Table of Contents | Previous | Next |
14. VOLUNTARY PROGRAMS
There are four voluntary evaluations in our database (Supported Work, Homemaker Health Aide, Maine’s Training Opportunities in the Private Sectors, and New Jersey’s Grant Diversion Project) that at least superficially appear similar. They all evaluated programs that paid a stipend to AFDC recipients who volunteered for temporary jobs that were intended to help them learn work skills.
Summary information about the results of these evaluations appears in the first two columns of Table 27. Impact estimates were only available from all four evaluations for earnings and the amount of AFDC received. Ten impact estimates are available for the 7th calendar quarter because one of the four evaluations (the Homemaker Health Aide evaluation) assessed somewhat different programs that were independently operated by six different states. One evaluation (Supported Work) did not estimate impacts for the 11th quarter so only nine observations are available for that quarter. Impact estimates are not available for the 15th quarter, and we do not provide impact estimates for the 3rd quarter because some members of the program groups were still actively participating in the programs then and some of the evaluations counted the stipends they were receiving as earnings.
The means and medians shown in Table 27 indicate that the evaluated voluntary programs increased the earnings of members of the program group by a modest amount in the 7th quarter and by somewhat more in the 11th quarter. Conversely, impacts on AFDC payments were larger in the 7th quarter than in the 11th quarter.
The standard deviations and maximum and minimums in the first two columns of Table 27 suggest that there is fairly substantial variation in these impacts. Whether this variation is due to systematic differences among the intervention or merely attributable to sampling error is investigated by Q-tests. Results from this investigation are shown in the second column of the table. The fact that the Q-test statistic is well below the critical value for a chi-squared for the 7th quarter earnings impacts implies that all the variation in these impacts can be attributed to sampling error. However, the Q-statistic is above the critical value for a chi-squared for the variation in 11th quarter earnings, suggesting that some of the variation in these impact results from systematic differences among the interventions. The hypothesis that the variation in the impacts in AFDC payment impacts is due entirely to sampling error is resoundingly rejected for both quarters.
We explored different possible explanations for variation in the impact estimates that is attributable to systematic differences among the interventions and, thus, is not a result of sampling error. The most promising explanation appears to be that, unlike mandatory welfare-to-work programs, more expensive voluntary welfare-to-work produce larger impacts than less expensive programs.
To examine this possibility, we used estimates of operating costs per program group member, which range between $2,500 and $16,000 (in year 2000 dollars), to divide the ten interventions into two subgroups of equal size: high cost programs and low cost programs. As indicated by Table 27, with the exception of the 7th quarter earnings impact, impacts were considerably higher for the high cost programs than the low cost programs. To investigate this further, we computed weighted regressions in which operating costs per program group member was the only explanatory variable. As expected, the coefficient on this variable is very small and statistically insignificant for the 7th quarter earnings impacts. The other three coefficients are statistically significant at the 5-percent level, however. More specifically, the results indicate that a $100 increase in expenditures per program group member results in impacts on earnings that are almost $5 larger in the 11th quarter and in impacts on AFDC payments that are about $5 larger in the 7th quarter and $2.50 higher in the 11th quarter.
The Q-test statistics in the right-hand column of Table 27 imply that, with the exception of the 7th quarter impacts on AFDC payments, the variation among the impacts for the low cost interventions is due to sampling error. The Q-test statistics in the fourth column indicate that the variation among the earnings impacts for the high cost interventions is also caused by sampling error, but that much of the variation among the AFDC payment impacts results from systematic differences among these interventions. With only five high cost interventions with which to work, we were unsuccessful in discovering the sources of these systematic differences.
| Table of Contents | Previous | Next |