As background for understanding the relationship between UI benefit receipt and return to TANF, Table 14 summarizes across all cohorts the observed rates of return to TANF by UI application and beneficiary status for TANF leavers who become unemployed. Rates of return to TANF are highest for UI applicants who do not become UI beneficiaries (see Figure 10). Rates of return to TANF by applicants who become UI beneficiaries are on par with TANF leavers who do not apply for UI after becoming unemployed. There are some variations across cohorts in mean rates of TANF return, but the across cohort, summary means are indicative of the relative rates of return to TANF for the three groups of interest: non-UI applicants, UI applicants who do not get benefits, and UI beneficiaries.
The overall mean rate of return to TANF among all newly unemployed TANF leavers is 42.7 percent. Considering subsets of this group, the mean TANF return rates are: 41.8 percent among those who do not apply for UI, 52.6 percent for those who apply for UI but do not get benefits, and 41.2 percent for those who apply for UI and do get benefits. The difference in mean rates of return to TANF among UI applicants who received benefits and those who did not was 11.4 percentage points. This suggests that among TANF leavers who become newly unemployed and apply for UI, beneficiaries return to TANF at a 21.7 percent lower rate than those who do not receive UI benefits.
|1997 Cohorts||2000 Cohorts||2001 Cohorts||Texas
|Newly Unemployed TANF Leavers||37,621||63,937||23,755||33,460||44,835||37,072||16,599||28,756||36,515||40,469||43,462||406,481|
|No UI Application||27,463||50,321||13,619||26,438||36,950||28,088||9,548||21,667||29,470||29,823||34,334||307,721|
|Apply for UI, no Benefits||4,415||5,284||4,118||2,562||5,487||3,696||3,290||2,483||4,964||4,372||3,748||44,419|
|Apply for UI, get Benefits||5,743||8,332||6,018||4,460||2,398||5,288||3,761||4,606||2,081||6,274||5,380||54,341|
|Return to TANF - Total||15,422||24,935||10,150||18,007||22,716||15,014||6,744||15,160||18,325||16,552||10,692||173,717|
|No UI Application||11,279||18,820||5,447||14,237||18,615||10,786||3,557||11,249||14,694||11,810||8,000||128,494|
|Apply for UI, no Benefits||2,179||2,700||2,212||1,678||3,114||2,011||1,663||1,604||2,767||2,217||1,226||23,370|
|Apply for UI, get Benefits||1,964||3,508||2,491||2,092||987||2,432||1,524||2,307||864||2,648||1,571||22,388|
|Return to TANF Total Rate (%)||41.0||39.0||42.7||53.8||50.7||40.5||40.6||52.7||50.2||40.9||24.6||42.7|
|No UI Application||41.1||37.4||40.0||53.9||50.4||38.4||37.3||51.9||49.9||39.6||23.3||41.8|
|Apply for UI, no Benefits||49.4||51.1||53.7||65.5||56.8||54.4||50.5||64.6||55.7||50.7||32.7||52.6|
|Apply for UI, get Benefits||34.2||42.1||41.4||46.9||41.2||46.0||40.5||50.1||41.5||42.2||29.2||41.2|
Return to TANF Rates by UI Applicant Groups
In addition to comparing unadjusted mean rates of return to TANF, the relationship of UI benefit receipt and the probability of return to TANF was estimated on cohort samples of TANF leaver UI applicants in regression models. Besides estimating the parameters on UI beneficiary variables, parameters on aspects of UI eligibility were estimated in the same models. Results presented in Table 15 suggest that among TANF leavers who are UI applicants, either quitting or being discharged from the prior job is associated with an increased probability of returning to TANF. On the other hand, receiving UI benefit payments is associated with a reduced probability of returning to TANF. Controlling for observable characteristics, receiving UI benefits was associated with a rate of return to TANF that was between 2.0 and 11.1 percentage points lower among UI applicants in the TANF leaver cohorts. These estimates are based on behavior up to four years after TANF exit. Any level of UI benefit receipt in that time is associated with a lower probability of returning to TANF.
Regression models were also estimated to measure the association between the future amount of TANF benefits received and job quit, discharge, UI monetary eligibility, and UI benefit receipt (Table 15). Results from these models suggest that quitting the prior job is not associated with a change in the amount of TANF received, but getting fired tends to be associated with an increase in the amount of future TANF benefits. There was no correlation between UI monetary eligibility and the amount of future TANF, but receiving UI benefit payments was strongly associated with reduced future TANF benefits. The estimated mean reduction was as large as $1,204 which was estimated for the Michigan 2001 cohort.
