Search Type Title Author/Investigator Search Advanced Search

Table of Contents | Previous | Next

Chapter Eight

INCOME AND POVERTY

8.1. BACKGROUND

As we move from the topics covered by the prior four chapters–welfare participation, participation in other social welfare programs, labor market outcomes, and fertility and marriage–to the focus of this chapter–income and poverty–we begin to consider the broader impacts of welfare reform on families and children. (Other outcomes that also capture broader concepts of well-being, such as material well-being and child well-being, are covered in subsequent chapters.) Income is one gauge of a family’s command over resources, and poverty is one widely used metric to identify the fraction of families with resources below a specified needs standard. Thus, for some, these outcomes are among the most important to consider when evaluating the effects of welfare reform.

The impact of welfare reform on income and poverty will be determined, in part, by the effects on the outcomes covered in Chapters 4, 5, and 6. The two most significant sources of income for low-income families with children are earnings and cash transfers from means-tested social welfare programs. If welfare reform raises both income sources, we would expect total family income to increase and poverty rates to fall. Conversely, income would rise and poverty would fall if welfare reform increased earnings and transfer payments. The relationship will not be exact because other sources of income (e.g., earnings from other family members; other non-means-tested public transfers, such as unemployment insurance or disability payments; or private transfers, such as child support payments) may increase or decrease as well, thereby reinforcing or undoing the contribution of changes in earnings and means-tested programs to the changes in income and poverty. In addition, if some policies work to lower dependence on welfare while others raise earnings from work, incomes may rise or fall depending on which change is greatest.

Whether poverty rates change depends on where in the income distribution any income changes take place. If income changes are small or occur only among those already above or below the poverty line, then the poverty rate would remain unchanged. Alternatively, income changes may be small on average but still lead to changes in the proportion of families classified as poor. For example, small increases in income may be associated with reductions in poverty if the income gains occur among those with incomes near the poverty threshold.

Considering only the trends in income levels and poverty rates for the population at risk of welfare utilization (i.e., female-headed households), we find that the trends since welfare reform are quite favorable. Figure 8.1 plots the levels of annual earnings and annual income (measured off the left y-axis) and the rates of poverty and welfare utilization (measured off the right y-axis) for female-headed families between 1990 and 1999. As discussed in earlier chapters, the welfare participation rate has been steadily declining since the mid-1990s, and earnings have been on an upward trajectory over the same period. Total family income shows the same pattern as earnings, indicating that the earnings increases have been large enough to offset the decline in welfare benefits (or that other income sources have increased).62 The income gains have resulted in a decline in the poverty rate for female-headed families, which in 1999 stood at its lowest level–35.7 percent–since 1959. The child poverty rate has also fallen over this period (Haskins, 2001).

Figure 8.1–Income, Earnings, Welfare Use, and Poverty for Female-Headed Families: 1990—1999

[D]

A more focused look at the welfare population, and those who leave welfare, suggests that this positive assessment on average may not tell the whole story. Data from the eight USDHHS-funded welfare leaver studies with survey data on household income indicate that post-exit incomes remain relatively low both soon after leaving welfare (6—8 months) and also after more time has passed (26—34 months)–in the range of $1,000 to $1,500 per month (USDHHS, 2001a).63  Over one-half of the sample (57 to 58 percent) of leavers in two states and one county (Missouri and Washington, and Cuyahoga County, Ohio) were classified as poor after leaving welfare, in both the short time horizon (6—8 months) and longer time horizon (26—34 months), although this was lower than the rate of poverty for the stayers as calculated with data for Washington State.64  The post-exit poverty rate was somewhat lower in Iowa: 47 percent based on cash income, and 41 percent including the value of food stamp benefits.

There is also some evidence from these leaver studies that those whose income is above the poverty line after leaving welfare would still be considered "near poor," although some may be better off than they appear once the full range of support available to the working poor are taken into account (e.g., Medicaid, food stamps, child care, and EITC) (Haskins, 2001). At the same time, there is a small- to moderate-sized group that has very low income after leaving welfare, even accounting for these forms of support which they may or may not take advantage of. As would be expected, most income (70—80 percent) derives from the earnings of the former recipient or other family members.

A comparison of an early cohort of leavers in Wisconsin (1995) with a cohort that left two years later, shows that the later cohort–which had more barriers to work–had lower earnings and higher poverty rates after leaving welfare than the earlier cohort of leavers (Cancian et al., 2000). By comparing outcomes prior to welfare exit with those after welfare exit, Cancian et al. (2000) find that nearly two-thirds of leavers in both the early and late cohorts had higher earnings in the year after welfare exit, but only one-third had higher combined income from their own work and public assistance benefits in the post-exit period. Thus, the earnings gains were not sufficient to offset the loss of welfare benefits for most leavers. (The impact on earnings of other household members is not known in this study.) For both cohorts, poverty rates based on combined recipient income were above 65 percent in the year after exit. The early cohort, which could be tracked over three years post-exit, did show a modest improvement in own income and poverty rates over time.

While the leaver studies provide a useful perspective on the experiences of those who left welfare at a particular point in time, they are not designed to assess the contribution that welfare reform made to the observed changes. To what extent are the changes observed in the national data or those captured in the leaver studies the direct result of the welfare reforms that began with waivers and continued as part of PRWORA? Like the other outcomes we have considered thus far, confounding factors such as the economy and other policy changes (e.g., the minimum wage and EITC) may have contributed to the observed changes. Thus, our goal in this chapter is to assess what we know about the effects of welfare reform on income and poverty after accounting for these and other confounding factors.

The causal impact of welfare reform on income and poverty has been evaluated using both experimental and econometric evidence. These two outcomes typically cannot be measured accurately by relying solely on administrative data, so both experimental and observational studies also rely on survey data to construct measures of income and to calculate poverty rates. Conceptually, the ideal income measure would capture all sources available to the recipient in her own name, as well as sources available through income pooled at the family or household level. Those income sources would include earnings, cash and noncash government means-tested transfers (e.g., welfare, food stamps, Supplemental Security Income, general assistance, and Medicaid), other government transfers (e.g., unemployment insurance, disability insurance, Social Security), private transfers (e.g., child support or alimony), and income from assets (e.g., interest and dividends). Income would also be measured net of taxes paid and tax credits received (e.g., EITC).65  Family income is typically defined as all income sources for the unit of individuals who are related by blood, marriage, or adoption. A cohabiting partner of the recipient might be considered part of the family group as well, especially if income is pooled. To the extent that income pooling occurs within the household, between related or unrelated individuals–for example, in a three-generation household–the income for the entire household would be measured.

The complexities of income measurement, both conceptually and in practice, mean that such an ideal measure is rarely available. In the studies we review in this chapter, the measures of income and poverty vary between and within the experimental and econometric studies. All the econometric studies we review rely on the CPS; thus, they use a fairly comprehensive measure of family income, although they typically do not account for taxes and tax credits and noncash transfers (including food stamps).

To draw on administrative data, experimental studies often measure income just for the recipient and count only earnings, cash assistance (e.g., welfare payments and any financial work incentives such as wage subsidies from the program), and food stamps. We refer to this concept as "combined income." In some cases, taxes and tax credits such as the EITC are imputed based on the administrative data, and some evaluations have access to administrative earnings records for other family members. Some experimental studies also collect information on a broader array of income sources directly from participants. These survey data alone, or in combination with administrative data, are used to measure a broader concept of income that might include other public and private transfers sources for the recipient, and earnings and unearned income of other family members, possibly just the spouse or partner and perhaps for all other adults in the household.

Thus, in comparing studies, it is important to keep in mind that the measurement of income (and hence poverty) may be incomplete (i.e., not all sources are measured or imputed) or suffer from underreporting of certain income sources for those that are measured.66  Measures of family income in some studies may include cash sources only, while income measures in other studies may include the value of in-kind benefits such as food stamps. Income may be measured before taxes, so that taxes paid or tax credits received (e.g., the EITC) are not accounted for in the measure of well-being. Income data collected through survey methods may be measured using a recall period as short as one month, or as long as a year. Finally, the unit of observation may vary, from the narrowest perspective of the recipient’s income, to the broadest perspective of income for the family or household.

