Missing the Mark? Imputation Bias in the Current Population Survey's State Income and Health Insurance Coverage Estimates
Michael Davern, Lynn A. Blewett, Boris Bershadsky, Noreen Arnold
The Demographic Supplement to the U.S. Current Population Survey (CPS) is used to produce state estimates of health insurance coverage and income. These estimates are used in federal allocation formulas that distribute $10-11 billion annually to states for the State Children's Health Insurance Program (SCHIP) and the Elementary and Secondary Education Act. The purpose of this article is to examine the CPS for evidence of bias in state estimates due to missing data imputation and estimate the extent of the bias for each of the fifty-one states and Washington DC. Comparing three years of CPS (1998-2000), to the Census 2000 Supplementary Survey and 1990 Decennial Census data benchmarks, we find evidence of bias in state estimates of earned income. We also extend the technique to the CPS state health insurance coverage estimates and find even more evidence of bias. In general, the "better off" states (those with higher insurance coverage rates or more income) tend to be even "better off" (have higher estimates of average income and coverage rates) after correcting for bias (and vice versa). We conclude by considering alternative strategies for the U.S. Census Bureau to alter its current imputation procedures.
Hot deck, allocation formulas, item nonresponse, missing data, 1990 Decennial Census, Census 2000 Supplementary Survey, American Community Survey