Table 13-18 shows probit estimates of the probability of leaving welfare for all July 1995 AFDC recipients. Estimates from two models that use different measures of past welfare receipt history are shown. The first model uses average spell length (ASL) and ASL-squared along with a series of dummy variables categorizing the number of spells of AFDC receipt the case had in the preexit period. The second model uses the long-termer, short-termer and cycler distinctions to synthesize the two concepts of spell length and number of spells. Both include a variable for the number of quarters for which the case had UI earnings during the preexit period, and controls for demographic characteristics of the case and for local economic conditions. The sign of the coefficient and the marginal effect of each variable on the probability of leaving welfare are given.
|Independent Variable||Model 1||Model 2|
|Average spell length||-(a)||-0.006|
|Average spell length squared||+(a)||0|
|Total # of spells of AFDC receipt = 1||-(a)||-0.005|
|Total # of spells of AFDC receipt = 2 or 3||+||0.018|
|Total # of spells of AFDC receipt= 4 or more||+||0.036|
|(reference group is those with no spells of AFDC receipt)|
|(reference group is short-termers)|
|# quarters with earnings before leaving||+(a)||0.011||+(a)||0.012|
|Age of case head||+(a)||0.008||+(a)||0.008|
|Age of case head squared||-(a)||0||-(a)||0|
|(reference group is white)|
|No high school diploma||-(a)||-0.037||-(a)||-0.037|
|At least some college||+(a)||0.027||+(a)||0.027|
|(reference group is high school diploma)|
|Age of youngest child||+(a)||0.008||+(a)||0.007|
|# of children under age 5||-||-0.038||-(a)||-0.039|
|# of children over age 5||-(a)||-0.021||-(a)||-0.024|
|Average other adult present in case||+(a)||0.038||+(a)||0.04|
|Milwaukee County resident||-(a)||-0.194||-(a)||-0.206|
|Resident of other urban county||-(a)||-0.057||-(a)||-0.06|
|(reference group is rural county resident)|
|Child receives SSI||-(a)||-0.025||-(a)||-0.028|
|Mother receives SSI||-(a)||-0.325||-(a)||-0.326|
|Unemployment rate in county July 1995||-(a)||-0.007||-(a)||-0.006|
|Log likelihood (restricted log likelihood is -24,069.95)||-21,105.04||-21,203.53|
|Likelihood ratio index||0.123||0.119|
|Percent of observations predicted correctly||67.4||67.1|
|a Coefficient is statistically significant at the 5-percent level.
(b) Coefficient is statistically significant at the 10-percent level.
Model 1 uses ASL and its square and dummy variables for the number of spells of AFDC receipt to characterize past welfare receipt history. 16 The categories of number of spells are: zero spells (reference group), one spell, two to three spells, and four or more spells.
Results show that longer average spell lengths are negatively associated with the probability of leaving welfare, but the marginal effect of a 1 month change in ASL has a small effect on the probability of leaving. The relationship is nonlinear, however, as ASL gets longer, the rate at which the probability of leaving decreases starts to slow. Those who had one spell are significantly less likely to leave AFDC than those with no prior spell. However, the size of the marginal effect is small, as a shift from no spell to one spell decreases the probability of leaving by only 0.5 percentage points compared to 14 percentage points. Those with two or more spells are not significantly more or less likely to leave welfare than those with no prior spells. These results suggest that when the length of time on welfare is accounted for, the number of spells of receipt does not have a big impact on the probability of leaving welfare. Those with one spell of AFDC receipt are significantly less likely to leave welfare than those with no prior spells, but those with more than one spell are no more or less likely to leave welfare. Cancian et al. (1999) found consistent results. They found that those with spells of over two years long were significantly less likely to leave welfare and that those with more than one spell were significantly less likely to leave welfare than those with only one spell.
Results from the second model corroborate this conclusion. In the fourth model, the spell length and spell number concepts of welfare receipt are combined into the cycler, long-termer, and short-termer classifications. The short-termers are used as the reference group in this model. Results show that long-termers are significantly less likely to leave welfare than short-termers, but there are no differences between cyclers and short-termers in the probability of leaving welfare. Long-termer status decreases the probability of leaving welfare by 10 percentage points, which is a sizable reduction.
To summarize estimations of the probability of leaving welfare, the distinction between long-term AFDC recipients and other types of AFDC recipients is an important one as long-termers are significantly less likely to leave AFDC. The estimates also show that the number of preexit quarters with earnings is consistently a strong and positive predictor of the probability of leaving AFDC across all four models.
"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)