# THE LABOR SUPPLY OF SINGLE MOTHERS

1067

over the 1984 to 1996 period that affected the labor supply of single mothers. We also provide summary statistics on program changes, discuss their theoretical impacts on labor supply, and introduce variables that measure particular policies. Section V compares the employment rates and other characteristics of sin- gle mothers and single women without children. Section VI in- vestigates how the employment of single mothers was affected by the policies we study. We also examine alternative explanations for our results and briey examine hours worked. Section VII provides an accounting of the contribution of different policy changes to the overall increase in employment of single mothers in recent years. We then offer conclusions in Section VIII.

# II. MODELING THE WORK DECISION

Our modeling approach combines some of the best aspects of structural methods and quasi-experimental or natural experi- ment type approaches. Beginning from a structural approach claries which variables should enter the work decision and the form in which they should enter. Our simple structural model also allows us to test some fundamental economic predictions and more convincingly simulate policy changes.^{7 }The quasi-experi- mental methods make transparent the assumptions that allow the identication of our key coefcients. By the appropriate use of control variables and simplifying assumptions, we identify our key parameters using only the sources of variation in our ex- planatory variables that we believe are exogenous.

We focus on employment because previous work has found that women are more responsive to wages and income in the decision to work than in the hours decision (see Heckman [1993]). The probability that a single woman works is just the probability that the expected utility when working U_{w }exceeds the expected utility when not working U_{n }_{w }; i.e., Pr[U_{w }> U_{nw }]. We take utility to be a function of income Y, nonmarket time L, an indi- cator for welfare participation P (which captures transaction costs or stigma), other demographic and other control variables X, and an additive stochastic term . Thus, the probability of work is just

7. Because of the simplications we make to improve the model’s tractability, one may not want to consider our approach fully structural. As with any struc- tural model, simulations that rely heavily on simplifying assumptions may give misleading results.