In those studies that calculate a poverty rate, the standard approach is to compare the measure of income to a needs standard that is usually specific to the family type (size and composition in terms of the number of adults and children). Almost all studies use the poverty cutoffs defined by the Census Bureau to calculate the official poverty rate, even though the income concept they use may differ from the official measure used by the Census Bureau (family pretax cash income). In some cases, the measure of income is more comprehensive than the official measure (e.g., the EITC and value of food stamps are included). In other cases, it is less comprehensive (e.g., income from assets, private transfers, and non-welfare-related public transfers are excluded). The income measure may also be based on a different unit of analysis, for example, individual income only, whereas the Census Bureau poverty cutoffs are intended to apply to a measure of family income. Thus, the concept of poverty as applied in these studies does not necessarily capture the same concept as the official measure, and the concepts are not necessarily comparable across studies.67

With this background, the next section focuses first on the results from random assignment studies with respect to income, income sources, and poverty. In that section, we also briefly summarize the results in Appendix A with respect to differential impacts for subgroups. A discussion of the econometric studies that consider these outcomes follows in the third section. The results from both types of studies are synthesized in the fourth section. A final section concludes the chapter.

8.2. RANDOM ASSIGNMENT STUDIES OF THE EFFECTS OF WELFARE REFORM ON INCOME, INCOME SOURCES, AND POVERTY

With the exception of the four programs that focus on other reforms (AWWDP, FDP, PPI and PIP), all the experimental studies reviewed in previous chapters also include at least one measure of income in their impact analyses, while some also include a measure of poverty. Table 8.1 summarizes the findings from this literature. For the relevant experiment and population served, the table indicates whether administrative or survey data are used to measure a particular income concept and then whether the income concept is used to calculate a poverty rate. In some cases, the same income measure is reported for different points in time to determine if impacts fade out or grow stronger with time; some studies report multiple income measures for the same or different follow-up periods. Since incomes are measured over varying time intervals, we also normalize all estimated impacts to a monthly concept.

In addition, we also consider the sources of income to the extent they are reported in the experimental studies. Since earnings are reported in Chapter 5, we do not repeat those results here. We do tabulate in Table 8.2 the impacts for the amount of welfare payments and food stamp payments, ideally for the same follow-up interval as the income measures. (Chapters 4 and 6 previously covered participation rates in welfare and food stamps, respectively.) Welfare payments are reported in all the studies that report income, and the same is true for food stamp payments with two exceptions: MFIP (where the Food Stamp Program was cashed out) and the Canadian SSP (where food stamps do not apply). For both income sources, the measures are typically based on administrative data for the recipient. In a few cases, survey data are used to collect information about benefits received for other family or household members. In a few studies, data on other sources of income (e.g., earnings from other household members, child support payments, and other public transfers) are collected as part of a follow-up survey. While we do not report those results in a table, we note in the text when significant impacts on these other income sources are measured.

Finally, to assess individual or family self-sufficiency, we also report in Table 8.2 program impacts for earnings’ share of total income. In some cases, this outcome is measured directly. In other cases, the fraction of the sample with 50 percent or more of income from earnings is reported. The earnings share is not available for all the studies that report income.

We follow the structure used in previous chapters and review studies by the reform policy or policies they evaluate, considering income, poverty, welfare and food stamp payments, and earnings share for each group of studies.

8.2.1. Programs That Focus on Financial Work Incentives

Three studies assess the impact of financial work incentives: WRP-IO, CWPDP, and MFIP-IO. CWPDP does not report a combined income measure, so the estimated impact in year three reported in Table 8.1 is based on the separate impacts for the recipient’s earnings (Table 5.1) and welfare and food stamp payments (Table 8.2). (Consequently, the significance level is not known.) The impact estimate for recipient combined income is close to 0, a negative $3 per month. Food stamp payments are the only combined income component that has a significant positive impact; earnings and welfare payments have negative but insignificant impacts. The negative welfare payment impact is despite the fact that there is a positive (but insignificant) impact on welfare use (Table 4.1), which reflects the fact that this program also reduced benefit levels.

Table 8.1–Estimated Impact of Welfare Reform on Income and Poverty: Random Assignment Studies

