A partially corrected estimate of medicaid enrollment and uninsurance: Results from an imputational model developed off linked survey and administrative data

Abstract

To improve the utility of estimates of Medicaid enrollment and uninsurance from the Current Population Survey (CPS) we use linked data from the CPS and the Medicaid Statistical Information System (MSIS) to build a probabilistic imputation model that partially corrects the public use data files for systematic under-reporting of Medicaid. We estimate the probability that a CPS survey case was enrolled in Medicaid, conditional on whether or not in the CPS the individual responded that they had Medicaid. We use the imputed data to develop adjusted estimates of Medicaid enrollment and uninsurance by demographic characteristics. The net Medicaid enrollment total using our imputation model for CY 2006 and 2007 is 41.0, compared to 34.0 million using the standard CPS variables. The resulting net adjusted uninsurance estimate is 4.5% below the unadjusted estimate.

Comparing Errors in Medicaid Reporting across Surveys: Evidence to Date

“…Some of the matching studies are published in refereed journals (Davern et al. , b; Klerman et al. ), and some are reported in working papers (Klerman et al.…”

Creating Improved Survey Data Products Using Linked Administrative-Survey Data

“…In this paper we do this by combining survey reported data with program administrative data and use the linked data to create models we can use to partially correct for measurement error in survey reported program participation. Previous studies have used linked data to: (1) examine sample coverage and accuracy (Bee, Gathright and Meyer 2015;Celhay, Meyer and Mittag, 2017), (2) impute variables (Davern et al 2009a), (3) substituteadministrative values for reported values (Nicholas and Wiseman,2010;Hokayem et al, 2015;Meyer and Mittag, 2015), (4) supplement survey reported data (Abowd et al 2006), and 5correct estimates (Davern et al 2009a;Schenker et al 2010;Mittag 2017). These studies demonstrate that data linkage can improve survey accuracy by illuminating problems in survey design and methodology.…”

“…Our analyses build on applications to Medicaid in Davern et al (2009a) and SNAP in Mittag (2017). Davern et al (2009a) link Medicaid administrative data to CPS data and then use the model developed on the linked data to impute an indicator of Medicaid enrollment in the CPS for subsequent years. Mittag (2017) employs a similar approach using SNAP administrative data linked to ACS data to correct estimates of food stamp receipt out of sample in the ACS.…”

“…However, because 47% of those linked do correctly report, survey reported Medicaid enrollment is a critical predictor of true enrollment in the imputation model. Stratifying on survey reported Medicaid status, Davern et al (2009a) estimated two models to partially correct for survey measurement error. The first model used a logistic regression to predict whether a person received Medicaid in MSIS given that they did not report having Medicaid in the survey (i.e., a false-negative model).…”

“…On the other hand only one percent of respondents in the CPS reported having Medicaid coverage that could not be confirmed through the linkage. This pattern results in a large net undercount (Davern et al 2009a). 3 Research on survey misreporting of SNAP has also found that a substantial share of recipients do not report receipt.…”