Comparing Alternative Models for Using Decennial Census Data in Saipe State Poverty Estimates

1. Introduction The Census Bureau's Small Area Income and Poverty Estimates (SAIPE) Program produces poverty estimates for various age groups for states, counties, and school districts. The state and county estimates are produced from various models applied to direct poverty estimates obtained from the March Supplement to the Current Population Survey (CPS). These models use predictor variables constructed from administrative data sources, demographic population estimates, and poverty estimates from the previous decennial census. The administrative data sources used include IRS tax file data, food stamp program participation data, and supplemental security income data from the Social Security Administration. The school district estimates are produced from simple synthetic updates to the previous census estimates with results controlled to the current county model-based estimates. This paper focuses on use of decennial census poverty data in the state poverty models. SAIPE state production estimates were previously released for income years 1993 and 1995-98. The " income year, " IY, refers to the year for which income is reported in the March CPS of the following year, the latter referred to as the " survey year. " (IY 1994 was skipped due to technical difficulties with applying the models arising from the transition to the " new " CPS sample derived from a sample redesign based on 1990 census results.) The models that produced these previous SAIPE estimates used 1990 census results as the previous census data. The model-based estimates with documentation are available from the SAIPE web site at As of this writing, new SAIPE state estimates are in production for IY 1999. In developing these estimates we faced some interesting new issues regarding our use of census data because the Census 2000 long form poverty estimates are also for IY 1999. If the census estimates could be regarded as unbiased estimates of true poverty, then at the state level the census estimates would nearly provide truth for IY 1999, because sampling error in the census estimates is very small at the state level. Census poverty estimates are not truth, however, because the census is known to be subject to various nonsampling errors, which could here be thought of as biases. CPS poverty estimates have their own nonsampling errors, but these are believed to be less important than the