THE LABOR SUPPLY OF SINGLE MOTHERS
 and Levine and Whitmore  who found strong effects of leads of waiver variables on caseloads.
E. Training and Child Care
The last three coefcient estimates in Table IV measure the employment effects of expenditures on training and child care. Higher expenditures on job search and other training and on child care are associated with a higher employment rate for single mothers. Training expenditures on education have a negative effect that is signicant in both samples. The job search coef- cients imply that an increase in expenditures of one thousand dollars (about two-thirds of average expenditures) would increase the employment rate for single mothers without young children by over four percentage points. Since single mothers without children young enough to exempt them from training programs make up about half of all single mothers, the overall effect would be over two percentage points. An increase in federal and state child care expenditures of ve hundred dollars per single mother with a child under six (slightly less than the mean in 1996) is associated with about a one percentage point increase in both weekly and annual employment. These effects are quite substan- tial per dollar expended. The training result on education is not surprising given the weaker results in the literature on classroom training and the possible short-term effect on employment as women are in classrooms rather than jobs.
F. Results by Education Group
Table IV also reports separate estimates for the effects of the policy variables for three education groups: less than high school, high school, and some college. We would expect a priori that the policy variables, which mostly capture taxes and benets received by low-income women, would have the greatest effect on high school dropouts, less of an effect on those with a high school degree, and even less of an effect on those with some college.22 Overall, the results by level of education are consistent with the hypothesized larger effects on the less educated. The derivatives tend to be much larger in absolute value for high school dropouts than they are in the full sample, and much smaller for those with
22. The estimates use a xed wage/hours distribution (that does not vary by education) to calculate the income and benet variables so that the explanatory variables are comparable across the columns.