				Income				Poverty
Name	Cases served	Data	Measure	Control mean	Impact	%	Normalize to monthly	Control mean	Impact	%
A. Programs that focus on financial work incentives
CWPDP	Single-parent recipients	A	Avg. annual recipient E+W+FS income in year 3 of 42-mo FU	$8,421	-$30n.a.	-0.4%	-$3
WRP-IO	Single-parent recipients and applicants	A	Avg. quarterly recipient E-W-FS income over 42-mo FU	$2,256	$5	0.2%	$2
		A	Avg. quarterly recipient E-W-FS income in last 3 mos. of 42-mo FU	$2,307	$7	0.3%	$2
		S	Avg. mo. HH income in mo. prior to 42-mo FU	$1,410	$145**	10.3%	$145
		S	Avg. mo. HH income + EITC in mo. prior to 42-mo FU	$1,501	$139**	9.3%	$139
MFIP-IO	Urban single parents recipients	A	Avg. quarterly recipient E-W-FS income in last 3 quarters of 10-quarter FU	$2,525	$243***	9.6%	$81	77.7	-8.3***	-10.7%
		A	Avg. quarterly recipient E-W-FS income + EITC/taxes in last 3 quarters of 10-quarter FU	$2,613	$299***	11.4%	$100	70.5	-6.6***	-9.4%
		S	Avg. mo. family income in month prior to 36-mo FU	$1,459	-$11	-0.8%	-$11
	Urban single parents applicants	A	Avg. quarterly recipient E-W-FS income in last 3 quarters of 10-quarter FU	$2,578	$138	5.4%	$46	66.0	-4.0***	-6.1%
		A	Avg. quarterly recipient E-W-FS income + EITC/taxes in last 3 quarters of 10-quarter FU	$2,556	$177***	6.9%	$59	59.7	-3.6***	-6.0%
		S	Avg. mo. family income in month prior to 36-mo FU	$1,838	$86	4.7%	$86
B. Programs that focus on financial work incentives tied to hours of work
New Hope (a)	Poor families employed FT at RA	A	Avg. annual recipient E+W+FS +EITC for year 1 of 2-yr FU	$14,561	$187	1.3%	$16	58.5	-5.2	-8.9%
		A	Avg. annual recipient E+W+FS +EITC for year 2 of 2-yr FU	$15,294	-$1,148	-7.5%	-$96	56.2	-6.9	-12.3%
	Poor families not employed FT at RA	A	Avg. annual recipient E+W+FS +EITC for year 1 of 2-yr FU	$9,843	$1,347***	13.7%	$112	89.3	-5.6**	-6.3%
		A	Avg. annual recipient E+W+FS +EITC for year 2 of 2-yr FU	$9,915	$1,298***	13.1%	$108	81.4	-8.2***	-10.1%
SSP (b)	Single-parent recipients	A	Avg. mo. recipient E-W income in Q 5 and 6	$932	$179***	19.2%	$179
		A, S	Avg. mo. family income at 18-mo FU	$1,286	$199***	15.5%	$199	89.8	-12.2***	-13.6%
		A, S	Avg. mo. family income in 6 mos. prior to 36-mo. FU	$1,432	$153***	10.7%	$153	86.2	-9.4**	-10.9%
SSP Plus (b)	Single-parent recipients	A, S	Avg. mo. family income in 6 mos. prior to 18-mo FU	$1,171	$156***	13.3%	$156
SSP Applicants (b)	Single-parent applicants	A, S	Avg. mo. family income in 6 mos. prior to 30-mo FU	$1,686	$286***	17.0%	$286	68.5	-11.3***	-16.5%
C. Programs that focus on mandatory work-related activities
LA Jobs-1st GAIN	Single-parent recipients and applicants	A	Avg. annual recipient E+W+FS in year 2 of 2-yr FU	$9,920	$136	1.4%	$11
		A	Avg. annual recipient E+W+FS +EITC/payroll taxes in year 2 of 2-yr FU	$10,262	$206	2.0%	$17	75.6	-4.5	-6.0%
		A, S	Avg. mo. HH income + EITC/payroll taxes in last month of 2-yr FU	$1,001	$86*	8.6%	$86	67.6	-9.7***	-14.3%
Atlanta LFA	Recipients and applicants	A	Avg. annual recipient E+W+FS in year 2 of 2-yr FU	$7,549	$191	2.5%	$16	87.1	-1.6	-1.8%
		S	Avg. mo. Recipient income +EITC + CC in last month of 2-yr FU	$699	$24	3.4%	$24	80.1	-1.2	-1.5%
Grand Rapids LFA	Recipients and applicants	A	Avg. annual recipient E+W+FS in year 2 of 2-yr FU	$7,746	-$303**	-3.9%	-$25	86.5	-1.2	-1.4%
		S	Avg. mo. Recipient income +EITC + CC in last month of 2-yr FU	$833	-$42	-5.0%	-$42	68.6	-0.4	-0.6%
Riverside LFA	Recipients and applicants	A	Avg. annual recipient E+W+FS in year 2 of 2-yr FU	$7,874	-$358***	-4.5%	-$30	83.5	-1.0	-1.2%
		S	Avg. mo. Recipient income +EITC + CC in last month of 2-yr FU	$867	$19	2.2%	$19	72.6	-6.3***	-8.7%
Portland	Recipients and applicants; no cases with substantial barriers	A	Avg. annual recipient E+W+FS in year 2 of 2-yr FU	$8,110	$238	2.9%	$20	83.4	-4.0	-4.8%
		S	Avg. mo. Recipient income +EITC + CC in last month of 2-yr FU	$843	$59	7.0%	$59	70.8	-6.4	-9.0%
Atlanta HCD	Recipients and applicants	A	Avg. annual recipient E+W+FS in year 2 of 2-yr FU	$7,549	$235	3.1%	$20	87.1	-2.0*	-2.3%
		S	Avg. mo. Recipient income +EITC + CC in last month of 2-yr FU	$699	$26	3.7%	$26	80.1	-0.6	-0.7%
Grand Rapids HCD	Recipients and applicants	A	Avg. annual recipient E+W+FS in year 2 of 2-yr FU	$7,746	-$91	-1.2%	-$8	86.5	-0.3	-0.3%
		S	Avg. mo. Recipient income +EITC + CC in last month of 2-yr FU	$833	-$28	-3.4%	-$28	68.6	0.1	0.1%
Riverside HCD	Recipients and applicants	A	Avg. annual recipient E+W+FS in year 2 of 2-yr FU	$7,768	-$619***	-8.0%	-$52	86.4	-0.2	-0.2%
		S	Avg. mo. Recipient income +EITC + CC in last month of 2-yr FU	$859	-$10	-1.2%	-$10	76.8	-3.5	-4.6%
Columbus Integrated	Recipients and applicants	A	Avg. annual recipient E+W+FS in year 2 of 2-yr FU	$8,332	-$41	-0.5%	-$3	79.3	0.0	0.0%
		S	Avg. mo. Recipient income +EITC + CC in last month of 2-yr FU	$806	-$9	-1.1%	-$9	76.1	-3.3	-4.3%
Columbus Traditional	Recipients and applicants	A	Avg. annual recipient E+W+FS in year 2 of 2-yr FU	$8,332	$29	0.3%	$2	79.3	-0.3	-0.4%
		S	Avg. mo. Recipient income +EITC + CC in last month of 2-yr FU	$806	$17	2.1%	$17	76.1	-4.5	-5.9%
Detroit	Recipients and applicants	A	Avg. annual recipient E+W+FS in year 2 of 2-yr FU	$8,892	$101	1.1%	$8	84.1	-1.2	-1.4%
		S	Avg. mo. Recipient income +EITC + CC in last month of 2-yr FU	$766	$10	1.3%	$10	79.1	-3.7	-4.7%
Oklahoma City	Applicants	A	Avg. annual recipient E+W+FS in year 2 of 2-yr FU	$5,238	-$137	-2.6%	-$11	92.8	-0.5	-0.5%
		S	Avg. mo. Recipient income +EITC + CC in last month of 2-yr FU	$737	-$40	-5.4%	-$40	74	0.6	0.8%
IMPACT Basic Track	Recipients and applicants, less job ready	A	Avg. annual recipient E-W-FS income over year 2 of 2-yr FU	$5,645	$336	6.0%	$28	91.2	-3.0	-3.3%
D. Programs that focus on financial work incentives and mandatory work-related activities
WRP	Single-parent recipients and applicants	A	Avg. quarterly recipient E-W-FS income over 42-mo FU	$2,256	$41	1.8%	$14
		A	Avg. quarterly recipient E-W-FS income in last 3 mos. of 42-mo FU	$2,307	$25	1.1%	$8
		S	Avg. mo. HH income in mo. prior to 42-mo FU	$1,410	$10	0.7%	$10
		S	Avg. mo. HH income + EITC in mo. prior to 42-mo FU	$1,501	$27	1.8%	$27
MFIP	Urban single-parent recipients	A	Avg. quarterly recipient E-W-FS income in last 3 quarters of 10-quarter FU	$2,525	$296***	11.7%	$99	77.7	-12.4***	-16.0%
		A	Avg. quarterly recipient E-W-FS income + EITC/taxes in last 3 quarters of 10-quarter FU	$2,613	$382***	14.6%	$127	70.5	-12.1***	-17.2%
		S	Avg. mo. family income in month prior to 36-mo FU	$1,459	-$24	-1.6%	-$24
	Urban single-parent applicants	A	Avg. quarterly recipient E-W-FS income in last 3 quarters of 10-quarter FU	$2,578	$162***	6.3%	$54	66	-6.9***	-10.5%
		A	Avg. quarterly recipient E-W-FS income + EITC/taxes in last 3 quarters of 10-quarter FU	$2,556	$187***	7.3%	$62	59.7	-6.1***	-10.2%
		S	Avg. mo. family income in month prior to 36-mo FU	$1,838	$75	4.1%	$75
TSMF	Single parent recipients	A	Avg. annual family E+W+FS over 4-yr FU	$8,849	$118***	1.3%	10
	Single parent applicants	A	Avg. annual family E+W+FS over 1-yr FU	$8,558	$10	0.1%	1
		A	Avg. annual family E+W+FS over 2-yr FU	$8,414	-$163	-1.9%	-14
FIP	Recipients	A	Avg. quarterly recipient E-W income in Q4 of 2-yr FU	$1,721	$98***	5.7%	$33
		A	Avg. quarterly recipient E-W income in Q8 of 2-yr FU	$1,907	$37	1.9%	$12
	Applicants	A	Avg. quarterly recipient E-W income in Q4 of 2-yr FU	$2,004	$218***	10.9%	$73
E. Programs that focus on other individual reforms
F. Programs that focus on TANF-like bundle of reforms (time limits with financial incentives, work-related activities, or both)
EMPOWER (c)	Recipients	A,S	Avg. mo. HH income in survey month approx. at 30-mo FU	$1,339	$80	6.0%	$80
IMPACT Placement Track	Recipients and applicants, more job ready	A	Avg. annual recipient E-W-FS income over year 2 of 2-yr FU	$7,502	$77	1.0%	$6	80.2	-1.0	-1.2%
VIP/VIEW	Recipients	A	Avg. annual recipient E-W-FS in year 2 of 2-yr FU	$6,482	$128	2.0%	$11
ABC	Single parent recipients and applicants	S	Avg. mo. HH income in month before survey 12-18-mo FU	$778	$0	0.0%	$0
FTP	Recipients and applicants	A	Avg. total recipient E-W-FS income in year 2	$6,358	$351*	5.5%	$29
		A	Avg. total recipient E-W-FS income in year 3	$6,137	$496**	8.1%	$41
		A	Avg. total recipient E-W-FS income in year 4	$6,310	$253	4.0%	$21
		A	Avg. total recipient E-W-FS income in 2nd Q of year 5	$1,674	-$52	-3.1%	-$17
		S	Avg. mo. HH income in month before 4-yr FU	$1,379	$89	6.5%	$89
Jobs First	Recipients and applicants	A	Avg. annual recipient E-W-FS income in year 2	$10,037	$1,121***	11.2%	$93
		A	Avg. annual recipient E-W-FS income in year 3	$10,647	$172	1.6%	$14
		A	Avg. annual recipient E-W-FS income in year 4	$11,249	-$132	-1.2%	-$11
		A	Avg. annual recipient E-W-FS + EITC/taxes income in yr. 3 and 4	$10,828	$150	1.4%	$13
		S	Avg. total recipient income in month before 3-yr FU	$1,022	$74***	7.2%	$74
		S	Avg. total income of other HH members in month before 3-yr FU	$442	$12	2.7%	$12

NOTES: For full program names and citations, see Table 3.4. In calculating the poverty rates, the official poverty lines are applied to the given income measure. This will not necessarily correspond to the official poverty definition. Abbreviations: A=administrative data; S=survey data; E=earnings; W=cash welfare payments; FS=Food Stamp payments; EITC=Earned Income Tax Credit; CC=out-of-pocket child care expenses; FU=follow-up; HH=household; Q=quarter. * = statistically significant at the 10 percent level; ** = statistically significant at the 5 percent level; *** = statistically significant at the 1 percent level. (a) Poverty line based on earnings-related income only (earnings, EITC, earnings supplement). (b) Results in Canadian dollars. (c) Phoenix site only, cash assistance.

