Printer Friendly

Effects on the AFDC-Basic caseload of providing welfare to two-parent families.

As of October 1990, the Family Support Act required all states to provide income assistance to two-parent families in which the principal wage earner is unemployed. Prior to this mandate, this program, known as Aid to Families with Dependent Children--Unemployed Parents (AFDC-UP), was optional for states. The decision to require states to offer AFDC-UP benefits was accompanied by considerable controversy about its effects on both marital stability and AFDC-Basic, the program that primarily serves single-parent families. The research described here, while not resolving the controversy about the effect of the AFDC-UP program on marital stability, provides new evidence on the effects of AFDC-UP on the number of families receiving AFDC-Basic. At a time when the current administration is planning to review the welfare system such evidence about the effects of past changes may be useful.

THE WELFARE-FAMILY STABILITY DEBATE

The debate over extending AFDC-UP to all states reflected the ongoing controversy over the effect of welfare receipt on family formation and stability. While some scholars argue that the availability of AFDC has discouraged the formation and stability of two-parent families,' others point to macro-economic and societal forces to explain changes in family structure. The latter contend that changes in family stability can be attributed to such trends as decreasing wages and increasing unemployment (which made men less desirable as marriage partners) and increased participation of women in the labor force (which gave women more independence from men).(3)

In the policy arena, empirical evidence from the income maintenance experiments conducted during the '60s and '70s has been cited to argue against the extension of AFDC to two parent families.(4) These experiments found an increase in separation among couples receiving income support.(5) Possible explanations for the destabilization associated with the receipt of assistance included the combination of marital stress resulting from prolonged unemployment and the certainty of continued cash support for the different family members even if the family separated.(6) However, proponents of the AFDC-UP program object to this use of the findings of the income maintenance experiments. Recent reanalyses have raised questions about the proper interpretation of data from the Seattle and Denver Income Maintenance Experiments (SIME-DIME).(7) In addition, some commentators have argued that SIME-DIME, regardless of how one analyzes the data, tested an income support program quite different from the one operated by the federal government as AFDC.(8)

The two sides of the argument have different implications for the effect of the AFDC-UP program on the AFDC-Basic caseload. If receipt of AFDC-UP has a destabilizing effect on the family, one would expect AFDC-Basic cases to increase. Almost all former UP families would be eligible for AFDC-Basic in the event of one parent's departure. This consequence is of special concern because, on average, single parent families use welfare benefits for a longer two parent families and thus cost more to support. If, on the other hand, the availability of AFDC-UP provides an alternative to couples that would otherwise have to separate in order to provide for their children, one would expect the AFDC-Basic caseload to decrease or grow at a slower rate.

Although the AFDC-UP program was made mandatory, debates about its likely effects continue, due partly to the paucity of empirical data. Debate often focuses on the effects of various welfare measures on "the poor," but the population in poverty is much larger than the group potentially eligible for AFDC benefits, covering many people who would not be eligible for AFDC; so program-based data may provide a more sensitive indicator of welfare program effects. Unlike the income maintenance experiments, the research described here was grounded in AFDC program data, thus making the results clearly applicable to AFDC families.

A STRATEGY FOR EVALUATING AN ENTITLEMENT PROGRAM

We applied separate interrupted time series analyses to data from two states to examine the relationship between the availability of UP benefits and the number of families receiving benefits through each state's AFDC-Basic program. This time series design took advantage of the natural start-and-stop interventions that have marked the AFDC-UP program. Although random assignment of eligible welfare families to benefit and non-benefit conditions could permit more conclusive findings, this was not a possibility because AFDC-UP was already an entitlement available to all families who qualified. (State experimentation with AFDC's work incentives and other penalty and reward structures has been encouraged by the federal government, however these experiments do not extend to the denial of all cash benefits to randomly selected families.(9) Moreover, the state-specific eligibility requirements for the AFDC-UP program-such as detailed evidence of work history-make it difficult to identify potentially eligible poor families in non-program based data series that contain information on marital status. Thus, as an alternative to both experimental research and other sorts of quasi-experiments, we used interrupted time series to exploit the period when the AFDC-UP program was optional.

States were first allowed to provide benefits to two-parent families in which the principal earner was unemployed in 1961. Between 1961 and 1990, some states offered the program for a time and then dropped it while others chose not to offer UP benefits for several years before adopting the program.(10) We reviewed data from all the states that reported these programmatic starts and stops after 1975. We did not consider states that had changes in their UP program in 1975 or before because we were unable to separate the AFDC-Foster Care caseload from the caseloads for the other AFDC components for that period.

