The rate of participation in programs that provide an economic `safety net` has been widely used as a bellwether of program adequacy. Improved design for income maintenance programs requires consistently-estimated models of behavior leading to participation in the program. Estimates of participation probabilities from households surveys are biased by response error; hence, estimates of behavioral models of participation are also biased. This fact motivates the research. Administrative-level records have been matched to survey data from the Survey of Income and Participation Program (SIPP). Extent of response error and models of response error conditioned on economic and demographic variables can be estimated from the match. Models of program participation can then be correctly estimated, accounting for response errors. The initial focus of this research is on two programs: Food Stamps (FS) and Aid to Families with Dependent Children (AFDC). Improved models of participation in these programs will afford better understanding of populations not currently reached by the program. Implications of program policy changes then can be more accurately predicted with the result that more efficient program designs can be developed. The methodologies developed from this research can be generalized to other subjects and afford a way for many applications to adjust for error in household surveys.
