THE MEDICAID NOTCH, LABOR SUPPLY AND WELFARE PARTICIPATION: EVIDENCE FROM ELIGIBILITY EXPANSIONS^*

Aaron S. Yelowitz

Abstract

            I assess the impact of losing public health insurance on labor market decisions of women by examining a series of Medicaid eligibility expansions targeted towards young children. These targeted expansions severed the historical tie between AFDC and Medicaid eligibility. The reforms allowed a mother's earnings to increase without losing public health insurance for her young children. Increasing the income limit for Medicaid resulted in a decrease in AFDC participation and an increase in labor force participation among these women. The effects were large for ever-married women, and negligible for never-married women.

            * I would like to thank seminar participants at the University of Chicago, Harvard University, the Institute for Research on Poverty, Massachusetts Institute of Technology, RAND Corporation, University of California - Berkeley, University of California - Los Angeles, University of California - Santa Barbara, University of California - Santa Cruz, and Yale University for helpful comments. Lawrence Katz and an anonymous referee provided extremely useful comments which led to a substantially improved paper. In addition, Janet Currie, David Cutler, Peter Diamond, Leora Friedberg, Jerry Hausman, Caroline Minter Hoxby, Hilary Hoynes, Mark McClellan, Lucia Nixon, Joseph Newhouse, Steven Pischke, Andrew Samwick, Douglas Staiger, Duncan Thomas, and especially Jonathan Gruber, Brigitte Madrian, and James Poterba provided insightful comments. I would like to thank the National Bureau of Economic Research Health and Aging group for use of their excellent computing facilities and the National Institute of Aging for financial support.

I. Introduction

            The United States welfare system for single parent families with children offers four main benefits: cash assistance through Aid to Families with Dependent Children (AFDC), health insurance coverage through Medicaid, food subsidies through Food Stamps, and public housing. In the past decade Medicaid benefits have become more important as medical costs have soared while cash benefits have failed to keep up with inflation. In fiscal year 1991, the combined Federal-State Medicaid expenditure of $21.9 billion on 12.6 million AFDC recipients exceeded cash payments of $20.9 billion to this group [U.S. House of Representatives 1993]. This paper investigates the hypothesis that losing Medicaid coverage is a deterrent to leaving welfare. If this is true, then proposals which extend health insurance coverage to children in low- and moderate- income families could have the additional effect of getting families off the welfare rolls.

            Medicaid provides a basic set of free or subsidized medical services to poor, eligible families. The program is federally subsidized and regulated but administered by the states, which have some leeway in defining the set of services offered. Traditionally, eligibility for Medicaid has been contingent on eligibility for AFDC -- that is, one simultaneously qualifies for Medicaid and AFDC by having net income under a state's income eligibility limit. The health insurance is retained as long as the AFDC recipient earns less than the "AFDC break-even level," the point where AFDC benefits are lost. Medicaid is entirely lost once earned income goes beyond the break-even level, generating a marginal tax rate in excess of 100 percent.

            This paper will explore some recent Medicaid expansions for children that explicitly sever the link between Medicaid eligibility and AFDC eligibility and generate sizable exogenous shocks to the budget set of some welfare recipients. Using recent data from the Current Population Survey (CPS), I offer new evidence on Medicaid's impact on labor force and AFDC participation. I find that the fully phased-in Medicaid reforms reduced the probability of participating in AFDC by 1.2 percentage points and increased the probability of working in the labor force by 0.9 percentage points. The effects of the reforms were quite strong among divorced and separated women, but not among never-married women.

            The paper is arranged as follows: Section II sets up the theoretical framework to analyze Medicaid's effect. Section III describes the legislative changes used to identify Medicaid's effect. Section IV describes the data extract, various years of the March CPS. Section V provides reduced form evidence of Medicaid's impact on labor force participation and AFDC participation. Section VI concludes.

