QUARTERLY JOURNAL OF ECONOMICS
ing at all in a given tax year, but for some could decrease weeks worked. If our goal is to provide a sharp test of theoretical pre- dictions, whether a woman worked last year is a better outcome measure. We report both measures with the expectation that the effects of many of the recent policy changes on weekly employ- ment will be smaller than on annual employment.
The employment rates reported in Table II exhibit a striking time pattern. For single mothers weekly employment increased by almost 6 percentage points between 1984 and 1996, while annual employment increased over 8.5 percentage points. Most of this increase occurred between 1991 and 1996.18 Focusing on the subsample of single mothers with young children, the employ- ment increases were even larger: 10 percentage points for weekly employment and 13.5 percentage points for annual employment. In contrast, the declines in both weekly and annual employment of about one percentage point for single women without children suggest that the rising employment of single mothers was not a result of better work opportunities for all single women. More- over, the timing of the employment increases suggest that policy changes in the 1990s are likely to have played a large role.
B. Comparing Single Mothers and Single Women without Children
Appendix 2 reports descriptive statistics for single women with and without children for the years 1984, 1988, 1992, and 1996. The table indicates that single mothers tend to be older and less educated and are more likely to be nonwhite than single women without children. The age of single women without chil- dren rises appreciably over the sample period, as does the edu- cation level of single mothers. The fraction of single mothers living with parents is stable, while the rate for single women without children falls. The rates of cohabitation rise for both single women with and without children.
A potential criticism of the Table II results (and our main
18. One concern in interpreting changes in employment for single mothers during the years 1992 to 1994, is that beginning in January 1994 the CPS used a redesigned questionnaire. For a description of this CPS redesign, see Cohany, Polivka, and Rothgeb , and Polivka and Miller . In Meyer and Rosen- baum  we assess the extent of any bias due to the redesign using the parallel survey which provides contemporaneous responses using the new and old surveys. We also employ ORG/March comparisons using the fact that redesign affected the two data sets at a different point in time. Overall, these comparisons indicate that the CPS redesign had a small effect that, if it leads to any bias, suggests that we slightly understate the recent employment increases of single mothers.