QUARTERLY JOURNAL OF ECONOMICS
tion, marital status, marital status interacted with a children indicator, the number of children under six and eighteen, the state unemployment rate, the state unemployment rate inter- acted with a children indicator, (for the March CPS only) controls for pregnancy, central city and unearned income, and (for the ORG only) controls for month and month interacted with a chil- dren indicator. Note that the difference-in-differences calculated by subtracting one YEAR ANYCHILDREN coefcient from another are hardly affected by including the controls.19 For ex- ample, between 1984 and 1996 the weekly employment of single mothers relative to single women without children rises 7.1 per- centage points without controls and 6.8 percentage points with controls.20 For annual employment, the difference-in-differences estimator for 1984 to 1996 suggests an 11.7 percentage point increase in the relative annual employment of single mothers without controls and an 11.3 percentage point increase with controls. Again, most of the increase occurs between 1991 and 1996. Therefore, these difference-in-difference estimates suggest
a potential role for policy changes, especially since 1991.
VI. POLICY VARIABLES AND EMPLOYMENT USING OUR SIMPLE STRUCTURAL MODEL
We now move on to our main approach that uses our simple structural model to distinguish between the different policies and to provide estimates that have a clearer interpretation. While some of the estimates rely on comparisons of single mothers and single women with children over time, other estimates use a variety of other sources of identifying variation in our key explana- tory variables. In some specications, the identifying variation comes from differences in taxes and benets for families of dif- ferent sizes and in different states, as well as changes in these taxes and benets over time, and differences in state living costs.
Table IV reports estimates of our structural model of the
19. Due to the difculty in gauging the magnitude of probit coefcient esti-
mates, instead we report derivatives of the probability of working with respect to each of the explanatory variables, averaged over the single mothers in the sample. Thus, differences in the average derivatives for the YEAR ANYCHILDREN variables give changes over time in the difference in employment between single women with and without children, analogous to the changes that can be calculated
from Table II.
20. The “without controls” results come from a weighted probit including only
the year dummies and YEAR