II. Theoretical Effects of Medicaid

            To analyze the effect of Medicaid on labor supply and welfare participation among potential welfare participants, I use a variant of the static labor supply model, which incorporates taxes and the welfare system. Assume that the consumer maximizes utility, U=u(Leisure, Other Goods). She faces a constant pre-tax wage, w⁰. The welfare and tax system create non-linearities in the budget set. At zero hours of market work, the mother receives a certain level of AFDC and food stamp benefits, known as the "guarantee" in addition to Medicaid. Figure I illustrates this budget set for a welfare recipient in Pennsylvania in 1989. As she begins to work, her AFDC and food stamp benefits are taxed away, so her after tax wage is w¹=(1-τ_AFDC)w⁰, where τ_AFDC is the marginal tax rate for earning income while on AFDC (which varies from 67 to 100 percent). Once she works more than H_Breakeven, the hours of work where the entire AFDC benefit is taxed away, she loses her AFDC eligibility, and therefore her Medicaid benefits, which creates a dominated part of the budget set. This discontinuous drop in benefits has been called the "Medicaid notch." Once the recipient works more than this level of hours, she faces an after tax wage of w²=(1-τ_FED)w⁰, where τ_FED is the marginal tax rate in the federal and state income tax codes. To determine what region of hours is dominated, however, we would need to know the value of the benefits she received from Medicaid.

            As states have expanded eligibility for Medicaid by increasing the income limit to a higher level than what an AFDC recipient could earn, the notch has moved. The coverage is still means-tested, but at a potentially higher level than the AFDC income limit. The change in the income limit yields several reduced form predictions for those eligible for the expansions by changing the budget set as illustrated in Figures I and II.

● Labor force participation increases (or remains unchanged if no behavioral response occurs), since the new opportunities on the budget set occur where the woman participates. This occurs because some women who were not initially working before the expansion begin to work. No one who is currently working should withdraw from the labor force, because that (Leisure,Other Goods) combination was available before the expansion.

● AFDC participation decreases (or remains unchanged if no behavioral response occurs), because the only new opportunities are where the woman leaves AFDC.

● AFDC participation decreases more than labor force participation increases. This occurs since some women will be located along the welfare part of the budget set (but not at zero hours of work) before the expansions, implying participation in both AFDC and the labor force. After the expansions these people could increase their hours and locate on the post-expansion part of the budget set, which we observe as exiting AFDC but having no effect on labor force participation. For women initially at zero hours of work, the two effects should be the same, since the only new opportunity the expansion offers is to exit AFDC and enter the labor force. For women initially off welfare, their hours may decrease, but they will not participate in AFDC, which they could have already done. Therefore, in aggregate, the effect on AFDC participation is larger than the effect on labor force participation.

● The effect on total hours of work is ambiguous. Hours could increase for women initially on AFDC, but could decrease for women initially off AFDC.

            Two other points deserve mention. First, the existence of the Medicaid "notch" is predicated upon no feasible insurance alternatives for these households. It may be reasonable to think that with the skill mix that the group possesses, the possibility of employer provided health insurance coverage is quite low. Second, welfare stigma potentially provides a second explanation for why the effect of extending Medicaid should be smaller on entry into the labor force than exits from AFDC. Households that previously worked because they found collecting AFDC very stigmatizing might withdraw from the labor force when given the opportunity to receive Medicaid without being on AFDC.

III. Legislative History and Identification

A. Overview of Legislation from 1986 to 1990

            The Medicaid expansions were legislated in response to the Omnibus Budget Reconciliation Act (OBRA) 1981, which severely reduced access to health care services for the poor by placing heavy restrictions on AFDC eligibility. As a result, nearly 40 percent of working AFDC families were removed from the welfare rolls and roughly two million people (nearly 500,000 families) lost Medicaid eligibility between 1979 and 1983. Starting in 1984, and especially from 1986 onward, Congress attempted to increase access to health care for pregnant women, infants and children through a series of Medicaid expansions. The legislation severed the link between Medicaid eligibility and AFDC eligibility by eliminating Medicaid eligibility criteria related to family structure and an individual state's AFDC income eligibility level.