 

Table 8.2–Estimated Impact of Welfare Reform on Income Sources: Random Assignment Studies (Click on this link to view table)

Vermont’s WRP-IO program shows no gain for the treatment group over the control group in recipient combined income, measured using administrative data over the full four-year follow-up or in the last quarter only. However, a large and significant treatment-control difference in household income is measured at the 42-month follow-up survey, equal to $139—$145 per month depending on the income measure. Panel A of Table 8.2 reveals that WRP-IO did not result in a significant difference in welfare or food stamp payments, and the program had no significant impact on earnings at the 42-month follow-up (Table 5.1). The only income source that shows a sizeable increase is reported earnings from other family members, but the treatment-control difference for this income component was not significant (not shown). Likewise, while the fraction of treatment group members with more than one-half of household income from earnings was higher than the control group, the difference was not significant (Table 8.2).

MFIP-IO shows some favorable impacts for both income and poverty. The MFIP-IO results are strongest for long-term recipients, with an estimated statistically significant increase in recipient combined income of $80—$100 a month (depending on whether measures are pre- or post-tax, including the EITC) and a reduction in the poverty rate of 7—8 percentage points. However, there is no gain in family income for the recipient group measured at the 36-month follow-up survey, a result that may be due to differential reporting bias between the treatment and control groups.68  Smaller effects were found for the income gains and poverty rate reductions in the MFIP applicant group, with the exception of family income at the 36-month follow-up where the impact was larger (i.e., a positive impact compared with the negative impact estimate for recipients). To the extent that MFIP-IO generated income gains, they were the result of higher welfare and (cashed-out) food stamp payments generated by the financial work incentives: MFIP-IO raised welfare and food stamp payments for both applicants and recipients by about the same dollar amount ($91—$97 per month, as shown in Table 8.2). There were no significant impacts on earnings for either MFIP-IO recipients or applicants (Table 5.1). Consequently, the earnings share measure declined, indicating a lower level of self-sufficiency as a result of MFIP-IO, an effect that was largest and significant for the MFIP-IO applicants (Table 8.2).

8.2.2. Programs That Focus on Financial Work Incentives Tied to Hours of Work

The two programs that evaluate financial work incentives tied to hours of work, with follow-up periods that range from 18 months to three years, both show positive and significant income impacts and corresponding significant negative poverty impacts. (See Panel B of Table 8.2.) In the case of New Hope, the effects are evident only for families not employed full-time at randomization. In the case of the Canadian SSP evaluations, the program with the most generous financial work incentives to reward full-time work, longer-term recipients (evaluated in two provinces), and new applicants (evaluated in one province) show large percentage gains in recipient combined income or family income–from 11 to 19 percent–for up to 36 months post-randomization.69  The SSP Plus results fall in between the results for the main SSP. The poverty rate, when reported, also falls by a substantial magnitude, between 9 and 12 percentage points.

New Hope and SSP, by operating outside the traditional U.S. and Canadian welfare systems, were designed to replace welfare payments with an earnings supplement. Both programs show significant decreases in welfare payments and increases in the earnings supplement. Food stamp payments, relevant only for New Hope, showed a significant decrease for families employed full time at random assignment, while the reverse is true for those not employed full time. Recall that the earnings impact for New Hope participants employed full time at random assignment were insignificant and negative. By year two, the positive impact on the earnings supplement just exceeded the negative impact on AFDC payments. These effects for earnings and welfare/earnings supplement combined to produce a negative impact overall on recipient combined income for this group. In contrast, those in New Hope not employed full-time at random assignment experienced significant positive earnings gains, and the average size of the earnings supplement more than offset the impact on AFDC benefits. Food stamp benefit payments also increased by the second year of follow-up. Consequently, this group experienced large and significant income gains overall for both years one and two. Finally, SSP produced increases in both cash transfer income (earnings supplements net of lost income assistance payments) and earnings. This combination of impacts on the sources of income produced the large increase in recipient combined income and family income.

8.2.3. Programs That Focus on Mandatory Work-Related Activities

All the programs that focus on mandatory work-related activities report measures of income and poverty at the end of two years, for both administrative and survey data in all but one case, where only administrative data is available (results summarized in Panel C of Table 8.1). Measures of combined recipient income based on administrative data show a significant negative impact for three of the NEWWS programs, ranging from —$25 to —$52 per month. The other evaluations all show smaller insignificant impacts, either negative (three NEWWS programs) or positive (five NEWWS programs, L.A. Jobs-First GAIN, and IMPACT). With the exception of Riverside LFA, the survey-based measures of recipient combined income (in some cases, accounting for the EITC, payroll taxes, and child care expenses) for the last month of the follow-up period all have the same sign as the measures based on administrative data and are similar in magnitude. The only statistically significant survey-based impact estimate is for L.A. Jobs-First GAIN, a 9 percent increase in income.

With a few exceptions, most of the poverty impacts are insignificant and small in magnitude.70  They are almost all negative, including the three significant impacts, suggesting these programs are somewhat more effective at raising incomes near the poverty threshold than at the bottom of the income scale. At the same time, several of the NEWWS programs resulted in a slight increase in the fraction with incomes below 50 percent of the poverty line as of the two-year follow-up, suggesting that those near the bottom of the income scale may be worse off (Freedman et al., 2000a).

Newly available data for the 11 NEWWS programs provides information on combined income (income from earnings, welfare, and food stamps plus the EITC, less payroll taxes based on administrative data) five years after the program began. The impact estimates are plotted in Figure 8.2 for follow-up years one to five. By year five, there is only one significant income impact in any of the sites (a negative impact in Riverside HCD; data are not available for Oklahoma). The year five impact estimates are almost equally divided between negative impacts, impacts close to zero, and positive impacts.

Figure 8.2–Impact Estimates for Combined Annual Income in 11 NEWWS Programs, Years 1 to 5

[D]

Based on Figure 8.2, there is no clear pattern in NEWWS site impacts associated with the program orientation or service delivery approach. Portland’s employment-focused, mixed—first activity model has the largest five-year income and two-year poverty impacts, but it excluded some recipients with substantial barriers to work. Thus, it is not clear if similar impacts would be found for a more disadvantaged population. The pattern over time for the various NEWWS programs is also not consistent, although the majority of the sites exhibit a fade-out effect over time. For example, Portland has increasing positive impacts in years two and three followed by the largest positive impact of all the sites in year 4 (just over $600 in annual combined income; p < 0.10). But the impact shrinks in year five. At the other extreme, Riverside HCD has negative and significant impacts exceeding $400 in all five years, although the year five impact is somewhat less negative than in year four. Some of the fade-out may be due to the control-group crossover that took place in years four and five, but this crossover did not apply to the Portland or Riverside sites and was probably not a significant factor in the other sites either (Hamilton et al., 2001).

Whether measured at the two-year or five-year follow-up, the modest, if any, income and poverty effects are consistent with the combination of the positive earnings gains produced by these programs (Table 5.1 and Figure 5.2) and high benefit reduction rates under the old AFDC rules that led to a significant reduction in welfare payments as earnings rose (Panel C of Table 8.2). Likewise, food stamp payment declines were also almost always significant. Since income, by and large, did not change, but the composition shifted from welfare benefits to earnings, it is not surprising that the majority of the programs also raised self-sufficiency as measured by the earnings share, available only as of the two-year follow-up (Table 8.2). Even so, the average share of income from earnings for treatment group members never exceeds 50 percent.

