The TRIM3 Microsimulation Model
Linda Giannarelli, Paul Johnson, Joyce Morton, and Laura Wheaton Urban Institute
Estimates of the number of children who are income eligible for WIC presented in this report were derived from the Transfer Income Model version 3 (TRIM3) microsimulation model using data from the March 1999 Current Population Survey (CPS). Microsimulation models like TRIM3 use database records on individuals and families to simulate the effects of complex, large-scale governmental tax, transfer, and health programs at the individual, family, state, and national levels. The TRIM3 model and its predecessors have been used by analysts to understand the potential outcomes of such public policy changes as welfare reform, tax reform, and national health care reform since the 1960s.
Microsimulation models operate on individual units rather than aggregate information. In the case of WIC and other social welfare programs, those units are typically individual economic units, such as a family—however the program defines it—or an individual. The database used as input to a microsimulation model contains records describing persons, households, or businesses. The simulation model applies a set of rules to each individual record in the database and simulates eligibility or the dollar amount of benefits to which the unit is entitled under a government program or the amount of taxes owed by the unit. The weighted individual results are then added together to obtain the aggregate result.
For example, microsimulation models may be used to estimate tax liability for proposed changes to federal income tax rules. To simulate total
tax liability, the model would use individual records on families from the input database and apply the detailed federal tax rules to each family in the database. In effect, the program fills out the tax forms for each family. TRIM3 counts dependents, adds up income, subtracts adjustments to income, subtracts the larger of itemized or standard deductions, subtracts personal exemptions, computes taxes on available income, and computes and subtracts tax credits to arrive at the final tax liability for each family. To obtain aggregate tax liability, each family’s tax liability is multiplied by its weight and then added to obtain the total.
In some cases, the primary input database for a microsimulation model may not exactly match the data needed to simulate the policy. For estimating eligibility and participation in transfer programs, monthly income is typically needed to simulate the income eligibility provisions of the programs. However, the March CPS data provide only annual income data. The TRIM3 model includes a procedure to estimate monthly income from annual income reports. Further, since people tend to underreport their participation in transfer programs in surveys like the CPS, the TRIM3 model makes adjustments to account for this underreporting using control totals from administrative records from transfer programs.
This appendix briefly presents the history and current capabilities of TRIM3. The remaining sections cover the two aspects of TRIM3 used most directly in the analyses in this report: (1) the allocation of reported annual income amounts across the months of the year and (2) the simulation of transfer program eligibility and receipt.
HISTORY AND CAPABILITIES OF TRIM3
TRIM3 has been developed at the Urban Institute with funding from the U.S. Department of Health and Human Services (DHHS) and other government and private funders. TRIM3 is a descendant of the first microsimulation model ever developed—the Reforms in Income Maintenance (RIM) model first developed in 1969 by members of the President’s Commission on Income Maintenance Programs. RIM was followed by the first TRIM model, which was operational in 1973, and by TRIM2, which was operational in 1980. The goals of TRIM and TRIM2 were to make the system increasingly comprehensive, flexible, self-documenting, and useful for quick-turnaround policy simulations of tax and transfer policies.
In 1995, the assistant secretary for planning and evaluation (ASPE) of DHHS began funding the development of TRIM3. This latest version of
TRIM grounds the simulation model on a computation platform that takes advantage of relatively recent advances in information technology, such as the rise of PCs, windowed user interfaces, client/server systems, open relational databases, object-oriented programming, and the Internet. TRIM3 was introduced in 1997 and is currently being used by researchers at the Urban Institute, DHHS-ASPE, and the Office of Management and Budget (OMB).
TRIM3 can be used to model the following U.S. tax and transfer programs:
Cash and in-kind transfer programs:
Supplemental Security Income (SSI)
Temporary Assistance to Needy Families (TANF)
Food Stamp Program
Child care subsidies (eligibility for subsidies and amount of potential copayment)
Public housing and subsidized housing programs (value of subsidy)
Health insurance programs
Medicaid and S-CHIP
Employer-sponsored health insurance
Federal income taxes
State income taxes
TRIM3 is uniquely qualified to estimate WIC eligibility because it can simulate eligibility and participation in Medicaid. This capability of TRIM3 is why it was chosen to make the estimations of WIC eligibility in this report.