Thus, eight states were originally considered: Colorado, Maine, Missouri, Montana, Oregon, South Carolina, Utah, and Washington. In six of the states--Maine, Montana, South Carolina, Utah, and Washington--the timing of other policy changes made our analyses impossible to interpret.(11) We report the analyses of the remaining two cases: Colorado and Oregon. In Oregon, we supplemented our examination of Basic caseloads with a look at the more sensitive measure of the number of new families added to the Basic caseload.

Sources of Data

The monthly level of the AFDC-Basic caseload for each state was derived using three series: the total AFDC caseload, the AFDC-UP caseload for the corresponding month, and the AFDC-Foster Care (AFDC-FC) caseload. The AFDC-Basic caseload was found by subtracting AFDC-UP cases and AFDC Foster Care cases from the total caseload reported for the particular month. Foster Care cases were only subtracted prior to October 1981, since these cases were not covered under AFDC after that date and were no longer included in its caseload totals. These calculations were based on data available in the Social Security Bulletin and data provided on microfilm by the Administration for Children and Families in the Department of Health and Human Services.

Before analysis, historical data were collected on a range of economic, demographic, and policy variables that might bear a meaningful relationship to welfare caseloads. (However, as described below in the section on analysis decisions, not all variables for which we collected data were included in the final set of models.) Economic data for each state were obtained from LABSTAT, an economic database maintained by the Bureau of Labor Statistics. These series included general unemployment as well as employment and wages in specific trades or industries. LABSTAT data were supplemented by historical series on initial unemployment claims and unemployment insurance exhaustions, which were taken from the Social Security Bulletin.

Demographic caseload predictors included state population, births, divorces, and births to unwed mothers. In most states, these data were not available on a monthly basis and the monthly values had to be estimated based on the annual totals. For example, monthly births to unwed mothers were estimated by multiplying total annual births to unwed mothers by the proportion of births to all mothers that occurred in a particular month.

Finally, we obtained information on policy variables by interviewing state officials. We also used these interviews to verify historical data on AFDC need standards and payment levels which were obtained from the Congressional Research Service and the Administration for Children and Families. Based on states' input, dummy variables were constructed to represent the implementation of the Omnibus Budget Reconciliation Act of 1981 (OBRA), the Deficit Reduction Act of 1984 (DEFRA), and state-initiated policy changes. OBRA sharply restricted AFDC eligibility and caused major declines in the AFDC-Basic caseloads. A second major policy change, the DEFRA, is believed to have mildly liberalized access to AFDC benefits, but also included elements regarding the counting of income from immediate family members that could have exerted some downward pressure on the Basic caseload.

Analysis Decisions

Unlike the interventions typically studied with an interrupted time series approach, the impact of UP policy on the AFDC-Basic caseload was expected to be gradual. Our analyses took this into account by incorporating an additional variable to index the effect of UP policy on the rate of growth in the caseload. The effect of UP on the rate of caseload growth was indicated by the coefficient for a variable that represented the product of the UP dummy variable and a time counter that was also incorporated in the model. When all other factors are held constant, the coefficient for the time counter can be interpreted as the monthly rate of increase in the caseload when UP is not present; the coefficient for the UP x TIME product variable is interpreted as the change in the rate of caseload growth associated with the presence of UP.

We used the AUTOREG procedure in the SAS/ETS program library for data analysis based on Extended Generalized Least Squares (EGLS) procedures using the two-step Prais-Winsten estimator.(12) EGLS procedures have two advantages. First, in contrast to ARIMA and other mathematical forecasting methods, the EGLS method allowed us to incorporate information about factors that might affect the Basic caseload to enhance chances of anticipating major turning points.(13) Second, unlike Ordinary Least Squares (OLS) analyses, the EGLS procedure accounts for serially correlated error that characterizes most time series data. Nonrandom, serially correlated errors violate the assumptions of OLS regression procedures for calculating standard errors and significance levels. The number of AFDC cases in any given month generally depends heavily on the number of cases in the previous month. Thus, the EGLS procedure is generally more appropriate for this type of data.(14)

To assess whether the degree of first-order serial correlation among the residuals was high enough to seriously violate the assumptions of OLS regression, we used the Durbin-Watson d statistic. The d statistic can vary between 0 and 4; the closer it comes to either extreme, the stronger the autocorrelation between residuals. In general, a d statistic close to 2 suggests that first-order serial correlation among residuals is negligible.(15) With the exception of the analysis of Oregon's new case openings, the d statistic for the caseload models indicated that EGLS procedures was more appropriate than OLS.

