EXPERIMENTAL DESIGN, THE CONLISK-WATTS ASSIGNMENT MODEL, AND THE PROPER ESTIMATION OF BEHAVIORAL RESPONSE.

In the Seattle and Denver Income Maintenance Experiments (SIME/DIME), a variant of the Conlisk-Watts Assignment Model was used to allocate families to experimental treatments. This paper discusses the theory of the Conlisk-Watts Assignment Model, its application to SIME/DIME, and the implications of the model for estimating experimental effects. Unlike simple random assignment, the Conlisk-Watts model results in an experimental sample in which household characteristics are not orthogonal to experimental treatments and experimental treatments are not orthogonal to each other. Because of this nonorthogonality, the assignment variables must be included as explanatory variables in regression models that estimate experimental-control differences in order to avoid biased estimates of these differences. In addition, if the assignment variables affect response, they must be interacted with treatment variables in order to generalize the experimental findings to the national population and to obtain unbiased estimates of differential response among programs. [ABSTRACT FROM AUTHOR]