How Effective Are Different Welfare-to-Work Approaches? Five-Year Adult and Child Impacts for Eleven Programs. Conclusions


The costs of services were relatively high everywhere. This was true for controls as well. Gross costs for employment-focused programs were similar to those for education-focused programs. However, net costs were lower for employment-focused programs. Similarly, LFA programs continued to have lower costs than HCD programs.

The NEWWS programs mostly left sample members worse off. Where there were gains to the welfare sample, they were very small. In half of the programs, the average earnings gains over the five-year period were smaller than reductions in welfare and Food Stamps. In Atlanta, which had the lowest welfare benefit levels, program group members had less to lose. As a result, both programs in this site produced earnings gains that were greater than the loss in combined welfare, Food Stamp, and Medicaid benefits. Detroit was the only other program to produce earnings gains greater than its reductions in transfer benefits. However, it bears repeating that the gains were very small and could easily have become losses had leisure and out-of-pocket work expenditures been included.

Except for Portland and Grand Rapids LFA, sample members without a high school diploma or GED at random assignment fared worse than those with these credentials regardless of the program approach. However, even among those who entered the evaluation with a high school diploma or GED, programs mostly left sample members worse off.

Although earnings increases were too small to cover benefit losses for program group members, in most programs the savings to government budgets were not large enough to cover their investments. The three programs that produced gains to government budgets were all employment-focused and had among the lowest net costs. Similar results were seen for sample members who entered the programs with a high school diploma or GED. That is, within this subgroup, the four employment-focused programs broke even or produced gains to government budgets, and all of the education-focused programs produced losses. For the subgroup without these credentials, all programs, except for the two in Atlanta, produced gains to government budgets. In general, the greater investment of education-focused programs did not pay off for government budgets or for program group members.

For the full samples, no program produced gains to both the welfare sample and government budgets. This was true for the educational attainment subgroups as well, except for nongraduates in Portland: Small gains ($1,084 over five years) accrued to nongraduates in that program, and the government budget experienced a moderate gain of $2,787. (Detroit produced a small gain to the government budget but broke even from the perspective of the welfare sample.)


1. See Hamilton et al., 1997; Scrivener et al., 1998; Farrell, 2000; Storto et al., 2000; and Scrivener and Walter, 2001.

2. See the two-year reports mentioned in footnote 1 for more detailed information on the calculation of unit costs.

3. As noted in Chapter 2, unlike those in other programs in this analysis, Columbus sample members were randomly assigned to the two programs prior to an orientation. These differences in sample composition are reflected in the cost and benefit outcomes.

4. A unit cost for basic education in Detroit is not available. Table 13.2 presents an "education and training" unit cost for Detroit, which combines vocational training, post-secondary education, and basic education. (See footnote "e" on Table 13.2.)

5. The Portland, Grand Rapids, and Detroit programs required welfare recipients with children as young as age 1 to participate in the NEWWS program, whereas the requirement extended only to recipients with children as young as age 3 in the other programs. The Grand Rapids and Detroit programs did not have particularly high child care costs.

6. A comparison of child care costs for the Columbus Integrated and Traditional programs does not support the idea that the case management approach alone was responsible for this result.

7. The Two-Year and Five-Year Client Survey data, along with welfare payment records, were used to estimate participation in employment-related activities that sample members took part in when they were not receiving welfare.

8. In Riccio, Friedlander, and Freedman, 1994, the GAIN costs were presented in 1993 dollars and were as follows: Alameda, $6,977; and Los Angeles, $6,402.

9. A portion of job search participation that occurred after the embargoes were lifted was considered program-related. The proportion of participation that was considered in-program was estimated based on the number of months that sample members received welfare in the post-embargo period. All participation in education and training activities was considered to be self-initiated.

10. The Riverside comparison includes only HCD and LFA program group members without a high school diploma or GED at random assignment.

11. Many of the techniques were originally developed for the benefit-cost analysis conducted as part of the MDRC Demonstration of State Work/Welfare Initiatives (for additional information, see Long and Knox, 1985). In this report the description of the analytical approach was adapted from previous MDRC reports (Riccio, Friedlander, and Freedman, 1994; Kemple, Friedlander, and Fellerath, 1995; Miller et al., 2000; and Bloom et al., 2000a).

12. In programs such as these, particularly education-focused programs, many costs are incurred early, when welfare receipt is heaviest; however, many benefits, such as earnings gains, continue to be realized in later years. Simply comparing the nominal dollar value of program costs with benefits over multiple years would be problematic because a dollar's value is greater in the present than in the future: A dollar available today, to either sample members or the government, can be invested and may produce income over time, making it worth more than a dollar available in the future. In order to make a fair comparison between benefits and costs over multiple years, it is essential to determine their value at a common point in time  for example, the present. In benefit-cost analyses, this is often accomplished by discounting, a method for reducing the value of benefits and costs accrued in later years relative to benefits and costs accrued in early years. In this analysis, the end of the first year following random assignment was used as the comparison point for the investment period. In other words, gains that were accrued after that point were discounted to reflect their value at the end of year 1. In calculating these discounted values, it was assumed that a dollar invested at the end of year 1 would earn a real rate of return of 5 percent annually (this assumption has been used in other MDRC benefit-cost analyses).

13. Some work-related expenditures were paid by the welfare agencies and are reflected as a cost to government budgets in the support services line.

14. The presentation of benefit-cost results in this report was adapted from previous MDRC reports (Miller et al., 2000; Bloom et al., 2000a).

15. Data limitations did not allow for an accounting of out-of-pocket costs paid by sample members for health care coverage.

16. These rates were based on published information on employers' compensation costs for 1995 from the U.S. Department of Labor, Bureau of Labor Statistics.

17. This analysis does not account for out-of-pocket child care costs. With the exception of the Portland program, no out-of-pocket cost differences were found between program and control group members in the programs for which Five-Year Client Survey data were available.

18. The federal Earned Income Tax Credit is a credit against federal income taxes for taxpayers with annual earnings below a certain level. For 1995, taxpayers with earnings up to $26,673 were eligible for the EITC. Not all eligible taxpayers receive the EITC. As has been the practice in earlier benefit-cost analyses performed by MDRC, the EITC "take-up" rate was estimated at 80 percent.

19. Income from earnings was used in calculating federal and state income taxes. Income from earnings and welfare benefits was used in calculating sales and excise taxes. Sales and excise tax rates were based on information from state tax boards.

20. This analysis does not include an accounting of public health care coverage for children through programs for low-income families, such as the Child Health Insurance Program.

21. Take-up rates for programs without Five-Year Client Survey data were estimated from those with these data.

22. Average statewide Medicaid costs per eligible month were calculated using 1995 data from the Health Care Financing Administration Web site (

23. The costs of administering welfare, Food Stamp, and Medicaid benefits were estimated using statewide administrative cost data.

24. One of the hypotheses of the case management experiment in Columbus was that integrated case management would allow more efficient delivery of services. As discussed in Scrivener and Walter, 2001, integrated case managers reported seeing such benefits. This analysis assumes that the costs of administering benefits were the same for both programs. Because the five-year administration net costs were relatively low ($379 for the Integrated program and $265 for the Traditional program), even if the Integrated program had realized some savings in this area, the overall cost of the program would still have been slightly higher than the cost for the Traditional program. That is, the combined net operating and transfer administration costs for the Traditional program were $4,037, about $150 less than the net operating cost of the Integrated program.

25. Both scenarios counted all participation as "in-program" for those who received welfare for the full period and all participation as "out-of-program" for those who did not receive any welfare during the entire period.