Economic policy: simulating effects of welfare reform on dependency, work, and poverty.

• Author: Jacobson, Jonathon
• Position: Statistical table - Abstract

Abstract

We use a version of the Simulation of Trends in Employment, Welfare, and Related Dynamics (STEWARD) model to generate provisional estimates of possible responses to Temporary Assistance to Need Family (TANF) plans for single mothers over a four-year period. Varying assumptions about the anticipation of TANF time limits, we find that there is considerable variability in the simulated level of TANF caseloads, but that the employment-related characteristics of the caseload are more stable. We find that, regardless of whether the TANF plan is aggressive or moderate in its incentives and requirements, the percentage of single mothers on TANF in private jobs for at least 30 hours per week is consistently below 10 percent. We also find that levels of extreme poverty among single parents are higher under the aggressive plan than the moderate plan, and under the assumption of forward-looking behavior than the assumption of myopia.

**********

With the shift in responsibility for welfare policy development from the federal government to the states, state officials have an urgent need for better information on how alternative reform policies will affect key outcomes such as program costs, caseloads, and their ability to meet the employment targets embodied in the recent legislation. Estimates of how the combination of time limits, work requirements, and enhanced work incentives will affect work behavior and welfare participation are particularly important. This paper reports on the continued development and application of a longitudinal microsimulation model, the Simulation of Trends in Employment, Welfare, and Related Dynamics (STEWARD), which was designed to produce such estimates. Unlike earlier models which had limited ability to examine behavioral responses, STEWARD explicitly models the time limits and changed work incentives. These policies are a central feature of the cash assistance programs states have been developing under the new welfare reform legislation.

The next sections briefly review the policy context and simulation methods. That is followed by a discussion of different approaches to designing TANF plans, how these plans are modeled in STEWARD, and how simulation estimates depend on some of the key assumptions regarding forward-looking behavior. The last section presents preliminary estimates of the effects of selected reform packages on welfare caseloads, employment-related outcomes, and poverty levels. Since the development of an improved version of the STEWARD model is an ongoing process, the reader should view the simulation estimates in this paper as provisional and illustrative only.

Policy Context

The Personal Responsibility and Work Opportunity Reconciliation Act of 1996 (PRWORA) made sweeping changes to the U.S. welfare system, including giving states great flexibility to design their own cash assistance programs for low-income families with children. States have begun to exercise this new authority in order to satisfy broad requirements of PRWORA and qualify for funding under the Temporary Assistance to Needy Families (TANF) block grant established by the act. However, even several months after the July 1, 1997 deadline for submitting TANF plans to the federal government, some states are operating essentially placeholder programs that they expect to change substantially during the next year. Nearly all states will continue to refine their programs for years to come. As state policymakers consider both large and small program reforms, they will need reliable estimates of the cost and distributional effects of alternative policies. The goal of STEWARD is to provide a policy analysis tool that combines the best information on how recipient work and welfare participation behavior varies depending on the personal characteristics and economic opportunities facing recipient families with information on the number and distribution of potential recipient families in each state.

What is Microsimulation: an Overview

A microsimulation model simulates how proposed changes to government assistance programs affect the programs and program participants. The model has two elements: (1) a micro database and (2) a computer program. The database is constructed from survey or administrative data with information on households in the population targeted by the government program. The model's computer program codes the rules of the government program under both the "baseline" policy, which is typically the current policy, and a "reform" policy, which is an alternative under consideration. The computer program also simulates what a caseworker doesBthat is, it determines whether a household is eligible for the government program and the benefits for which the household would qualify. In addition, the computer program simulates a household's behavioral response, determining whether the household will participate in the program and how much the adult members will work. Processing all the households in the database as they make work and welfare decisions over 48 months, the model counts participants to estimate the caseload of the government program, adds up their benefits to estimate program costs, and tallies changes work behavior. By performing these operations under both baseline and reform policies and comparing the results, the model estimates the effects of the proposed reform on caseloads, costs, and employment levels. The model can also estimate the distributional effects of the reform on outcomes such as poverty rates.

Steward Conceptual Structure

Overview of the STEWARD Model

The STEWARD model was first developed by Mathematica Policy Research, Inc., between 1993 and 1995 with funding from the Assistant Secretary for Planning and Evaluation, U.S. Department of Health and Human Services (DHHS). The purpose of the model was to simulate, on the national level, responses of female single parents over time to changes in AFDC and related policies. STEWARD generated estimates of changes in AFDC participation, food stamp participation, and labor supply response to changes in welfare, health, and tax policies, as discussed in Beebout et al (1994). The model also simulated the participation of AFDC recipients in the Job Opportunity and Basic Skills (JOBS) education/training program.

The original STEWARD model database consisted of longitudinal records from the National Longitudinal Survey of Youth (NLSY), 1979 to 1982 and 1986 to 1989, and the Panel Study of Income Dynamics (PSID), 1985 to 1988. These data were weighted to be representative of the national population of single mothers in the U.S. Current Population Survey (CPS) as of 1992. Each of the 3,194 single mothers in the sample was observed monthly for 48 months, and observed changes in marital status or fertility were used to account for demographic transitions over time.

The original STEWARD's model of labor supply and program participation was a structural model based on the work of Keane and Moffitt (1991, 1995). Each single mother's preferences for work are described by a utility function that is quadratic in hours worked per week and in disposable family income per month (Table 1, equation 1). The individual's preference for hours of work depends on personal characteristics (equation 2). Hours worked per week are assumed to be constant within a month and take one of four distinct values: zero, ten, twenty, thirty, of forty (equation 3). Disposable income equals the sum of earnings, AFDC and food stamp benefits, other non-labor income, and a valuation of private health insurance or Medicaid, minus federal income and payroll taxes, child care costs, transportation costs, and employee health insurance premiums (equation 4). Earnings per hour were imputed using a wage equation for persons who did not report a wage in the database (equation 5), and analogous equations were used to impute child care costs (equation 6), and health coverage probabilities (equation 7). Also included in the utility function were "stigma" equations capturing the psychic costs of participating in the AFDC and food stamp programs; these depend on personal characteristics plus variables describing the duration of participation in AFDC, food stamps only, or neither program (equations 8 and 9). The approach for estimating the parameters of the utility function involved the use of logit maximum likelihood procedures and is described in detail in Beebout et al (1995); we have since re-estimated these equations using a slightly different specification.

During 1996 and 1997, Mathematica received funding from the Smith Richardson Foundation, the Food and Consumer Service of the U.S. Department of Agriculture, and DHHS to develop a new version of STEWARD, called MATH-STEWARD (for Micro Analysis of Transfers to Households). Unlike the original STEWARD, MATH-STEWARD relies on data from the early 1990s through its use of three years of longitudinal data from the 1992 panel of the Survey of Income and Program Participation (SIPP). MATH-STEWARD will simulate program participation, labor supply, and child care outcomes for a broader population of households than the original STEWARD, and it will be able to generate simulation estimates for individual states. It will include a new set of behavioral equations estimated from SIPP data. Table 2 compares the major design features of the original STEWARD model and MATH-STEWARD.

While our original intention was to use the MATH-STEWARD model to generate estimates for this paper, the current schedule of model development has prevented us from achieving this goal. Instead, we have used the original STEWARD model to simulate the effects of illustrative TANF programs on the 1992 population of single mothers in the United States. These simulations incorporate two enhancements of the original model's simulation capabilities, the first related to the anticipation of TANF time limits, and the second related to the simulation of...