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


As noted above, it is important to consider whether other household members received income from work or other sources and may have contributed toward the purchase of food, clothing, shelter, and other necessities for respondents and their children and helped respondents to experience a higher standard of living than they could have attained from their income alone.(17) Higher household income may also be associated with favorable outcomes for children, an issue explored in Chapters 11 and 12.

Respondents to the Five-Year Client Survey reported whether other members of their household received income from employment, welfare, Food Stamps, child support, SSI, social security, or other sources at the end of year 5. (See Appendix Table E.4, lower panels.) About 55 percent of control group respondents in Grand Rapids, Portland, and Riverside and about 40 percent in Atlanta reported living with someone who received income from one or more of these sources. Most programs did not affect the likelihood of living with others who received income, although the decrease for Riverside LFA (of 4.2 percentage points) was just above the 10 percent level of statistical significance.(18)

A large majority of control group respondents reported that they were living in a household in which at least one person worked for pay: from 65.5 percent in Atlanta to 80.3 percent in Grand Rapids (Appendix Table E.4, upper panels). Furthermore, a fairly large proportion of control group membersВ  42 to 43 percent in Grand Rapids, Portland, and Riverside and 28 percent in AtlantaВ  lived with at least one person who was working for pay (Appendix Table E.4, lower panels). Fewer than half of these respondents (or from 10 to 19 percent of all control group members) reported that they were jobless but other household members were working for pay (Appendix Table E.5, lower panels).

Programs did not affect the likelihood of living with another person who was the sole wage-earner in the household or living in a household in which the respondent and another household member were working for pay. Similarly, programs had no impact on the proportion of sample members who lived in a household in which monthly earnings for other household members totaled $1,000 or more (results not shown).


1.В  Morris et al., 2001; Bos et al., 1999; Knox, Miller, and Gennetian, 2000.

2.В  See, for example, Appendix Table E.4, which displays income receipt at the end of year 5, based on survey data. A relatively small percentage of respondents reported receipt of income from sources other than earnings, welfare, and Food Stamps. The main exceptions are child support payments in Grand Rapids and Portland.

3.В  See Chapter 10 for a discussion of out-of-pocket expenses for child care.

4.В  It was beyond the scope of this evaluation to measure the EITC and payroll taxes directly. Instead, these outcomes were estimated on the basis of sample members' measured earnings, rules for calculating the EITC and taxes, and assumptions about the percentage of sample members who applied for the EITC on their federal income tax return. For each sample member, EITC calculations for years 1 to 5 use the tax rules for the calendar years in which the quarters 4, 8, 12, 16, and 20 occurred. Tax years range from 1992 through 1996 to 1995 through 1999, depending on when sample members were randomly assigned. Calculations assume an 80 percent take-up rate (see Scholz, 1996). Specifically, sample members' EITCs were estimated for each year of follow-up and then multiplied by 0.8. Estimated payroll taxes were calculated by multiplying earnings by 7.65 percent, the tax rate during most of the follow-up period.

5.В  As noted above, it was assumed that 80 percent of sample members with earnings received EITCs. To test the effects of this assumption, sample, sample members' combined income was recalculated, assuming that 100 percent of sample members with earnings received thereceived the EITC. The higher rate made the impact estimates on combined income more positive by about $170 for Portland and $90 for90 for Riverside LFA. However, impact estimates for other programs changed verychanged very little. (Results not shown.)

6.В  P-values were 0.11 for Atlanta LFA and HCD and 0.15 for Portland.

7.В  The decreases for Grand Rapids LFA (-$1,433) and Riverside HCD (-$2,387) were statistically significant. The decreases for Grandfor Grand Rapids HCD (-$1,247, p-value = 0.11) and) and Riverside LFA (-$875, p, p-value = 0.16) were) were just above the 0.1 level of statistical significance.

8.В  See Riccio, Friedlander, and Freedman, 1994, for GAIN; Hamilton and Friedlander, 1989, for SWIM; Freedman et al., 2000b, for Jobs-First GAIN; and Kemple, Friedlander, and Fellerath, 1995, for Project Independence. See Bloom and Michalopoulos, 2001, for a synthesis of these and other studies.

9.В  Trends in average combined income for the program or control group may be misleading because they do not include other sources of income for sample members and their households. However, it is reasonable to assume that excluding these sources did not bias the cumulative or yearly estimates of impacts on combined income.

10.В  Appendix Table E.2 shows the cumulative effects of the programs on income from earnings and public assistance (that is, without estimated EITC payments and payroll taxes) over the first three years and for the last quarter of year 5.

11.В  For Portland, only the impact for year 4 was statistically significant, although p-values for years 2 (.139) and 3 (.102) were just above the .1 level of statistical significance. The increase in year 5 of $470 had a p-value of .245. Detroit's program also led to higher income during years 2 to 4, but the differences were small and not statistically significant.

12.В  Appendix Table E.3 shows similar results for the last quarter of year 5. Since the programs in general had small effects on employment during the last quarter of year 5, they also had relatively small effects on these outcomes. However, several, several of the programs continued to significantly reduce the proportion of people who were on welfare and not working. Notably, Portland, Portland increased the proportion of sample members who were working and not receiving welfare by 5.8 percentage points relative to the control group, by far the largest effect on this measure of self-sufficiency.

13.В  The other three employment-focused programsIn addition, Atlanta and Grand Rapids LFA both led to 1 percentage point gains in this measure. The fact that people were not able to leave welfare when they went to work implies that they earned fairly little, perhaps because they worked part time or earned close to the minimum wage. In addition, sample, sample members may have received welfare and earnings at different times during a quarter as they transitioned from welfare to work or from work back to welfare. MDRC did not investigate these issues.

14.В  Across all sites, program and control group members received income from Food Stamps only (no earnings or welfare payments) during 2 to 7 percent of follow-up quarters. Programs did not affect the likelihood of having income from Food Stamps only.

15.В  Appendix GH compares employment levels and impacts calculated with UI earnings and survey data. As discussed in Appendix G, survey respondents in Riverside and Portland reported considerably more employment than was recorded on UI earnings records. Furthermore, as will be discussed in greater detail below, according to responses to the Five-Year Client Survey, virtually all sample members had some type of income in the month prior to completing the survey. (See Appendix Table E.4.)

16.В  Some respondents also reported earnings that had not been recorded by the UI system. See Appendix H.

17.В  Some household members who received income may not have contributed financial support to sample members and their children. For instance, people receiving SSI or other types of assistance may have neededhave needed to use all of their income to meet their own needs.

18.В  A larger reduction for LFA nongraduates was statistically significant. The Portland program led to a program-control group difference of +6.0 percentage points that was not statistically significant (p-value = .21). The p-value for the Riverside LFA impact was .13.