            The legislation allowed Medicaid coverage to be means-tested to some percentage of the Federal poverty line (FPL), usually 100 or 133 percent. As illustrated in figure II, a Medicaid expansion to 133 percent of FPL would move the Medicaid notch from $9,000 to $13,380. Before the expansion, this household loses Medicaid between $9,000 and $10,000 of gross earnings, and total income does not reach this level again until gross earnings nearly double.      Several pieces of legislation expanded access to health care for children. Legislation in 1986 and 1987 gave states several options for expanding their Medicaid program, while legislation in 1989 and 1990 mandated more extensive coverage. Table I illustrates the generosity of the expansions at the beginning, middle and end of the period that I analyze, by showing the age limit to qualify for Medicaid and the Medicaid income eligibility limit for an infant (the income limit for older children was usually lower). The earliest legislation gave states the option to carry out the expansions to children under two. The table shows that by January 1988, half the states had expanded eligibility. By the end of 1989, every state had adopted some form of expansion and there was a great deal of variation in Medicaid eligibility across states based on the age of the child. The later mandates increased the income threshold to 133 percent and the age limit to six. Thirty-two states were required to adjust their income threshold and thirty-seven states were forced to increase their age limit. Finally, the mandates in 1991 expanded eligibility to children over the age of six to 100 percent of the poverty line. By December 1991, all states extended Medicaid coverage to children up to age eight, though the income eligibility limits for infants varied substantially.

            These reforms resulted in a dramatic increase in Medicaid eligibility and coverage. Figure III [Medicaid Source Book: Background Data and Analysis 1993] shows a sharp rise in the number of children covered by the Medicaid expansions (beneficiaries without cash assistance) starting in 1988, whereas the number of children enrolled in the Medically Needy program and AFDC program remained quite stable. This data, based on administrative records from the Health Care Financing Administration, shows that by 1991, three million children were covered by Medicaid resulting from the expansions. In addition, Currie and Gruber [1994] find that between 1984 and 1992, eligibility for Medicaid increased by 100 percent. By the end of the period, one-third of all children were eligible for Medicaid. They find Medicaid coverage from the expansions was flat until 1988, and rose steeply thereafter. This dramatic increase in coverage helps to motivate looking at other possible consequences of the reforms, such as the effects on labor supply and AFDC participation.

B. Parameterization of Health Insurance Expansions

            The reforms in Medicaid boil down to changing the income limit for eligibility (which I shall measure as a percentage of the poverty line). Figure II illustrates the measure I incorporate: how much the Medicaid income limit is increased from the reforms over its previous AFDC level. This is formulated as:

(1)       GAIN% = max(MEDICAID% - AFDC%,0).

GAIN% parameterizes how drastically severed the tie is between Medicaid eligibility and AFDC eligibility. MEDICAID% is the income eligibility limit from the Medicaid reforms, while AFDC% is the income limit from AFDC. GAIN% can be equal to zero for two reasons. First, the child may be in a household that is ineligible for the Medicaid expansion (based on the child's age), so that the link between AFDC and Medicaid is not severed. Second, the household might be eligible for the expansion, but MEDICAID% may be less than AFDC%. After the appropriate deductions and tax rates are accounted for, the income level that a recipient can earn from AFDC% can exceed 100 percent of the FPL. GAIN% explicitly models the fact that the expansions should have less impact in a generous AFDC state since the recipient could have worked and earned some amount of money before the expansion.

            Measuring MEDICAID% is straightforward: it is equal to zero if the household is ineligible for the expansion, while it is equal to some percentage of the FPL (typically 75, 100, 133 or 185 percent) if the recipient qualifies for the expansion. AFDC% depends on a state's need and payment standard, which are two state-specific income limits used to determine AFDC eligibility. The information used to compute AFDC% changes over time (through changes in the need and payment standards) as well as through changes in the ordering of the certain deductions for welfare payments, known as disregards. It also varies by family due to family size, work expense deductions and child care deductions. Furthermore, the Family Support Act of 1988 (FSA) affected the calculation of the AFDC income limit in several ways. After incorporating this detail, AFDC% is calculated as:

(2)       AFDC% = [(PAYMENT/BRR) + DISREGARD + WORKEXP + DAYCARE]/POV

where PAYMENT stands for the state's payment standard (one of the income limits used to determine AFDC eligibility), BRR for the benefit reduction rate (which is 2/3 for the first four months of work), DISREGARD for the standard deduction, WORKEXP for work expenses, DAYCARE for total child-care deductions, POV for the dollar amount of the FPL (appropriately adjusted for family size, and inflated by 1.15 for Hawaii and 1.25 for Alaska). DAYCARE expenses are calculated as:

(3)       DAYCARE = (1 + ½*↿(post-FSA))*DEDUCTION*CHILDREN

where ↿(●) is defined as an indicator variable equal to one after the passage of FSA and zero otherwise, DEDUCTION is the child care expense deduction while on AFDC, and CHILDREN is the number of children in the family.

