THE LABOR SUPPLY OF SINGLE MOTHERS
effects of tax and welfare policy on the probability that a woman works. These specications provide estimates of the parameters in expression (5) of Section II, and can be used to obtain estimates of the effects of the different policy changes during the 1984 –1996 period. These specications also provide coefcients that can be used to summarize the effects of a wide range of policies and that can be used to simulate other policies. In addition to the variables shown in Table IV, each of these probits include the control variables reported in Table III (except for the YEAR ANY- CHILDREN interactions) along with a large number of family
composition variables listed variables imply that we are family types to identify our
in the table notes. These control not using simple differences across coefcients. We are using changes
treated. We (1) and (5).
All of the coefcients on the income variables have the signs that are implied by our simple structural model and are signi- cantly different from zero.21 Lower taxes and maximum welfare benets increase employment, while higher welfare benets if a woman works (due to lower implicit taxes on earnings) increase employment. Rather than restricting the income variables to enter the work/nonwork decision as a single expected income variable, we have allowed the coefcients on the different compo- nents of income to differ. It is, thus, encouraging that the coef- cients on the income tax and welfare variables have roughly the same magnitude, as expected. The one exception to this rule is that the coefcient on Welfare Benets if Work in the weekly employment equation is substantially larger than the other in- come coefcients.
The Income Taxes if Work coefcient implies that a one thousand dollar reduction in income taxes if a woman works increases employment last week by 2.7 percentage points, and increases employment last year by 4.5 percentage points. Both of these effects are strongly signicant. These coefcients indicate elasticities of the participation rate with respect to the return to
21. We examined the importance of allowing for correlation among the error terms at the level of state year ANYCHILDREN using STATA. These standard errors are very close to those without this correction for clustering.