QUARTERLY JOURNAL OF ECONOMICS
Part of the difculty is the uncertainty about individual knowledge of Medicaid rules and their valuation of the benets. We have tried a large number of alternative specications, none of which indicates a large effect of Medicaid. A full accounting of these results can be found in Meyer and Rosenbaum [2000b].
D. Welfare Waivers and Time Limits
The AFDC waiver variables have the expected effect on em- ployment, and their coefcients are signicantly different from zero. Both the implementation of a time limit on welfare receipt and the actual termination of benets under a work requirement or time limit waiver are predicted to increase employment by between 1.4 and 4.8 percentage points. However, until the last years of our sample, the overall importance of such waivers is small. Even by 1994, only 5 percent of single mothers lived in states with a time limit, and less than half of 1 percent lived in states that had begun to terminate benets.
One should be cautious in interpreting the waiver coef- cients, especially in attributing effects to the implementation of particular provisions of recent waivers or the termination of cases per se. The perception of welfare changes by potential welfare recipients, the attitudes of case workers, and differences in state implementation of policies likely play a large role in inuencing the welfare caseload and consequently employment. It is also econometrically difcult to disentangle which provisions of a waiver are the most important, since states typically imple- mented several changes to their AFDC programs under waivers at the same time. The reported coefcients are partly the effect of the particular actions coded and partly a proxy for other changes going on in the states.
Recognizing these limitations, the strength of the evidence here for a causal interpretation of the waiver results is much greater than in the studies of welfare caseloads. First, we use implementation dates, rather than application or approval dates, which are at best loosely related to when provisions are enforced. Second, when we account for state intentions to reform welfare as indicated by whether or not a state has made a major waiver application, this variable has little effect. Third, one or two year leads of our time limit and termination variables have small and insignicant coefcients, suggesting that the provisions per se, rather than publicity or administrator attitudes lead to the em- ployment increases. This result contrasts with those of Blank