C. Econometric Model and Identification Strategy

            The primary evidence I present on the effect of the Medicaid notch comes from probits that model labor force and AFDC participation. The model is specified as:

(4)       LFP_ikjt^* = β₀ + β₁GAIN%_ikjt + β₂X_i + β₃TIME_t + β₄STATE_j + β₅KIDAGE_k + ε_ikjt

where (4) is the underlying index function for the probit (and i indexes mothers, k indexes the youngest child's age, j indexes states and t indexes time). A similar equation is used with AFDC^* on the left-hand side. Here, GAIN% is the independent variable of primary interest (with β₁ hypothesized to be positive here, and negative for AFDC participation). X_i contains other covariates, including the number of children under age six, mother's age and its square (divided by 100), mother's education and its square, and dummy variables for family size, black, divorced, separated and residence in a central city. Dummy variables for time, state and youngest child's age are represented by TIME, STATE and KIDAGE. In practice we do not observe the underlying value LFP_ikjt^*, but instead only observe the discrete outcome:

(5)       LFP_ikjt =         1 if LFP_ikjt^*≥0

                                    0 if LFP_ikjt^*<0.

Assuming ε_ikjt∼N(0,1) and denoting Φ(●) as the cumulative normal function gives the following probability:

(6)       Prob(LFP_ikjt) = Φ(β₀ + β₁GAIN%_ikjt + β₂X_i + β₃TIME_t + β₄STATE_j + β₅KIDAGE_k).

            The reforms in Medicaid create three dimensions of variation that I exploit to identify Medicaid's effect. The laws create variation in the budget constraint for mothers with children of different ages within a state, across states, and over time. The most intriguing dimension is within state. The expansions provided health insurance coverage to a young child that was conditioned on his or her birthday. By conditioning on the child's birthday, the expansions create "treatment" and "control" groups to gauge the effects of moving the income eligibility limit for Medicaid to a higher level. The treatment group is families with children fortunate enough to be born after the birthday cutoff, while the control group is families with children who were born before the birthday cutoff. To illustrate, consider a state that carried out the OBRA 1987 provisions to the maximum extent possible, as soon as possible. Then, the following mothers with children of different birthdays get the following "treatment" on July 1, 1988 in that state.

            Besides variation in the eligibility and length of coverage, the laws generate variation in the income limit where the recipient loses coverage for her children. For instance, after July 1, 1991, the mother could face the following earnings schedule for losing Medicaid coverage (conditional on her children being eligible for the expansions):

In other words, a mother with a five year old can earn up to 133 percent of the FPL before losing Medicaid while a mother with a six year old can only earn up to 100 percent of the FPL.

            The expansions also created variation across states and over time, because different states carried out the expansions at different times. Examining differences in AFDC and labor force participation across states or over time may not be entirely convincing, however, since other events might occur simultaneously. For example, changes in macroeconomic conditions over time or variation in economic conditions across states could offer incentives to participate in AFDC independently of the Medicaid expansions. By including dummy variables for state, time and youngest child's age in equation (6), we account for some of these unmodelled stories and obtain the "difference-in-differences" estimator. While it is hard to indict this estimator, interactions between STATE, TIME and KIDAGE could still drive the observational difference. For example, changing economic conditions may have affected mothers with older children more than mothers with younger children. Also, some states rebounded from the recession more quickly than others. If these stories are driving the differences in the outcomes, then we can include second order interactions of STATE*TIME, KIDAGE*TIME and STATE*KIDAGE to gauge the impact of Medicaid. By including these interactions in equation (6), we obtain the "difference-in-difference-in-differences" estimator.

            This identification strategy represents an important departure from previous work which has tried to assess Medicaid's effect. Blank [1989], Winkler [1991] and Moffitt and Wolfe [1992] tried to identify Medicaid's effect by valuing health insurance because the tie between AFDC and Medicaid was much stronger for the periods they examined. Valuing an in-kind benefit such as health insurance is a daunting task, because Medicaid's value should incorporate health status, risk aversion, scope of medical services offered, access to care, insurance copayments and deductibles. In addition, some studies have used health status to value Medicaid, but this approach likely attributes changes the potential wage and changes in preferences towards work to the value of Medicaid.

