Association for Public Policy Analysis & Management

Research Objective: We first compare estimates of health insurance status (covered/not covered) and coverage type (e.g., non-group coverage, Medicaid) as reported using questions from the American Community Survey (ACS) to known coverage status and type as indicated in health plan enrollment records. Second, we explore the offsetting influence of misreporting coverage across individuals with different insurance types in the CPS. Finally, we project the impact of coverage type misreporting for estimates of the prevalence of insurance coverage derived from state-specific and national benchmark data.

Study Design:Using administrative records from a private health plan, individuals known to be enrolled in a range of different coverage types (private and public, including the marketplace) were selected and randomly assigned to a test survey mimicking the ACS.

Population Studied:Non-elderly adults enrolled in a Minnesota health plan were surveyed by telephone in May and June of 2015. Sample were randomly selected from five different coverage type pools or strata: employer-sponsored insurance, non-group coverage, qualified health plans (QHP) from the Marketplace, Medicaid and MinnesotaCare (Minnesota’s basic health plan). The analysis dataset includes data on non-elderly 1,834 people from 1,110 households assigned to the ACS survey treatment.

Preliminary Findings: Among individuals indicated by enrollment records to be covered by one of the five coverage types, overall 97 percent reported some type of comprehensive coverage. There was some variation by strata; 96 percent of public enrollees reported being covered compared to 98 percent for all other plan types. Reporting accuracy of particular coverage type varied by strata. Using the standard set of ACS questions 93 percent of individuals with ESI reported this type. Preliminary results show that reporting accuracy for public coverage was about 72 percent and for non-group (including from the marketplace) reporting accuracy was about 84 percent. We then explore how well we are able to correctly defining coverage type using three additional items not currently included in the ACS: whether the coverage was from the marketplace, whether there was a premium and, if so, whether the premium was subsidized.

We also examine the offsetting influence of misreporting coverage across individuals with different insurance types in the ACS. We then project the impact of these offsetting influences of survey misreporting on benchmark survey estimates of the prevalence of insurance type. To facilitate generalizability, we contrast CHIME ACS result with and without the MinnesotaCare strata given that only small group of states have public programs that charge a premium and bear a unique name.

Conclusion: CHIME results corroborate past findings that self-reported coverage status (insured/uninsured) in the ACS is quite accurate. There is no evidence of a drop-off in self-reported accuracy of coverage status (insured/uninsured). Accuracy of self-reported coverage type is generally problematic and appears to be more complicated following the introduction of new health insurance products and new pathways to enrolling in coverage. Understanding overall bias in survey estimates of coverage type requires information about misreporting of both private and public insurance.