only a quarter reported having Medicaid coverage, a result inconsistent with the "point in time" interpretation.
On the other hand, among those who lacked Medicaid at the time of the survey but who were covered by Medicaid at some point during the previous year, only a third reported Medicaid coverage – a result inconsistent with the "previous-calendar year" interpretation.