IV. The Data Set

            The data set, which consists of repeated cross sections, was constructed using the March Current Population Survey (CPS), from the years 1989 through 1992. These years covered the period when the Medicaid expansions occurred. The CPS is a timely, nationally representative survey interviewing many households (approximately 57,000 per month). Its March Annual Demographic file contains retrospective information on labor force participation and welfare participation. The sample contains 16,062 single mothers between the ages of 18 and 55 with at least one child under 15 present. I use a smaller range of children's ages (only up to 14) than previous studies (usually up to 18) for two primary reasons. First, during the four year window that my data spans, 1988 to 1991, the expansions never affected children over age eight, so using children up to age fourteen should provide an adequate control for within state variation in the benefit schedule. Second, and more importantly, I was concerned with the possibility that older teenage children may form their own families and collect welfare benefits independently of their mother, so that modeling the joint labor supply decision might be appropriate when older children are present.

            To each mother's record I linked the youngest child's age, which is used to impute eligibility and generosity of the expansions (along with time and state of residence). I therefore compare labor market outcomes of mothers with any child eligible to mothers with no child eligible. The data concerning the Medicaid expansions was compiled from documentation provided by the Intergovernmental Health Policy Project that contains detailed information on the date of implementation, range of ages the expansions covered, the new Medicaid notch, and any phase in schedule for the expansion.

            To assign eligibility it was necessary to impute a birth month and birth year to each child, since the CPS only asks the child's age as of March 1 of the survey year. To do this, I assigned each child a birth month, randomly drawn from the year in which they could have been born, based on a uniform distribution. This random assignment is a compromise because it induces measurement error in the independent variable, GAIN%. This measurement error is more important for children born in the year of some birthday cutoff, who then have a chance of being misclassified, while it is less important for children born more than one year above or below the cutoff date.

            Table II shows the means of the variables used in the estimation. Forty percent of children in these households collect Medicaid, while a smaller fraction of households, 32 percent, participate in AFDC. The labor force participation rate for the mothers is 68 percent, and roughly half of those working, 36 percent, are covered by employer provided health insurance. While the demographic makeup stays fairly stable across the years (not shown), columns (2), (3) and (4) show that there are observable differences between never married, divorced and separated mothers. I therefore include dummy variables for marital status in all specifications. The never married women are younger and more likely to participate in welfare than divorced or separated women. Slightly less than one-quarter of the mothers in the sample did not complete high school, while slightly less than one-third of the sample attended college. The table also shows that 42 percent of the sample was eligible for the Medicaid expansion based on the state Medicaid rules, time period, and age of the youngest child (but not family income). The takeup rate was 29 percent among all newly eligible families, and was 47 percent among newly eligible families that lacked employer provided health insurance. These rates are similar to the findings of Currie and Gruber [1994b].

            Finally, Table III shows the independent variable of policy interest, GAIN%. It is positive for approximately 16 percent of the sample (compared to the 42 percent who have children who are eligible for some expansion coverage). Conditional on GAIN% being positive, its mean is 20.8 percent. When the expansions have "bite," the notch is moved up by 20.8 percent of the FPL, on average. The maximum GAIN% is 85 percent, which helps show that the AFDC% can be quite important in reducing the generosity of the expansion. This table also shows how GAIN% evolved over time. The fact that the expansions became more generous is illustrated through the average movement in the income eligibility limit increasing more than six-fold between 1988 and 1991. In addition, different regions of the country had different responses to the expansions. Since the South tended to offer less generous AFDC benefits, it should not be surprising that the expansions had the most impact there. Finally many of the women in the sample had no change in their budget set resulting from the expansions. This should be kept in mind when evaluating the actual changes in labor force and AFDC participation during the time span.