THE ALLOCATION OF REPORTED MONTHLY INCOME AMOUNTS ACROSS THE MONTHS OF THE YEAR
To simulate the number of people adjunctively eligible for WIC through other transfer programs, TRIM3 requires monthly rather than annual income amounts. However, the March CPS file asks respondents
about annual income from various sources. To estimate monthly income, the TRIM3 system takes the annual income amounts for each person and “allocates” them across the months of the year. The exact procedures for this allocation vary by type of income and are most detailed for earnings. The following sections explain how monthly income totals from various sources are generated by TRIM3.
Weeks of Work and Unemployment
The monthly allocations for earnings and some types of unearned income are based on the estimated distribution of weeks worked and/or weeks unemployed across the months of the year. CPS respondents report several items of information that are used to make these allocations: the number of weeks of employment, the number of employers during the year, the number of weeks of unemployment (looking for work), and the number of stretches of looking for work. TRIM3 allocates the reported weeks of work and/or the reported weeks of unemployment across the months in a way that is consistent with the other reported information. For example, if a person reports 26 weeks of work, 26 weeks of unemployment, and 1 stretch of unemployment, then all the weeks of unemployment will be placed consecutively. The exact placement of the weeks (for instance, whether to start the 26 weeks of unemployment in January, July, or somewhere in between) is largely random. However, the procedure is controlled so that monthly unemployment rates generated from the TRIM3 CPS-based estimates have the same trend over the year (although not necessarily the same exact levels) as the actual unemployment rates reported by the Bureau of Labor Statistics from the monthly CPS.
Once the weeks of employment and unemployment have been allocated across the year, TRIM3 uses this information to allocate earnings across the weeks of employment. All the combined reported annual earned income—wages, farm income, and self-employment nonfarm income—is allocated evenly across the weeks of employment during the year. Thus, if a person reported working all 52 weeks, the earnings will be distributed evenly across those weeks. If a person reported working all 26 weeks, the earnings will be distributed evenly over those weeks.
In general, annual unemployment compensation income is divided evenly across weeks of unemployment. However, for a randomly-selected 51 percent of the annual unemployment compensation recipients, the monthly amounts are lagged by one month to capture real-world delays in receipt of unemployment compensation relative to the start of a spell of unemployment. The percentage is based on data from the Survey of Income and Program Participation (SIPP).
Allocation of annual workers’ compensation income across months of the year depends upon how much compensation the individual unit reports receiving. If the unit reports receiving more that $7,800 (for the 1998 reporting year) it is assumed that the unit received compensation throughout the year and workers’ compensation income is divided evenly across the months of the year. If it is less than that amount, the assumption is that it was not received in every month. In that case, 20 percent of recipients are randomly selected to receive the workers’ compensation in one month, and the remaining 80 percent have their workers’ compensation income divided evenly over weeks of non-work (either unemployed or not in the labor force). These percentages were calculated from SIPP data which include monthly reports on workers’ compensation receipt. The $7,800 threshold for determining whether workers’ compensation income is divided evenly across the months of the year or not was originally based on SIPP data of monthly workers’ compensation totaled across the year, and has since been updated for inflation.
Child Support and Alimony
Annual income amounts from child support and alimony are allocated across the months of the year based on patterns of receipt of income from these sources as reported in SIPP. SIPP data were used to develop “look-up” tables that give, for different ranges of combined annual alimony and child support income, the percentage of recipients getting that income in 1 month, 2 months, 3 months…12 months. These tables are used to guide how an individual unit’s annual report of child support and alimony income is allocated across the year, depending upon which range of com-
bined child support and alimony the unit falls into. For each recipient, TRIM uses random numbers to pick a number of months of receipt of child support and alimony, and then uses random numbers to assign those months within the year (not necessarily consecutively). The assignments are made such that the resulting percentage distributions of recipients by months of receipt match the percentages in the look-up table. Once each unit is assigned months of the year for child support and alimony receipt, their annual incomes from these sources are divided evenly across these months.
Allocating Asset Income
All types of asset income—interest, dividends, and rent—are divided evenly across the months of the year.
Allocating Other Unearned Income
All other income amounts—including Social Security and Railroad Retirement, government pensions, private pensions, veterans’ benefits, “other” income, and contributions—are assumed to be received in 12 equal monthly amounts.
