Poster Paper: Title: The Case of the Missing Medicaid Enrollees: Identifying the Magnitude and Causes of the Medicaid Undercount in the SIPP

Friday, November 7, 2014
Ballroom B (Convention Center)

*Names in bold indicate Presenter

Jamie Rubenstein Taber and Brett O'Hara, U.S. Census Bureau
A number of papers have detailed a substantial underreporting of Medicaid coverage in the Current Population Survey (CPS) as well as other surveys.  Undercounting the number of Medicaid enrollees leads to biased estimates of the costs and benefits of Medicaid enrollment in a wide range of contexts.  Undercounting Medicaid enrollees may result from characteristics of the survey instrument, stigma of enrollment in means-tested government insurance, confusion on the part of the survey respondent, or other factors.  Each of these factors could also lead to an undercounting of Medicaid enrollees in the Survey of Income and Program Participation (SIPP), an important source of data for policy, sociological, and economic analysis.  However, to date, no systematic research exists on whether the SIPP also undercounts the number of Medicaid enrollees and by how much.  This paper bridges that gap by quantifying the Medicaid undercount in the SIPP and analyzing what demographic and socioeconomic characteristics and program characteristics increase the probability of a respondent failing to report Medicaid or other means-tested public programs. 

To investigate the extent and determinants of Medicaid undercount in the SIPP, we perform descriptive and multivariate analyses.  For our analysis, we must compare the SIPP survey data to administrative data, and accordingly, we link 2009 data from the 2008 SIPP panel to administrative Medicaid enrollment records from the Medicaid Statistical Information System (MSIS).  To link these data, we use unique person-level identifiers developed by the U.S. Census Bureau, which are available on internal Census Bureau data.  We perform all analyses separately for children, non-elderly adults, and elderly adults. 

For the descriptive analysis, we calculate the number of Medicaid enrollees enrolled anytime in 2009 using three sources: the SIPP without imputed health insurance values, the SIPP with imputed values, and the MSIS civilian non-institutionalized population.  Next, we use the linked SIPP and MSIS data to calculate the percentage of individuals for which the two sources agree, the percentage who incorrectly report no Medicaid coverage, and the percentage who incorrectly report Medicaid coverage.  We further examine reported coverage for people who fail to report Medicaid despite being enrolled in the program.  This information gives insight into which other insurance categories may suffer from an overcount.  We also tabulate the percentages of individuals falsely reporting having Medicaid and incorrectly reporting not having Medicaid by a number of demographic and socioeconomic characteristics as well as characteristics of the state Medicaid programs. 

For the multivariate analysis, we estimate two logistic models.  The first model predicts the probability that a person has Medicaid given that the person did not report Medicaid.  The second model predicts the probability that a person does not have Medicaid given that the person reports Medicaid.  In these models, Medicaid indicators are a function of socioeconomic and demographic characteristics, length of time enrolled in the program, and characteristics of the Medicaid program.  These results will provide both guidance for where to focus efforts to improve the SIPP and a better understanding of SIPP data for other researchers.