Using the EGLS procedure, we attempted to develop two kinds of regression models in each state: a "dummy variable" model and a "predictive" model. Dummy variable models employed data from periods before and after a UP policy change. In developing models using this approach, we generally adjusted first for major policy changes (e.g., implementation of OBRA) and then included such variables as population and unemployment claims or unemployment insurance exhaustions. Before accepting a dummy variable model for re-estimation with the UP variables, we required that it meet two criteria: (1) the full model (including the autocorrelation coefficient that accounts for serial correlation of errors) had a squared multiple correlation coefficient of .90 or higher and (2) all regression coefficients were significant and had the expected signs. Like other authors that have used AFDC models to form conclusions about program impact, we rejected models with coefficients that were not theoretically sensible in order to seek unbiased estimates of the UP parameter.(16) For example, a model with a negative unemployment rate coefficient, associating increased unemployment with decreased caseloads, would be rejected.

To augment this basic approach, we also developed a second, "predictive" model that we validated by predicting the 12 months of data before the UP policy change. If a model issued predictions for this 12-month period that were not significantly different from the observed values at the .05 level, we re-estimated the model incorporating the 12 months of test data and determined whether the model significantly over- or under-predicted the actual caseload after the UP policy change. Since these models were tested for their predictive capacity, we included those variables that achieved significant coefficients and appeared to improve prediction over the test period. The data series to which these models were applied were shorter than those used for the dummy variable models and probably covered periods of limited variation in some variables, yielding models that predicted adequately while including fewer variables than the comparable dummy variable models.

The major limitation of the modeling approach we used is the absence of consensus on the factors that drive AFDC caseloads. However, our goal was not to explain the AFDC-Basic caseload, but rather to predict what the caseload would have been without the UP intervention. Thus, we selected models primarily on the basis of their statistical strength, although-as explained above-we also sought coefficients with sensible signs. As with any regression, because the coefficient for the UP term could, in theory, change substantially depending on the set of other variables included in the model, we interpreted only the direction, not the size of the coefficient. In order to test the sensitivity of our models to different variables, we explored various combinations of demographic, economic, and policy factors. Although we only report the strongest models developed, these models were consistent with other models for the states.

RESULTS

The results of our analyses in Colorado and Oregon associate the presence of a UP program with a decrease in the rate of growth in the number of families receiving AFDC-Basic.

The Oregon Models

The strongest evidence that the presence of the UP program reduces the rate of growth in the Basic caseload comes from Oregon, which reinstated UP in 1986 after it had been suspended for several years. Unlike other states, the UP intervention in Oregon was not confounded by other major policy changes, since it occurred after both OBRA and DEFRA. Moreover, the association between the reinstatement of the UP program and a reduction in the rate of growth was found in both dummy variable and predictive models of the two dependent variables: AFDC-Basic caseload and new Basic openings." As shown in Figures 1 (caseload) and 2 (case openings), the dummy variable models predicted that the dependent variable (caseload or openings) would grow at a faster rate if the UP program had not been implemented. (See Tables 1 and 2 for the models.)

[TABULAR DATA 1 & 2 OMITTED]

In the period after the AFDC-UP program was reinstated in February 1986, the predictive models over-predicted the AFDC-Basic caseload. All but the first four post-reinstatement months are significantly less than predicted (at the .05 level) in the model of Oregon's AFDC-Basic caseload. In the model of AFDC-Basic openings, fifteen of the twenty-three months following the reinstatement of the UP program are significantly less than predicted.

Although the unemployment rate was decreasing when the UP program was reinstated, the inclusion of economic control variables, such as retail employment, unemployment rate, and unemployment insurance claims, did not change the results of the Oregon analyses. Thus, changes in the economy are not a likely alternative explanation for the association between the UP program and the decreased rate of growth in the Basic caseload.

