THE LABOR SUPPLY OF SINGLE MOTHERS
To assess the effect of cutting the AFDC benet, one needs to incorporate the effects of all three of the welfare variables and the Medicaid if Work variable. When the AFDC maximum benet and payment standard are cut, they not only reduce benets if one does not work, but also reduce benets if one does work. They also decrease the likelihood that a working mother will be on welfare at all, thereby reducing both her Medicaid eligibility and her AFDC transaction and stigma costs. When we do the full calculations, we nd that a 10 percent cut in the maximum benet ($324 annually) increases both the annual and weekly employment rate by about 1.0 percentage points.
Despite a more detailed calculation of welfare incentives than most past work and the use of panel data techniques, we think there are important potential sources of bias in these esti- mates. We should also note that by dividing the effect of welfare into income when working and when not, and by estimating a separate term for transaction costs/stigma we are putting the theoretical predictions to a more severe test than most work. As discussed in Section IV, the Welfare Benets if Work variable and the Probability of AFDC Receipt if Work variable are more dif- cult to calculate precisely than our other variables. The larger coefcient on the Welfare Benets if Work variable could also be due to the scale of this variable being inappropriately low. The earnings distribution used to calculate expected benets puts most of the weight on earnings levels where welfare benets would be low or zero. It is very likely that we should use an earnings distribution that puts greater weight in the left tail, since women who work while on welfare rarely report all of their earnings to the welfare ofce [Edin and Lein 1997]. The reasons for possible bias in the Probability of AFDC Receipt if Work variable are similar. The coefcients on these two variables tend to both be large in the same specications with their opposite signs canceling each other out.
We nd little effect of Medicaid on the employment decisions of single mothers. Theory predicts that the Medicaid if Work variable will have a positive effect on employment. The variable has the opposite effect from this prediction in both samples, although the coefcient estimates are small and usually are not signicantly different from zero. This result is not completely unexpected given the weak and conicting ndings in past work.