The Small Area Health Insurance Estimates (SAHIE) program at the U.S. Census Bureau produces estimates of numbers and proportions of those with and without health insurance coverage for demographic groups within states and counties. The demographic groups are defined by age, sex, and income, and in addition, for states by race and ethnicity. Income groups are defined in terms of income-to-poverty ratio (IPR), which is the family income divided by the appropriate federal poverty level.
The estimates are for groups that some of them overlap or are contained in one or another domain. However, actual modeling is done at a “base” level at which domains are disjoint, and are chosen so that the estimates needed for publication can be obtained as needed by aggregation. For example, for states, was done the actual modeling for the full cross-classification of:
– Four age categories: 0-64, 18-64, 40-64, 50-64; a fifth age category, 21-64, was added in release year 2014.
– Five race/ethnicity categories: all races, Hispanic, White not Hispanic, Black not Hispanic and Hispanic (any race)
– Three sex categories: all sexes, male, female
– Five income groups: all incomes, and IPR categories 0-138%, 0-200%, 0-250%, and 0-400%; a sixth IPR category, 138-400%, was added in release year 2012.
The 2013 SAHIE were updated in order to provide a basis of comparison for the 2014 SAHIE which, for the first time, used more up-to-date Medicaid and Children’s Health Insurance Program (CHIP) source data. Both data, original and updated is contained in this dataset.
The SAHIE model is an “area level” model (Rao 2003) in that it uses survey estimates for areas or domains of interest rather than individual responses, and it uses other data that are aggregates rather than for individuals. Each part of the SAHIE model is similar to a well-known small area-level model, the Fay-Herriot model. The Fay-Herriot model is a hierarchical model in which the variables of interest occupy a “middle” level, between high-level parameters such as regression coefficients, and observed data.
The following primary data sources were used for states and counties:
– Census 2010
– Single-year ACS (American Community Survey) direct estimates
– 5-year ACS direct estimates
– Federal Tax Returns data
– Supplementary Nutrition Assistance Program data
– Medicaid/Children’s Health Insurance Program participation data
– Demographic population estimates
– County Business Patterns