The Colorado Models

As the broken line in Figure 3 indicates, the dummy variable model of Colorado's AFDC-Basic caseload predicted that the caseload would rise more steeply in the absence of the UP program. The predictive model applied to the post-suspension period indicated that the actual Basic caseload was higher than would have been expected if the UP program had continued. (Table 3 shows the dummy variable and predictive models for Colorado.)

[TABULAR DATA 3 OMITTED]

We identified two other possible explanations for our finding in Colorado, but with further analysis considered them unlikely arguments against the effect of AFDC-UP. First, state administrators reported that one large county had systematically transferred former UP cases to the AFDC-Basic caseload under the category of families eligible as a result of parental incapacity. If this were responsible for the subsequent sustained increase in Colorado's caseload, we might have expected to see larger portions of children eligible due to parental incapacity after the UP program was suspended than before. However, quality control data from the Administration for Children and Families indicated that this proportion did not change markedly and may have decreased slightly between periods before and after the UP policy change. Thus, the notion that the post-UP increase in Colorado's AFDC-Basic caseload is attributable to simple reclassification of former-UP families appears inconsistent with available data.

The implementation of DEFRA is another possible reason for the change in the rate of growth of the Basic caseload after 1985. However, it is also not a likely explanation. Colorado's standard for paying benefits at the time was far lower than its need standard, so liberalizing the standard of need was unlikely to have markedly increased the number of cases receiving payments.

DISCUSSION

The utility of the time series models derives in large part from their reliance on AFDC program data rather than data on the low-income population in general. Because we used AFDC program data, we were able to model the effects of state welfare policies-such as Oregon's policy to link the need standard to the minimum wage-as well as macroeconomic changes. Although the models support the importance of economic changes (such as unemployment in the Colorado model) and demographic variables (such as births to unwed mothers in the Oregon models), these variables are linked to the number of families receiving benefits by a relatively indirect chain of events. In contrast, the caseload models indicated strong effects for the state welfare policies that affected large numbers of families eligible for AFDC benefits. The evidence found for the effects of changes in other state welfare policies strengthens the argument that dramatic changes in UP policy could affect the caseload at least as strongly as the macroeconomic and demographic varialbes.

As shown above, our time series analyses indicated that the UP program was associated with a decrease in the Basic caseload's rate of growth. Although the dampening effect seen on the Basic caseload may be seen as evidence for AFDCUP's effects on the formation or maintenance of two-parent families, our analyses only indirectly inform the debate over UP's effects on family stability. However, as described above, it is logical to link the withdrawal of UP benefits to increases in AFDC-Basic caseload because most single-parent families resulting from the dissolution of a former UP family would be eligible for AFDC-Basic.

The best rationale for the UP program does not hinge on evidence that the benefits enhance family stability; providing such benefits is amply justified on the grounds of equity. However, it is important that the benefits also do no harm. Thus, although our analyses do not provide conclusive support for the argument that the UP program encourages family stability, it is notable that we found no evidence to suggest that the provision of welfare benefits to two parent families is associated with higher numbers of single parent families joining the caseload.

Acknowledgments: The authors thank the following individuals for their assistance with the caseload data, comments on an earlier report of these analyses, or advice on the analyses: Burt Barnow, Douglas Besharov, Emmett Dye, Tom Fraker, Daniel Friedlander, Barbara Goldman, Robert Greenstein, Mark Lipsey, Mark Rom, Lori Schack, and Joseph Wholey. The views expressed in this article do not necessarily represent those of the U.S. General Accounting Office.