8.2.4. Programs That Focus on Financial Work Incentives and Mandatory Work-Related Activities

Four programs combine financial work incentives and work requirements with results for income and poverty rates for follow-up periods that range from two to four years. (See Panel D of Table 8.1.) With the exception of WRP, the combined programs produce statistically significant income gains and, when measured, poverty reductions for at least part of the population served over both short and longer horizons. Recipients benefit more than applicants in the MFIP and TSMF studies, while the reverse is true for Iowa’s FIP. The significant impact for FIP recipients as of the fourth quarter fades by the eighth quarter, from $33 per month (significant) to $12 per month (insignificant).71  In the case of MFIP, a more comprehensive survey-based measure of family income shows a negative impact for recipients, compared with the sizeable gain in combined income measured with administrative data, and a somewhat larger, but insignificant, impact for applicants. As discussed above, the difference in estimated impacts for recipients using administrative data compared with survey data is likely a result of differential reporting biases for treatment versus control group members (Miller et al., 2000). The MFIP impacts based on administrative data are similar in magnitude and are not statistically different from those for MFIP-IO.

The changes in income sources that produce the overall income impacts differ to some extent across these programs. In WRP, which had among the least generous financial work incentives, there were significant declines in welfare payments, and food stamp payments showed no change (Panel D of Table 8.2). The significant earnings increase essentially offset the transfer payment decline, leading to a significant increase in the share of income from earnings, but no change in income. TSMF and FIP also lowered welfare and food stamp payments, but somewhat larger earnings increases resulted in modest income gains overall. (The impact on the earnings share was not measured for these two programs.) In contrast, MFIP raised earnings but also increased combined welfare and food stamp payments, so that the gains in combined income were even larger than programs that decreased welfare income. Consequently, there is no change in MFIP in self-sufficiency as measured by the earnings share.

8.2.5. Programs That Focus on TANF-Like Bundles of Reforms

As shown in Panel F of Table 8.1, four of the six programs that include TANF-like bundles of reforms–EMPOWER, IMPACT, VIP/VIEW, and ABC–find no significant impacts on income or poverty (which is only measured in one study), with follow-up periods that range from 12 to 30 months. Of these four programs, IMPACT had the largest positive (and significant) impact on earnings, but it also resulted in a significant reduction in welfare and food stamp payments (Panel F of Table 8.2). It also resulted in increased self-sufficiency as measured by the earnings share. As noted in Table 3.5 (see notes b, c, and f), a sizeable fraction of the control group in three of these four studies believed that the time limits applied to them, which may bias the estimated program impacts toward zero.

FTP and Jobs First–programs for which the time-limit feature was better understood– demonstrate positive income results earlier in the follow-up period that tend to disappear as time limits are reached. For FTP, the positive and significant impact on recipient combined income peaks in year three–the year recipients first begin to reach the time limit. It becomes insignificant but still positive in year four, as more recipients reach the time limit, and turns negative and insignificant in the first part of year five. Jobs First showed an even larger positive income impact for recipient combined income as late as year two, but this positive impact dissipated by year three, and turned negative by year four.72  More comprehensive measures of income based on administrative or survey data show the same pattern for both programs. The pre— and post—time limit impacts for FTP and Jobs First suggest that reaching the time limit is associated with an income decline, as welfare benefits are exhausted and recipients must rely more on their own earnings.

Both programs had a significant positive impact on earnings (Table 5.1) and a significant reduction in welfare payments, at least by the final follow-up period, three to five years past randomization (Table 8.2). FTP also measured a significant increase in child support payments (not shown). Only FTP analyzed the earnings share, and the results show a significant increase in self-sufficiency as measured by recipient earnings as a fraction of recipient combined income (earnings, welfare, and food stamps) and as a fraction of household income by year four. This impact is no longer significant, however, by the second quarter of year five. The structure of the benefit levels and income disregards in the Connecticut program made it more generous than FTP in Florida, which helps explain the larger initial increase in income and welfare and food stamp use in Jobs First (as of the two-year follow-up).

8.2.6. Subgroup Differences

As summarized in Appendix A, the experimental studies demonstrate both similarities and differences in the impacts of reform policies on income and income sources between subgroups, defined by various measures of disadvantage. Because the studies often analyze different subgroups, it is difficult to draw broad inferences of the impacts of individual reform policies or welfare reform more generally for subgroups with specific characteristics.

Only one of the evaluations that focus exclusively on financial work incentives considers any subgroups, and it is hard to generalize from the specific patterns from that study. The two programs that focus on financial work incentives tied to hours of work are not consistent, with some impacts larger for the more disadvantaged while the reverse is true for other impacts. A larger number of subgroup analyses for programs that focus on mandatory work requirements, including the pooled analysis of the NEWWS programs (combined with nine others), suggests that these programs can have differential impacts on subgroups. The pooled NEWWS analyses indicate that income gains are largest and that welfare payment declines are smallest for the least disadvantaged. The L.A. Jobs-First GAIN results differ from this pooled finding in that the program impacts did not vary by subgroups. The subgroup patterns for programs that combine financial work incentives and work requirements or that focus on TANF-like bundles of reform are mixed, with examples of larger impacts for both the least and the most disadvantaged.

8.3. ECONOMETRIC STUDIES OF THE EFFECTS OF WELFARE REFORM ON INCOME AND POVERTY

Compared with the large number of econometric studies that have examined welfare caseloads (reviewed in Chapter 4), only a handful of econometric studies have examined family income, and even fewer have examined poverty. (See Table 8.3.) All the studies in this literature rely on the CPS, the primary nationally representative data source with information on annual individual and family income. The Annual Demographic Supplement to the March CPS is the source used by the Census Bureau to calculate poverty rates on an annual basis; hence, these studies can implement a measure of income (annual family pretax cash income) and poverty that follows the concepts employed by the Census Bureau in its calculations.

Like the caseload studies that rely on survey data, these analyses use aggregated microdata or individual-level microdata to model the level or log of income and the poverty rate as a function of the existence (approval or implementation) of a waiver or TANF as a bundle. In one case, the effect of a specific TANF policy–time limits–is also considered. The welfare policy variables generally follow the "modified dummy variable" approach used in the caseload literature. The study population either includes all women or female-headed families in a given age range (typically 16—54). In one case, the sample is children under 16. The models typically include controls for the business cycle (e.g., current and lagged unemployment rate), demographic characteristics of the recipient (e.g., age, education, and race/ethnicity), other policy variables (e.g., maximum welfare benefit level, minimum wage, and EITC), and state and year fixed effects. Some models also include state-specific time trends.

In the remainder of this subsection, we first discuss the findings from these studies for a specific welfare policy, time limits; we then discuss the findings for waivers or TANF as a bundle. Results for income and poverty rates are discussed in turn.

8.3.1. Effects of Specific Reforms: Time Limits

Grogger’s (forthcoming) study is the only econometric analysis of income to consider the effect of a specific TANF reform, in this case, time limits. In addition to estimating the impact of reform as a bundle (discussed below), Grogger’s model of income is estimated with a dummy for the implementation of a time limit and an age interaction.73  The point estimates suggest that time limits lower incomes by 3—6 percent (linear and log model, respectively) when the age of the youngest child is 13 or above (Panel A of Table 8.3). For each year below age 13, time limits further reduce income or leave it unchanged. However, none of these estimates is statistically significant. Given that Grogger finds significant negative effects of time limits on welfare use and only modest positive effects on earnings, family income might be expected to decline in states with time limits in effect. The insignificant effects may mean that other sources of income increase enough to offset the welfare declines. The estimated impact is also limited to the period before most recipients began reaching the time limit. Thus, the impact may change as more recipients exhaust their benefits.

