THE LABOR SUPPLY OF SINGLE MOTHERS
over the full sample period. The only exception to this generali- zation is that the job search/other training coefcient is larger and the child care coefcient is smaller and statistically insigni- cant over the shorter time span. These results are among the most important in this paper, because they indicate that (1) the
urry of welfare reform measures after 1993 has not falsely led to
our main results, and (2) the extended recovery of the 1990s is not an alternative explanation for our main results.
Next, we examine a sample of only single mothers. This specication identies the effects of the income variables through changes across states and for different family sizes. In the case of the Income Taxes if Work variable, we are largely using the variation from the last few years when the EITC for women with one child was nearly unchanged but the EITC for women with two or more children rose in large steps. Thus, identication comes from using women with one child as a control group, and chang- ing the treatment that women with two or more children receive. With single mothers only, the year indicators remove the time trend in welfare receipt and benets, and the state indicators remove time-constant differences in state welfare benets and much of these state cost of living differences in the income vari- ables. Thus, the variation in welfare benets used to identify the coefcients is now changes in state-level benets. This identi- cation approach examines the employment response to fairly subtle or short-run features of the welfare and tax laws. These policy changes may be overwhelmed by other factors in these specications. Despite these potential difculties, much of the income tax effect remains, although the estimates are much smaller. While the effect of taxes is still signicant in the March CPS data, the drop in the coefcient and larger standard error leads the ORG coefcient to be insignicantly different from zero. The welfare benet coefcients are now no longer signicant. The AFDC transaction cost coefcient, however, remains signicant in the ORG data, while the Medicaid coefcient has the expected sign, but remains small and insignicant in both samples.
In the third set of specications of Table V, we only include single mothers with a child under six (and single women without children). The derivative estimates for the tax and welfare vari- ables, including waivers, are often substantially larger in magni- tude for these single mothers with young children, especially for the tax variable in the ORG sample. These specications are of particular interest, because the effects of increased employment