NOTES

(1.) George Gilder, Wealth and Poverty (New York: Basic Books, 1981); Charles Murray, Losing Ground: American Social Policy, 1950-1980, (New York: Basic Books, 1984). (2.) Frances Fox Piven and Richard A. Cloward, Regulating the Poor: The Functions of Public Welfare (Pantheon, 1971). (3.)David T. Ellwood, Poor Support: Poverty in the American Family (New York: Basic Books, 1988); Daniel P. Moynihan, The Negro Family: The Case for National Action (Washington, DC: U.S. Department of Labor, March 1965); William J. Wilson and Katherine M. Neckerman, "Poverty and Family Structure: The Widening Gap between Evidence and Public Policy Issues," in Fighting Poverty: What Works and What Doesn't, edited by Sheldon Danziger and Daniel Weinberg (Cambridge, MA: Harvard University Press, 1986), pp. 232-259. (4.) Russell Long, cited in Sanford Schram and Michael Wiseman, "Should Families Be Protected from AFDC-UP?" Institute for Research on Poverty discussion paper 860-88 (Madison, WI: Institute for Research on Poverty, 1988). (5.) Michael T. Hannan, Nancy Brandon Tuma, and Lyle P. Groeneveld, "Income and Independence Effects on Marital Dissolution: Results from the Seattle and Denver Income-Maintenance Experiments," American Journal of Sociology, 84(1978): 611-633. (6.) John Bishop, "Jobs, Cash Transfers, and Marital Instability: A Review and Synthesis of the Evidence," Journal of Human Resources, 15(1980):301-334. (7.) Glen G. Cain and Douglas A. Wissoker, "A Reanalysis of Marital Stability in the Seattle-Denver Income-Maintenance Experiment," American Journal of Sociology, 95(1990):1235-1269; Michael T. Hannan and Nancy Brandon Tuma, "A Reassessment of the Effect of Income Maintenance on Marital Dissolution in the Seattle-Denver Experiment," American Journal of Sociology, 95(1990): 1270-1298. (8.) John Bishop, op. cit.; U.S. General Accounting Office, Welfare Reform: Projected Effects of Requiring AFDC for Unemployed Parents Nationwide (GAO/ HRD-88-88BR) (Washington, DC: Author, 1988). (9.) Michael Wiseman, "Welfare Reform in the States: The Bush Legacy," Focus, 15(1993): 18-36. (10.) Congressional Research Service, State Use of the Aid to Families with Dependent Children-Unemployed Parent [AFDC-UP] Program: An Overview, 87-969 EPW (Washington, DC: Author, 1987). (11.) In Montana, Utah, and Washington, the timing of the change in UP policy coincided with the states' implementation of the Omnibus Budget Reconciliation Act of 1981 (OBRA) which sharply restricted AFDC eligibility and caused major declines in the AFDC-Basic caseloads. In Maine, the Defcit Reduction Act of 1984 (DEFRA) was implemented at the same time as UP. DEFRA is believed to have mildly liberalized access to AFDC benefits, but also included elements that could have exerted some downward pressure on the Basic caseload. Missouri's suspension of the UP program lasted only two years and children in former UP families continued to be eligible for Medicaid during the suspension. South Carolina's implementation of the UP program in 1986 occurred concurrently with a major increase in the need standard. (12.) SAS Institute, SAS/ETS User's Guide, Version 6 (Cary, NC: Author, 1988). (13.) Spyros Makridakis, Steven C. Wheel right, and Victor E. McGee, Forecasting: Methods and Applications, 2nd ed. (New York: John Wiley and Sons, 1983). (14.) Charles W. Ostrom, Jr., Time Series Analysis: Regression Techniques, 2nd ed. (Newbury Park, CA: Sage Publications, 1990). (15.) Charles W. Ostrom, Jr., op. cit. (16.) Burt S. Barnow, Estimating the New Jersey AFDC Caseload. Prepared by ICF, Inc., for the U.S. Department of Health and Human Services, Office of the Secretary, Assistant Secretary for Planning and Evaluation (Washington, DC: U.S. Department of Health and Human Services, 1988); Steven Garasky, "Analyzing the Effects of Massachusetts' ET Choices Program on the State's AFDC-Basic Caseload," Evaluation Review, 14(1990):701-710. (17.) OLS procedures are used to model Oregon's new case openings because the value for the Durbin-Watson d statistic did not indicate that the GLS procedure was required. Unlike the caseload, which generally depends heavily on its previous value, the number of new case openings in a particular month can be viewed as an independent observation.

LESLIE J. C. RIGGIN, Direct all correspondence to: Leslie J. C. Riggin, U.S. General Accounting Office, 1244 Speer Blvd., Suite 800, Denver, Colorado 80204.
COPYRIGHT 1995 Elsevier Science Publishers
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 1995 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Title Annotation:Aid to Families with Dependent Children
Author:Riggin, Leslie J.C.; Ward-Zukerman, Betty
Publication:The Social Science Journal
Date:Jul 1, 1995
Words:3958
Previous Article:Factors shaping the probability of community vs. four-year college entrance and acquisition of the B.A. degree.
Next Article:Indian self-determination: the charge to Indian journalists.
Topics:

Terms of use | Privacy policy | Copyright © 2021 Farlex, Inc. | Feedback | For webmasters |