SIMULATING ELIGIBILITY AND RECEIPT OF TRANSFER PROGRAM BENEFITS
Simulation estimates made in this report require information on both the amount of income received from transfer programs (to determine a unit’s total income) and eligibility and participation in these programs (to determine if a unit is adjunctively eligible for WIC). The estimates made in this report involved simulations of three different transfer programs: TANF, the Food Stamp Program, and Medicaid (including State Children’s Health Insurance Programs (S-CHIP) funded Medicaid expansions).1 For each of these programs, TRIM3 simulates eligibility, participation, and levels of benefits received. Results of the simulations are saved as new
variables on the data records, augmenting the CPS-reported information. One benefit of a comprehensive model like TRIM3 is that simulated variables can be “passed” from one part of the simulation to the next. In this case, the results of the TANF simulation are used by the FSP and Medicaid simulations.
The eligibility simulations are very detailed and attempt to mimic how eligibility rules would apply to the units in the CPS if they actually sought program assistance. Demographic tests (including immigrant-status tests2), asset tests,3 and income tests are applied to the extent possible given the information available in the CPS data. If a program’s rules vary by state, the rules that are applied to a particular household are the rules in effect in that household’s state of residence. The TRIM3 simulations refer to a detailed database on program rules in each state, for each year and each program. The detailed TANF rules are based on the information in the Urban Institute’s Welfare Rules Database,4 which is derived from review of each state’s caseworker manuals and/or regulations. The Medicaid rules capture the real-world state-specific variation in rules such as the percent-of-poverty thresholds and medically needy income thresholds. They also capture the difference between the S-CHIP-funded Medicaid expansions and the separate state S-CHIP programs. People who enrolled in S-CHIP-funded Medicaid expansions are adjunctively eligible for WIC but those who enrolled in separate state S-CHIP-funded programs are not, unless they are otherwise income eligible.
Each simulation uses the “filing unit” that is most appropriate to that particular program. The filing unit is the group of people who together would file for and potentially receive the benefit. For TANF, the filing unit is a narrowly defined family, with subfamilies treated separately from the primary family. In the Food Stamp Program, the filing unit may be the entire household or it may be smaller, particularly when a broader house-
The CPS includes data on citizenship and nativity but does not include immigrant legal status. Legal status is imputed through complex procedures that hit targets for legal versus illegal immigrants, based on demographic and employment characteristics.
Vehicle assets tests are not modeled, because there is no CPS variable for the value of a vehicle, and no imputation has yet been incorporated into TRIM3 to create such a variable.
The Welfare Rules Database is available on line at <newfederalism.urban.org/wrd>.
hold includes an AFDC/TANF unit. For Medicaid, the filing unit may be either a family or an individual.
Eligibility simulations are performed on a month-by-month basis. Thus, a family might be eligible for a program in one month but not another, or eligible for a different amount in one month than another. Note, however, that only one set of rules is stored in the database for each year. The rules used for each year are the rules in effect for the majority of that year.
For additional details about simulating eligibility for the transfer programs, see the TRIM3 web site at <trim3.urban.org>.
Modeling the Participation Decision
For each transfer program, TRIM3 determines whether a particular filing unit that is eligible for that program will in fact apply for and receive the benefits from that program. The CPS respondents do report whether the household received food stamps or TANF and whether any of the members were enrolled in Medicaid (again, these reports are on an annual basis). However, there is serious underreporting of transfer program benefits (Wheaton and Giannarelli, 2000). The public-use CPS captures only about 61 percent of the 1998 TANF caseload, 67 percent of the Food Stamp Program caseload, and 68 percent of the Medicaid caseload. The TRIM3 simulation corrects for this underreporting of transfer benefits.
Participation methodologies vary somewhat among the simulated programs, but several general methods apply to all programs. Units that are eligible for assistance and report receiving assistance on the CPS are assigned to participate. Note that responses that were “allocated” (imputed) by the Census Bureau are not considered to be actual reports. Units that are simulated to be ineligible for assistance are not assigned to participate, even if they report receiving that assistance according to the CPS data. Additional eligible nonreporters are selected to participate in such a way that the simulated caseload matches the caseload reported from administrative records as closely as possible in terms of size as well as key characteristics—typically unit type, benefit level, citizenship status, and state. Thus, the final TRIM3 version of the CPS data corrects for the under-reporting of transfer benefits.
For additional details about how TRIM3 models participation for each of the transfer programs, see the project’s web site at <trim3.urban.org>.