While the goal of SAIPE is providing estimates of numbers of people in poverty in various groups, for many purposes poverty rates are more readily interpreted. We provide these rates but regard them as inferior to the estimates of numbers of people in poverty because of the unknown quality of our estimates of the required denominators. We provide what we regard as "illustrative" confidence intervals around the poverty rates, which are computed as if the poverty universe estimates were "true", i.e., without error.

The state models estimate ratios of number of people in poverty to population, as measured in Current Population Survey (CPS) Annual Social and Economic Supplement (ASEC), for the groups of interest. We convert them to estimates of numbers of people in poverty by multiplying by demographic estimates of the population, as covered by the CPS ASEC, for these groups. The county models directly estimate logarithms of numbers of people in poverty in the groups of interest. The computation of poverty rates based on the model-based state and county estimates of numbers of people in poverty requires estimates of the number of people in the relevant poverty universes. Because the poverty numbers are consistent with the CPS ASEC definitions, the poverty universes must also be. The poverty universes for CPS ASEC estimates of the numbers of people in poverty have specific definitions that exclude various population subgroups from the baseline noninstitutional population (the sampling frame for the survey). We base our approximations of these groups on county-level postcensal estimates (as of July, after the postcensal year for which the poverty estimates are created) of resident population and of the group quarters population by group quarters type and by age group provided by the Census Bureau Population Division. For our 1999 estimates, we based our approximation on the Census 2000 population. Procedures for state and county estimates are different from each other in postcensal years, and from census years.

We derive state level estimates of the poverty universes for postcensal years in five steps:

1. First, we compute 3-year average ratios of the CPS ASEC poverty universes to the CPS ASEC populations of each state directly from the CPS ASEC. The CPS ASEC population includes the civilian noninstitutional population of the United States and members of the armed forces in the United States living off post or with their families on post, but excludes all other members of the armed forces. In addition, the CPS ASEC treats college students as residing in their parental homes.
2. Second, we estimate the CPS ASEC population for each state from postcensal population estimates, by subtracting the appropriate group quarters populations.
3. Third, the resulting state estimates of the CPS ASEC population are raked to the national CPS ASEC population, by age group, through a simple ratio adjustment.
4. Fourth, we apply the ratios of poverty universes to the estimated CPS ASEC population calculated in step one.
5. Finally, we rake the resulting state estimates of the poverty universe populations, by age group, to the national CPS ASEC poverty universes through a simple ratio adjustment.

We derive county level estimates of the poverty universes for postcensal years in four steps:

1. First, we modify the county resident population estimates to exclude residents of institutions, military group quarters, and college dormitories from the various age groups of interest. We use estimates by age, supplied by Population Division to make these modifications. Our estimates also exclude residents of college dormitories from the poverty universe in the county where the college is located.
2. Second, we tabulate the ratio of the poverty universe to the non-institutional population for each age group in each county from the most recent decennial census.
3. Third, we apply the ratios, created in step two, to the modified July county population estimates from step one, to obtain initial poverty universe estimates.
4. Finally, we rake this estimated county poverty universes to total the state poverty universes previously estimated by a simple ratio adjustment. This final adjustment implicitly distributes college dormitory residents across the counties proportionately by the size of the age group being estimated.

Before income year 2000, when controlling the estimates of the poverty universe to CPS ASEC results, we implicitly generate population estimates at the county level that reflect adjustment for the 1990 census undercount. A second effect of the process is to shift the July population estimates to the CPS ASEC estimates that reflect the population as of March. Population estimates based on Census 2000 do not reflect any adjustment for undercount.

We form state and county-level 1989 poverty universe estimates for the computation of illustrative model-based poverty rates by multiplying 1990 census county poverty universes for the groups of interest by the ratio of the CPS ASEC national total to the 1990 census national total for each group.