Table 8.3–Estimated Impact of Welfare Reform on Income and Poverty: Econometric Studies

										Other controls
Study	Data	Sample population	Begin	End	Outcome	Dep. var.	Policy var.	Coeff. (s.e.)	% effect	Economy	Demogr. and Geogr.	Fixed Effects	Policy
A. Income
Moffitt (1999)	CPS aggregated	women 16-54	77	95	Annual pre-tax family cash income	Level	Any waiver	393 (474)	1.3	U, U-1		S, Y, State time trends	B
	CPS aggregated	women 16-54, educ<12	77	95	Annual pre-tax family cash income	Level	Any waiver	-240 (909)	-0.8 (a)	U, U-1	A, E	S, Y, State time trends	B
		women 16-54, educ=12					Any waiver	569 (909)	1.9 (a)
		women 16-54, educ=13-15					Any waiver	870 (909)	2.9 (a)
		women 16-54, educ>16					Any waiver	-898 (909)	-3.0 (a)
Schoeni and Blank (2000)	CPS aggregated	women 16-54, educ<12	76	98	Annual pre-tax family cash income	Log	Any waiver	0.061 (0.013)	6.1	U, U-1, EG, each *E	A, E, A*E, R	S, Y, state time trends, Y*E	B, B*E
		women 16-54, educ=12					Any waiver	-0.006 (0.011)	-0.6
		women 16-54, educ>12					Any waiver	-0.011 (0.009)	-1.1
		women 16-54, educ<12					TANF	0.031 (0.031)	3.1
		women 16-54, educ=12					TANF	0.022 (0.027)	2.2
		women 16-54, educ>12					TANF	-0.011 (0.021)	-1.1
Bitler, Gelbach and Hoynes (2001)	CPS micro data	women 16-54	84	98	Annual pre-tax family cash income,	Level	Any waiver	-78 (480)	-0.2	U, U-1, EG	R, MSA, CC	S, Y	B
					CPS families		TANF and ever had waiver	-781 (1,412)	-1.5
							TANF and never had waiver	-602 (1,304)	-1.2
		women 19-54, educ < 12			Annual pre-tax family cash income,		Any waiver	1,466 (536)	5.5
					CPS families		TANF and ever had waiver	-70 (1,090)	-0.3
							TANF and never had waiver	-1,431 (974)	-5.4
		children under age 16			Annual pre-tax family cash income,		Any waiver	-331 (647)	-0.7
					CPS families		TANF and ever had waiver	-313 (1,854)	-0.7
							TANF and never had waiver	-833 (1,746)	-1.8
Grogger (forthcoming)	CPS micro data	female headed families 16-54	78	99	Annual pre-tax family cash income	Level	Any reform (waiver or TANF)	702 (570)	3.6	U	A, E, R Young child A, # kids	S, Y, State time trends	B, MW, EITC
							Any reform * Age of youngest child	-55 (81)	-0.3
							Modified dummy>0 if time limit in place	-480 (986)	-2.5
							Modified dummy>0 if time limit in place * (age of youngest child - 13)	31 (85)	0.2
			78	99	Annual pre-tax family cash income	Log	Any reform (waiver or TANF)	0.098 (0.028)	9.8	U	A, E, R Young child A, # kids	S, Y, State time trends	B, MW, EITC
							Any reform * Age of youngest child	-0.007 (0.003)	-0.7
							Modified dummy>0 if time limit in place	-0.056 (0.040)	-5.6
							Modified dummy>0 if time limit in place * (age of youngest child - 13)	0.000 (0.003)	0.0
B. Poverty
Schoeni and Blank (2000)	CPS aggregated	women 16-54, educ<12	76	98	Poverty rate (annual pre-tax family cash inc.)	Level	Any waiver	-0.024 (0.006)	-8.2	U, U-1, EG, each *E	A, E, A*E, R	S, Y, state time trends, Y*E	B, B*E
		women 16-54, educ=12					Any waiver	0.001 (0.005)	1.0
		women 16-54, educ>12					Any waiver	0.001 (0.004)	1.6
		women 16-54, educ<12					TANF	-0.022 (0.013)	-7.8
		women 16-54, educ=12					TANF	-0.011 (0.012)	-8.4
		women 16-54, educ>12					TANF	0.003 (0.009)	4.6
Bitler, Gelbach and Hoynes (2001)	CPS micro data	women 16-54	84	98	Annual pre-tax family cash inc. is below poverty	Level	Any waiver	-0.006 (0.003)	-4.2	U, U-1, EG	R, MSA, CC	S, Y	B
					line (CPS families)		TANF and ever had waiver	-0.006 (0.007)	-4.0
							TANF and never had waiver	-0.004 (0.007)	-2.6
		women 19-54, educ < 12			Annual pre-tax family cash inc. is below poverty		Any waiver	-0.031 (0.010)	-8.3
					line (CPS families)		TANF and ever had waiver	-0.046 (0.022)	-12.3
							TANF and never had waiver	0.004 (0.021)	1.1
		children under age 16			Annual pre-tax family cash inc. is below poverty		Any waiver	-0.008 (0.005)	-3.5
					line (CPS families)		TANF and ever had waiver	-0.018 (0.009)	-8.3
							TANF and never had waiver	-0.013 (0.008)	-5.9

NOTES: Abbreviations: U=unemployment rate; U-1=lagged unemployment rate; EG=employment growth; A=age, E=education, R=race, MSA=Metropolitan Statistical Area (urban), CC=Central city; B=maximum welfare benefit, MW=minimum wage, EITC=Earned Income Tax Credit; S=state; Y=year. (a) Percentage effects calculated using mean for entire sample.

8.3.2. Effects of Reform as a Bundle

Income

All four studies summarized in Panel A of Table 8.3 estimate models of the determinants of annual pretax family cash income, either in a linear or log model. Moffitt (1999) covers the shortest interval of time and finds no statistically significant effects of waivers (as approved) on income for women age 16—54. Schoeni and Blank (2000), with four additional years in their time series, find that the existence of a waiver (as implemented) raises family income by 6 percent for women with fewer than 12 years of schooling (a statistically significant effect), while the effect for TANF (as implemented) is 3 percent for the same group but statistically insignificant. There is no effect of waivers or TANF on the incomes of women with higher levels of education, which would be expected given that this group is less likely to be affected by welfare reform.

Like Moffitt (1999), Bitler, Gelbach, and Hoynes (2001) (hereafter BGH) find no effect of a major state waiver (or TANF) as implemented on income for all women. There is also no significant impact on income for the sample of children in the CPS. Like Schoeni and Blank, BGH do find a significant positive effect of any waiver on income for women with less than a high school education, but no effect of TANF (where the estimated impact of TANF is differentiated by whether the state previously had implemented a waiver). The estimated waiver impact on income for less educated women based on their model estimated in levels–a nearly 6 percent increase–is comparable to Schoeni and Blank’s estimate based on a log-linear model. BGH also estimate alternative models using different living arrangement concepts to measure income. In addition to the results based on the traditional CPS concept of family reported in Table 8.3, they also consider income for the household, personal income (not relevant for the child sample), and family income, where income from related subfamilies is either pooled or not pooled with the primary family’s income. The results show some sensitivity to the assumption about income pooling, with more pooling in the sample of less educated women leading to bigger impacts of waivers or TANF in waiver states. This result follows from BGH’s estimated impacts of welfare reform on the number of adults in the household and families "doubling up," discussed in Chapter 7.

Grogger (forthcoming) combines waivers and TANF into one measure of any reform and finds that family income among female family heads rises by 4—10 percent for women whose youngest child is under age one, depending on whether the model is estimated in levels or logs (although the coefficient is only statistically significant at conventional levels in the log model).74  The interaction between the reform dummy and age of the youngest child suggests that the impact on family income declines with the youngest child’s age. (Again, this effect is significant only in the log model.) The larger effect Grogger finds may result from the fact that he limits his sample to female-headed families. This population is likely to be most affected by welfare reform, hence, the larger estimated impact.

All these models suffer from the collinearity problem discussed in earlier chapters. There simply is too little variation left to precisely estimate a TANF effect once state and year fixed effects are included in these models. Schoeni and Blank (2000) address this problem using changes in estimated year effects (in regression models with other controls) between 1995 and 1998 for less educated versus more educated women. This strategy does not lead to any more precisely estimated effects of TANF on the income of low-skill women.

Schoeni and Blank (2000) also consider the distributional effects of waivers and TANF by estimating equations of the log of the 20th and 50th percentiles of family income for women 16 to 54, and the 20th/50th ratio (not shown). Waivers are estimated to raise family incomes at the 20th and 50th percentiles by about 8—10 percentage points (a statistically significant effect) for women who drop out of high school. TANF, however, has a statistically significant and positive effect only on the 50th percentile of family income for women in the lowest education category. The lack of an effect at the 20th percentile leads to an estimated widening of the 20th—50th gap because of the implementation of TANF.

Poverty

Only Schoeni and Blank (2000) and BGH (2001) model the poverty rate, with results shown in Panel B of Table 8.3. Consistent with their findings for income, Schoeni and Blank report a statistically significant decline in the poverty rate for women with fewer than 12 years of schooling, equal to about 8 percent associated with the implementation of waivers and 8 percent associated with the implementation of TANF. They find a similar size effect for TANF using their alternative residual change methodology. There is no evidence to suggest that waivers or TANF affected the poverty rate for women with 12 or more years of schooling.

