1. Whether this will turn out to be the case is an empirical matter. The evidence to date is not so clear that women with greater labor market skill have necessarily left the rolls. See Cancian et al. (2000), Danziger (2000), Loprest and Zedlewski (1999), Moffitt and Stevens (2001), Oellerich (2001), and Zedlewski and Alderson (2001).
2. See Brauner and Loprest (1999) and Acs and Loprest (2001) for reviews. For studies that examined heterogeneity among leavers, see Cancian et al. (1999), Moffitt and Roff (2000), and Ver Ploeg (this volume).
3. See Moffitt and Stevens (2001) for a study of how the types of women on welfare have varied over the business cycle in the past, and whether the change in the types of women on welfare after 1996 was different than what would have been expected from the effects of the economy alone.
4. See, for example, Ellwood (1986), Ellwood and Bane (1994), and Gottschalk and Moffitt (1994a). The U.S. Department of Health and Human Services (2000) uses a total-time-on definition of welfare dependence as well.
5. In some of their discussion, Ellwood and Bane (1994:40) suggest that cyclers are a subset of long-termers rather than constituting a parallel category. This is a slightly different definition of what is meant by a "long-termer," as will be discussed in this chapter.
6. We do not list H(i,t) as a source of heterogeneity because, at any point in time, it arises completely from the other three exogenous factors we have listed. There will be no need to distinguish between state dependence and heterogeneity here, given the goals of the analysis. Furthermore, we ignore initial conditions problems because the data will allow us to observe all women reasonably close to t=0, the start of the process.
7. The initial conditions at t0 also must be included but, as noted previously, the data will start reasonably close to t0 =0, so no conditioning is necessary.
8. Mathematically, let y(i,t) = m(i) + e(i,t), where y(i,t) is earnings for individual i at time t, m(i) is permanent earnings, and e(i,t) is per-period transitory earnings. Assume that an individual goes off welfare in any period t if y(i,t)>b, the welfare benefit. If e(i,t) has the same variance for all individuals but individuals differ in their level of m(i), then the rate of turnover of an individual will be directly proportional to how close m(i) is to b. Both those with very low m(i) (relative to b) and those with very high m(i) will have low turnover rates, while those with m(i) close to b will have high turnover rates.
9. Some argue that more time on welfare also increases the perception by employers that an individual has low job skills.
10. Gottschalk and Moffitt (1994b:Table 1), indeed, found that the individual-specific level of permanent earnings is negatively correlated with the variance of earnings around that level.
11. The inclusion of both left-censored and right-censored spells, and the counting of their lengths as the lengths of a spell, is appropriate in the application here because such a spell concept is the appropriate one for a decomposition of a total-time-on measure defined over a fixed calendar interval. The only danger is that, because censored spells will be shorter than their completed counterparts, there will be an undercount of individuals with long completed spells. To the extent that the labor market skill measures that will be the main focus of our analysis are more weakly correlated with these censored spell lengths than their uncensored counterparts, our correlations of spell length with skill will be weakened. However, as we shall describe, our calendar interval is 10 years long and hence there are few censored spells relative to the total number of spells.
12. As noted in the Introduction, Bane and Ellwood in some passages suggest that long-termers are those with high total-time-on, for example.
13. A difference in the Ver Ploeg and Cancian studies, on the one hand, and the Stevens study and this study, on the other, is that the former were point-in-time samples composed of families on the rolls at a point in time, whereas this study and that by Stevens contain all women ever on welfare in a 10-year period. The former studies will omit short-term spells not in progress at the point in time at which the sample is drawn.
14. The interviews gathered information on AFDC recipiency for the year prior to interview in a list format prior to 1993 and in an event-history format in 1993 and after, the latter format providing the start and end dates of all spells since the last interview. The calendar period for which recipiency is available is therefore January 1978 through the 1996 interview date. We use only 1978-95 calendar years.
15. Despite this, all tables in this chapter were also estimated over the entire 18 years of data. With that longer period, there are a substantially greater number of cyclers than reported, and average spell lengths of long-termers are shorter. However, none of the critical results on the differences in labor market characteristics by T, N, L, and long-termer/short-termer/cycler status reported are different.
16. We exclude women who have missing data for any of the 120 months (276 women are excluded for this reason).
17. The Appendix to this chapter is comprised of four tables with auxiliary data that will be referred to throughout the chapter.
18. Exact figures are given in Table 14-A1.
19. Exact figures are given in Tables 14-A2 and 14-A3. As noted in the last section, left-censored and right-censored spells are included as "spells" in these tabulations. However, only 3.4 percent of the sample was on AFDC in the first month of age 20 and only 2.6 percent were on in the last month of age 29.
20. Administrative data may show more turnover because of administrative churning.
21. Because earnings and wages are measured only annually, the measures are all computed only over those years when the woman was not on AFDC at all (because otherwise, some of the earnings and wages might have been earned while on AFDC). For each woman, her mean employment (whether worked at all in the year), annual earnings, weekly wages, and hourly earnings are computed for each year she is off AFDC, then averaged to obtain a mean for her non-AFDC periods. The figures in Table 14-1 represent the means of these figures, taken over all women in the sample.
22. Note that, in this classification, short-termers could have higher T than long-termers if a recipient in the former category has two spells and a recipient in the latter group has only one. This illustrates, once again, that this typology is not perfectly correlated with T (nor should it be, by concept, as discussed earlier). Nevertheless, despite this possibility, long-termers will be seen to have much larger T than short-termers on average.
23. This implies that an even larger percentage of the point-in-time caseload would be long-termers.
24. See Stevens (2000) and Ver Ploeg (this volume: Chapter 3) for exceptions. Using Maryland administrative data, Stevens estimated a smaller fraction of cyclers (about 20 percent) and a larger fraction of short-termers (50 percent), but about the same fraction of long-termers (30 percent). Using Wisconsin data, Ver Ploeg finds that cyclers constitute 14 percent of the caseload, while long-termers constitute 55 percent and short-termers constitute 31 percent.
25. Exact figures are given in Table 14-A4 as are the figures for Definitions 1 and 3.
26. See Table 14-A4. Cyclers have much shorter mean and median spell lengths than long-termers, as expected.
27. The T cutoff was chosen to divide the cycler sample in half, that is, the approximate median T was used.
28. It is possible that selection bias is at work and that those long-termers who are observed to have worked have above-average wages. However, the employment rates of the two groups are not far different, suggesting that this justification is unlikely to be a major source of the explanation.
"01.pdf" (pdf, 472.92Kb)
"02.pdf" (pdf, 395.41Kb)
"03.pdf" (pdf, 379.04Kb)
"04.pdf" (pdf, 381.73Kb)
"05.pdf" (pdf, 393.7Kb)
"06.pdf" (pdf, 415.3Kb)
"07.pdf" (pdf, 375.49Kb)
"08.pdf" (pdf, 475.21Kb)
"09.pdf" (pdf, 425.17Kb)
"10.pdf" (pdf, 424.33Kb)
"11.pdf" (pdf, 392.39Kb)
"12.pdf" (pdf, 386.39Kb)
"13.pdf" (pdf, 449.86Kb)