|1997 Cohorts||2000 Cohorts||2001 Cohorts||Texas
|Florida||Texas||Florida||Michigan||Ohio (*1)||Texas||Florida||Michigan||Ohio (*1)||Texas|
Return to TANF (%)
|Amount of TANF ($)|
|Notes: The dependent variables in the return to TANF models take the value of one for return to TANF, else zero. These linear probability models were estimated by ordinary least squares regression, as were the amount of TANF models. The latter were estimated on samples of TANF returnees. (*1) For the Ohio 2000 and 2001 cohorts, the regression parameter estimates on indicator variables for quit and discharge are based on UI benefit year begin (BYB) dates on or before December 31, 2002. However, the estimation samples also include new UI payment data received by the Upjohn Institute in December 2007 for UI claims with BYB dates between January 1, 2003 and December 31, 2005. These data included information on monetary eligibility and beneficiary status, but not on job separation reason or other characteristics. While there were sufficient observations to estimate parameters on quit and discharge for both models on the 2000 Ohio cohort and the model of probability of return to TANF for the 2001 Ohio cohort, there were not enough observations to estimate similar parameters in the amount of TANF model for the Ohio 2001 cohort. Characteristics variables in models on all cohorts are specific to each state and take advantage of all administrative data available. Variables included in all models are: age, educational attainment, race, prior earnings, local labor market conditions, and year:quarter of BYB.
** Difference significantly different from zero at the 95 percent confidence level for a two-tail test.
* Difference significantly different from zero at the 90 percent confidence level for a two-tail test.
The relationship between UI benefit receipt and the probability of return to TANF was also estimated in regression models on a pooled sample of 19,758 observations constructed from the 2000 TANF leaver cohorts for Florida, Michigan, and Ohio (see Table 16, Model I 2000). Similar regression models, excluding variables for quit, discharge, and job search exemption, were estimated on pooled data from Florida, Michigan, and Ohio for the 2000 and 2001 cohorts 14 (see Table 16, Model II 2000 and Model III 2001). The estimation samples were 21,377 and 17,343 for the 2000 and 2001 cohorts respectively. Variables for quit, discharge, and job search exemption were not included in the latter models, because data on these variables were not sufficiently available for the Ohio 2001 cohort.(1)
Model I provides additional evidence that job quits and discharges increase the likelihood of returning to TANF. In all three pooled models, UI benefit receipt was associated with a reduction in the mean rate of return to TANF. Parameter estimates on the UI beneficiary variable range between -9.6 and -11.6 percentage points. This range includes the simple unadjusted difference of means, -11.4 percentage points, computed between UI beneficiaries and non-beneficiary UI applicants (Table 14). Collectively, these estimates suggest that among TANF leavers who apply for UI, beneficiaries rate of return to TANF is about 20 percent lower than the 52.6 percentage rate for UI applicants who do not receive benefits.
|Variable||Cohort Parameter Estimates (*1)|
|Florida TANF Leaver
Michigan TANF Leaver
Ohio TANF Leaver
|Job Separation Reason, Quit
Job Separation Reason, Discharged/Fired
|Education, Less than High School Graduate
Education, High School Graduate or GED
Education, Some College
Education, Bachelors Degree or Higher
|Weekly Benefit Amount (WBA, $10)
WBA at Maximum
Entitlement Length of UI Benefits (Weeks)
|Earnings in UI Base Period ($1,000)
Earnings Total less than $10,000 in UI Base Period
Earnings in High Quarter of UI Base Period ($1,000)
|Multiple Employers in Any Base Period Quarter
Employed in 0 or 1 Consecutive Quarters Before BYB
Employed in 2 to 4 Consecutive Quarters Before BYB
Employed in 5 to 8 Consecutive Quarters Before BYB
Employed in 6 to 12 Consecutive Quarters Before BYB
Job Search Exempt
|Unemployment Rate as of the UI BYB Month
Unemployment Rate Change Over Benefit Year
|Notes: The dependent variables in these models take the value one for return to TANF, else zero. These linear probability models were estimated by OLS. (*1) For Model I 2000 the Ohio data include UI claims with benefit year begin dates (BYB) on or before December 31, 2002. For Model II 2000 and Model III 2001 additional UI data was used for Ohio with BYB dates between January 1, 2003 and December 31, 2005. However, the added Ohio data included only payment information from UI administrative files. Therefore, variables for quit, discharge and job search exemption were excluded from Model II 2000. Data on other exogenous characteristics for Ohio in Model II 2000 and Model III 2001 were obtained from the TANF and wage record data sets. Characteristics variables included age, educational attainment, race, prior earnings, local labor market conditions and year: quarter of BYB. Sample sizes are: 19,758 (Model I 2000), 21,377 (Model II 2000) and 17,343 (Model III 2001). Adjusted R-squares are: 0.11(Model I 2000), 0.10 (Model II 2000) and 0.11 (Model III 2001).
** Significantly different from zero at the 95 percent confidence level for a two-tail test.
* Significantly different from zero at the 90 percent confidence level for a two-tail test.
(1) Demographic variables for age, gender, and educational attainment are included in the three regression models by sets of indicator variables (Table 16). To represent each demographic variable a full set of categorical indicators was included with the restriction that the sum across all categories for a variable of the sample proportion times the category indicator is constrained to zero. OLS estimation with this restriction permits inclusion of a separate intercept term in each model. Estimated coefficients on categorical indicator variables are interpreted relative to the variable mean.