BGH likewise find that the implementation of waivers significantly reduced poverty rates for less educated women by 8 percent, but there was no significant reduction for all women. The implementation of TANF on top of waivers further reduces poverty for the sample of less educated women, but the effect is zero for TANF implemented in states with no prior waivers and for TANF as a whole for all women. For the sample of children, only TANF implemented on top of waivers results in a statistically significant reduction in the poverty rate. Again, there is some sensitivity of the magnitude and significance of the impacts, depending on whether income is measured at the household or family level and depending on the concept of the family.

8.4. EVALUATING THE EFFECTS OF WELFARE REFORM ON INCOME, INCOME SOURCES, AND POVERTY

The econometric studies and demonstration studies reviewed thus far provide a varied picture of the impact on both income and poverty of specific welfare reform policies and groups of policies implemented simultaneously. What lessons can we learn from these varied studies, and can the different results be reconciled? To what extent can we expect welfare reform as a whole and the specific policy and program components to affect the incomes of recipients and the incomes of their families, as well as the poverty rate?

8.4.1. Effects of Specific Reforms

Of the studies reviewed in this chapter, the experimental evaluations are the most informative about the effects of specific reforms, but then only for a limited group of policies, principally financial work incentives, work requirements, and time limits. One econometric study also considered the impact of time limits on income only. However, these studies are not informative about the impact of other policy and program components of welfare reform, including sanctions.

Financial Work Incentives

Based on the results of eight high-quality experimental studies, we conclude that generous financial work incentives alone or in combination with work requirements, or those that are tied to hours of work, have the effect of raising incomes, with a corresponding reduction in poverty, an effect that is sustained up to three or four years (the length of the longest follow-up period). This effect is particularly strong in a program like MFIP (about $100—$125 per month in recipient combined income) with or without mandated work-related activities or a program like SSP (up to $200—$300 Canadian per month in family income or roughly $150—$200 U.S.) where income supports were linked to a requirement of full-time work (30 hours per week). Although these programs also produced significant reductions in poverty, the poverty rate for the treatment group that benefits from the higher income still exceeds 50 percent or more.

Part of the income gains result from increased benefit payments, given more generous disregards and benefit reduction rates (or the earnings supplement), and part of the income gains may stem from earnings gains, especially in programs that promote greater work effort through the incentive structure and work mandates. At the same time, the increase in welfare payments may mean that self-sufficiency as measured by the share of income from earnings will not necessarily improve and may even move in the direction of greater dependency. When incentives are weak, however, as in the case of the Vermont WRP program, there may be no income gains or poverty declines.

Mandatory Work-Related Activities

For work requirements alone–at least as implemented in the NEWWS evaluation, L.A. Jobs-First GAIN, and Indiana IMPACT–the results for income and poverty are mixed. Most programs show no statistically significant or economically meaningful effects on income or poverty, particularly compared to the large gains observed for programs that include generous financial work incentives. In the case of the five-year NEWWS follow-up, the impact estimates for recipient combined income are evenly divided in sign, and only one negative impact estimate is statistically significant. The two-year impact estimate for a broader measure of household income in L.A. Jobs-First GAIN is the only statistically significant positive impact. Impact estimates for poverty, which are only available for the 13 studies for the two-year follow-up, are more consistently negative, but at most two estimates are statistically significant depending on the income measure used. Thus, there is some evidence that these programs may modestly reduce poverty by raising incomes for those just below the poverty line. Evidence from NEWWS that deep poverty may increase in some programs is a counter to this more favorable assessment. These results on income, poverty, and self-sufficiency should not be too surprising, given that the earnings gains from these mandatory work programs are accompanied by reductions in welfare payments. As a result, these programs tend to raise self-sufficiency as measured by the share of income from earnings.

The programs that combine work requirements and financial work incentives produce gains in both earnings and cash assistance, thereby contributing to more favorable effects for income and poverty in contrast to the programs with work requirements only. This suggests that it is the financial work incentives that can be credited with this result. Two of the programs–MFIP and WRP–with their two treatment contrasts allow a more direct test of the role of the incentives component versus the work requirement component. For these two programs, when the impacts on income and poverty based on incentive-only programs shown in Panel A of Table 8.1 are compared with the alternative programs that combine incentives and work requirements shown in Panel D, the major share of the strong income gains and poverty reductions for MFIP and WRP can be attributed to the incentive component of the program. For example, the gain in recipient combined income for long-term recipients is $81 per month in the MFIP incentives-only program and $99 when incentives are combined with work requirements. In the case of WRP, the broader measure of household income is statistically significant and positive only for the incentives-only program, unlike MFIP where the combination of financial work incentives and work requirements produces favorable effects as well.

Time Limits

Programs that only implement time limits have yet to be experimentally evaluated. The econometric evidence provided in Grogger (forthcoming) suggests that time limits serve to reduce incomes. However, given the time period covered by the study, these estimates pertain to the impact of time limits before they are binding for most recipients. Moreover, none of Grogger’s point estimates are statistically significant.

FTP and Jobs First, with follow-up periods that include the period prior to time limits being reached and after time limits are reached, provide some insights into the mechanical effects of time limits. They both suggest that the favorable income gains observed in the pre—time limit period fade and are then reversed in the post—time limit period. We are not able to assign a significance level to these estimated negative impacts. These inferences, while not derived from the experimental design of these two studies, are nevertheless suggestive that time limits serve to reduce income as recipients begin to exhaust their benefits. The impact of FTP and Jobs First on poverty is not reported and therefore cannot be assessed with these studies.

8.4.2. Effects of Reform as a Bundle

The econometric studies reviewed in Section 8.3 suggest that welfare reform as a whole, and specifically TANF, has resulted in an increase in family pretax cash income for low-skilled women and single women with children, where the estimated impact is largest for the latter group. Focusing on the point estimates, Schoeni and Blank’s (2000) results suggest that family incomes for women with less than a high-school education increased by about 3 percent in the post-TANF period, and by a smaller amount (2 percent) for women who completed high school. Bitler, Gelbach, and Hoynes (2001) find a nearly 6 percent increase in income for women with less than a high school degree from implementing a waiver (and no significant effect for TANF implementation). Grogger’s (forthcoming) estimate translates into an 8 percent increase in income for a female-headed family whose youngest child is age 3 as a result of implementing a waiver or TANF. The effect falls as the age of the youngest child increases, reaching an impact of 5 percent for single mothers whose youngest child is age 7 and zero when the youngest child is age 14.

Grogger’s larger estimate is consistent with the fact that more of the sample of single-headed families are at risk of welfare compared with the sample of all women. To the extent that women without children and married women are mostly unaffected by welfare reform, Schoeni and Blank’s estimates will be biased downward from the true impact on women at risk of welfare receipt. Then again, by selecting on marital status and the presence of children, Grogger’s approach may be biased to the extent that welfare reform changes the composition of the at-risk population. Grogger argues that this bias will be negative, so that his estimates may be, if anything, too low. It must be stressed, however, that all of the econometric studies essentially measure the impact of reform as a bundle prior to the period when time limits become binding. Thus, they must be viewed as "pre—time limit" impacts.

Assuming these studies are correct in placing a bound on the impact of welfare reform as a bundle on family income from 3—9 percent in the pre—time limit period, how does this inference compare with the findings from the demonstration studies? Several methodological differences between the econometric analyses and the demonstration studies should be reiterated. First, the income concepts may differ, with less comprehensive income concepts often used in the demonstration studies compared with the CPS-based studies. However, in most demonstration studies, the most significant drivers of change in family income are changes in the recipient’s earnings and cash transfers; there are few instances where earnings from other household members or income from other sources, as reported in surveys, differ between the treatment and control groups during the follow-up period. Thus, the absence of data on other sources of income for the recipient or other family members will not necessarily have a large negative bias on the income impacts. Second, the demonstration studies, as noted in Chapter 3, cannot be used to estimate the full effects of TANF, because they do not capture program entry effects, i.e., they do not capture the impacts for potential welfare recipients. They also do not represent the weighted combination of reforms implemented under TANF. Even if they did, a demonstration program may not reproduce the effects of a large-scale program.

