Income, poverty, and population estimates from the decennial censuses enter into four phases of the estimation process:

• Investigating the relationships between administrative data and income and poverty. Because the census yields direct county estimates with relatively small sampling variance, it is helpful in exploring the relationship between the dependent variables and their predictors, although this view is limited to a single year (1989 for the 1990 Census, 1999 for Census 2000).
• Providing a potentially valuable predictor variable. We use the decennial census estimates and administrative data to make postcensal estimates.
• Evaluating estimates from alternative models with decennial census estimates. By constructing models that predict income and poverty for the census year, we can assess the models' performance.
• Computing critical components of variance that are needed in several phases of the modeling process for subsequent estimate years (only for SAIPE 2004 and prior year estimates).