Evaluating a Structural Model of Labor Supply and Welfare Participation: Evidence from State Welfare Reform Experiments

This paper assesses the ability of a structural labor supply model to predict the impacts of a welfare policy change by studying two state welfare reform experiments conducted in Minnesota (MN) and Vermont (VT) during the mid-1990s. I estimate and evaluate a static discrete choice model of labor supply and welfare participation that incorporates heterogeneity in preferences, fixed costs of work, and disutility associated with welfare take-up. Although this type of labor supply models has been commonly estimated and applied to welfare and tax policy simulations, there have been very few attempts to verify the predictive ability of the existing models. I use the experimental impacts of the welfare policy change in each state as a benchmark for the structural model’s predictions. This approach is similar in spirit to LaLonde (1986). The utility parameters of the model are estimated using data from the MN control group. First, based on the parameter estimates, I make predictions regarding labor supply, welfare participation, and government costs under the treatment group program in MN and compare them with the observed effects of the MN experiment. Next, I apply the parameter estimates to the VT control group and compare the predicted and observed impacts of the policy change in VT. The results show that the model fits the estimation sample very well, but is unable to replicate the observed treatment effects on labor supply and welfare participation outcomes of the two experiments. Consequently, the effect on net government costs is under-predicted by 31 to 93 percent in MN. The prediction biases are even larger for the VT sample.