With these caveats in mind, we cautiously compare findings across the two types of studies. Since the last year of data analyzed in any of the econometric studies is 1999, only a small share of sample members in the CPS are likely to have had their benefits cut off because of time limits. Thus, the econometric results might be usefully compared with the impact estimates from the pre—time limit impacts of the demonstration studies that evaluated time limits combined with other policies (e.g., work requirements and financial work incentives). Panel F of Table 8.1 provides examples of such programs and their early impacts, which range from small (0—2 percent for VIP/VIEW and ABC up to two years post-randomization) to moderate (6—11 percent for recipient combined income in FTP and Jobs First). Since few states have implemented financial work incentives as generous as Connecticut’s, the impacts on the high end are not likely to be observed for TANF as implemented. Thus, the impacts estimated using the CPS data seem plausible given the range of the small and moderate impacts in the experimental studies with more modest financial work incentives.

One caution, however, is that the positive impact of TANF as a bundle on incomes of either recipients or the at-risk population may disappear once the TANF time limits have had an opportunity to produce both behavioral and mechanical effects. The demonstration studies of programs that include time limits suggest that the initial income gains are not sustained once time limits become effective and income from welfare payments goes to zero. At the same time, it is possible that the earnings gains may increase with time off welfare as former recipients obtain additional labor market experience. Such longer-term earnings gains may be sufficient to offset the welfare benefit losses. Additional years of data for the CPS can be used to explore which effect may dominate for the larger population at risk of welfare.

This same analysis applies to the estimated impact of welfare reform on the poverty rate. Schoeni and Blank’s econometric estimates suggest a 2 percentage-point reduction in the poverty rate for women with less than a high school education, which translates into an 8 percent reduction in the poverty rate. Bitler, Gelbach, and Hoynes (2001) find a similar impact estimate for the poverty rate for women who drop out of high school based on implementing waivers and an additional reduction in the poverty rate when TANF is implemented in states that had a major waiver. Unfortunately, none of the demonstration studies in Panel F of Table 8.1 reported effects for poverty. Judging from the income effects, we would certainly expect the demonstration studies with time limits to generate reductions in the poverty rate in the short run corresponding to the income gains. Likewise, if incomes fall after time limits become binding, the poverty rate may be expected to rise. Thus, while the estimated effect from the econometric studies is plausible in light of the experimental evidence, the antipoverty effect of TANF as a whole may be short-lived.

8.5. CONCLUSIONS

The studies reviewed in this chapter suggest that some welfare reform components can raise incomes and reduce poverty, although this result is not associated with all policy components, and there is reason to believe that some of the initially favorable effects will not persist over a longer horizon. The econometric and experimental evidence suggests that welfare reform as a bundle has raised incomes for disadvantaged women and lowered their poverty rate, at least in the short run before time limits have become binding. As time limits begin to affect a larger share of the recipient population, this outcome may not be sustained with the decline in welfare income that accompanies reaching the time limit. In addition, these favorable effects in the short run may mask important distributional changes. There is some limited evidence from econometric and experimental studies that reductions in poverty may be accompanied by an increase in the rate of deep poverty.

Generous financial work incentives–high earned income disregards and low benefit reduction rates inside the welfare system or earnings supplements outside the welfare system–generate the strongest income gains and antipoverty effects, especially when the incentive structure encourages full-time work. However, these programs can lower self-sufficiency because the increased welfare payments or earnings supplements may exceed the increased earnings. Work requirements alone have relatively weak effects on family income and poverty, but they do raise self-sufficiency by increasing the fraction of income from earnings. Finally, time limits, once they become binding, may erase income gains made possible by generous financial work incentives associated with working while on welfare.

While the antipoverty effectiveness of policy components such as financial work incentives appears to be quite robust, the income levels are relatively low and consequently the rates of poverty are relatively high among those who benefit from all the welfare reforms considered in this chapter. For example, in the experimental studies that evaluate financial work incentives (Panels A, B, and D of Table 8.1), the poverty rate for the treatment group falls below 50 percent just one time for the study populations and time periods measured in the table. The fraction with very low incomes may not move much at all, and many of those raised above the poverty line still remain "near poor." Changes in poverty status over time as a result of welfare reform remain relatively unexplored.

It is important to keep in mind that much of this chapter has focused on income measures that are rarely as comprehensive as would be desired to evaluate changes in well-being. Most econometric studies consider only family income before taxes and exclusive of in-kind benefits. Many of the experimental evaluations likewise use a concept of income limited to earnings and social welfare benefits (e.g., welfare and food stamp benefits in the U.S. experiments) for the recipient. Even if a comprehensive income measure were available, it may not fully reflect the individual’s or family’s command over resources. For that reason, in the next chapter, we focus on results for broader measures of well-being, which may provide a better gauge of living standards than what can be gleaned from examining income alone.

---

62Haskins (2001) also documents similar trends for single mothers with children.(back)

63Of the USDHHS-funded leaver studies, only those conducted in five states (Arizona, Illinois, Iowa, Missouri, and Washington), the District of Columbia, and two counties (Cuyahoga, Ohio, and San Mateo, California) collected information on income. Poverty rates were constructed only for Iowa, Missouri, Washington State, and Cuyahoga County.(back)

64Similar evidence is obtained from other recent reviews of leaver studies (see, e.g., Loprest, 1999; GAO, 1999c; and Cancian et al., 1999a).(back)

65Some would also argue that income should be measured net of work-related expenses such as out-of-pocket child care and transportation costs, but this is rarely done in practice. (See for example, Citro and Michael, 1995, for a proposed modification to the official poverty measure that takes this approach.) If these costs are deducted from income, a single mother moving from welfare to work may experience a decline in net income if the increase in her earnings is not large enough to offset the loss of welfare benefits and the increased out-of-pocket work-related expenses.(back)

66Some would argue that income, especially the way it is usually measured in surveys, may not be the best indicator of material well-being (Edin and Lein, 1997; Meyer and Sullivan, 2001; Haskins, 2001). More limited information on other measures of well-being, such as material hardship, food insecurity, and housing problems, will be covered in Chapter 9.(back)

67The lack of correspondence with the official U.S. poverty measure is not necessarily problematic, given the concerns with the validity of that measure (see Citro and Michael, 1995). The lack of comparability across studies is more of an issue.(back)

68For the same MFIP-IO sample with administrative data and survey data on combined income (earnings plus welfare, including the food stamp cash-out), the mean survey report of combined income for the AFDC (control) group exceeds the administrative data mean for combined income, while the reverse is true for the MFIP-IO (treatment) group (Miller et al., 2000, Table 4.6). Thus, the impact estimate (treatment-control difference) for monthly combined income is $109 (p < 0.05) based on the administrative data versus $12 based on the survey data. There is almost no difference between the treatment and control groups in the mean value of the other income sources (e.g., earnings of other household members, child support, and other income) collected in the survey data.(back)

69The impacts are shown in Canadian dollars. Converting to U.S. dollars using the rate $1Canada = $0.75U.S., the absolute changes in income are among the largest for the programs in Panel C.(back)

70The results for the NEWWS programs are similar when household income is considered instead of recipient income (Freedman et al., 2000a).(back)

71The fade-out of the FIP income effects for recipients by year two are likely to be repeated in the follow-up at three and one-half years post-randomization. Results for quarter 14 reported in Fraker and Jacobson (2000) show that average quarterly earnings are $8 more for the treatment group, while average quarterly FIP welfare benefits are $24 less. A combined recipient income result is not reported for the follow-up through quarter 14.(back)

72When income data for Jobs First are examined by quarter, the quarterly combined income impact falls from $266 (p < 0.01) to $150 (p < 0.01) between quarters 7 and 8 as the first recipients begin to reach the time limit, a difference that is equal to about 5 percent of the control group mean in quarter 7. The impact declines further from $152 (p < 0.05) in quarter 9 to $16 in quarter 10. The treatment-control difference remains insignificant through quarter 16 and is sometimes negative.(back)

73See the discussion in Chapter 4 of the interpretation of the age interaction with the time-limit dummy in Grogger's (forthcoming) analysis.(back)

74In Grogger's specification, the main effect of any reform (waivers or TANF) applies only to women whose youngest child is less than one. For these women, the interaction term between any reform and age of youngest child is zero.(back)

 

Table of Contents | Previous | Next