V. CPS Results

A. Primary Specification: All States

            Table IV presents the primary evidence of the effect of the Medicaid notch on labor force and AFDC participation, through the coefficient estimate on GAIN%. In these specifications decoupling Medicaid from AFDC results in a significant positive effect on labor force participation and a significant negative effect on AFDC participation. The estimated effect of the fully phased-in Medicaid expansion is calculated by comparing the actual labor force and AFDC participation in 1991 for the sample (66.5 and 34.8 percent respectively) to the predicted rate when GAIN% is set equal to zero and evaluated for the 1991 observations (yielding predicted rates of 65.6 and 36.0 percent for the models that include STATE*TIME effects). Interestingly, this exercise shows that the actual change in the probability of labor force participation was small: around nine-tenths of one percent, which translates into an increase of 1.4 percent in the labor force pool. The fully phased-in Medicaid reforms reduced the AFDC caseload by approximately 3.5 percent. This effect is dominated in the aggregate caseload numbers by the effect of the recession and other factors. Between 1988 and 1991, the average number of families that participated in AFDC each month increased 17 percent, from 3.74 million to 4.37 million [U.S. House of Representatives 1993]. The reason that the fully phased-in expansions did not generate more exits from the welfare rolls is that only 29 percent of the 1991 observations had their budget constraint changed. The remaining 71 percent either had a child who was too old for the expansion or lived in a generous AFDC state.

            The other covariates enter as expected: the number of children under age six, the square of the mother's age, the black and central city indicators usually enter the labor force participation equation as negative and statistically significant. The mother's age, divorced, separated, and the square of education enter into the labor force participation equation as positive and significant. These covariates are similar in significance to Winkler [1991], who uses an earlier year of the CPS. When I evaluated the marginal effect of these coefficients, the magnitudes were similar to the results from the linear probability model in column (1) of table VII. In particular, mother's age, education and education squared all increased labor force participation by less than two percentage points. Being divorced or separated (relative to never-married), having one less child under the age of six, and not living in a central city all increased the probability of labor force participation by more than five percentage points.

            Table V simulates the change in probability from different policy changes (in percentage points) generated by the models that include STATE*TIME interactions in the second and fourth columns in table IV. Notice that severing the link to AFDC eligibility has stronger effect on reducing AFDC participation than increasing labor force participation. As shown in section 2, this is an implication of the model since the work and welfare decisions are not mutually exclusive: some women who were on AFDC and working will now leave AFDC, which shows up as no change in labor force participation.

            The first row evaluates the effect of increasing GAIN% by 25 percent of the FPL over its current level. This simulation, applied to the whole sample, increases the probability of labor force participation by 3.3 percentage points (an 4.9 increase in the labor force pool) and decreases the probability of AFDC participation by 4.6 percentage points (a reduction in the AFDC caseload of 14 percent). The second row presents the estimated change in participation among those for whom the Medicaid expansions changed the budget constraint, meaning that GAIN% is positive. Among this group (which is 16 percent of the CPS sample), the expansions increased the probability of working by 3.0 percentage points and decreased the probability of welfare participation by 4.5 percentage points. If these movements were applied to all female headed households, they would translate into a 14 percent reduction in the AFDC caseload and into a 4.4 percent increase in the labor force pool. The third and fourth rows, applied to the 1991 observations only, show the result of two final policy changes: means-testing Medicaid at $20,000 and $15,000, instead of at the recipient's current 1991 AFDC income limit. If these policy changes were applied to all female headed households, they imply reductions in the AFDC caseload of 22 and 6.6 percent, respectively. Since many recipients did not, in fact, have their budget constraint changed, these simulations suggest that there is substantial room for additional reform.

            In relation to previous work, the movements from the fully phased-in expansions are larger than the effects found by either Blank [1989] or Winkler [1991] (who find very small effects), but less than the effects found by Moffitt and Wolfe [1992]. This is not surprising, because the variation that both Blank and Winkler used, average expenditure, may not capture much of the value of Medicaid to any specific family, while the health measures used by Moffitt and Wolfe could partly attribute changes in preferences and changes in the wage that affect labor supply to Medicaid.

            Finally, to qualify for AFDC, a recipient faces a benefit reduction rate of 67 percent during the first four months of work and 100 percent thereafter. Therefore, I have also attempted to use a benefit reduction rate of 100 percent in constructing GAIN%. While the substantive findings remain unchanged, the estimated effect from different policy reforms of moving the Medicaid notch falls by approximately 40 percent. It turns out that using the benefit reduction rate of 67 percent is more appropriate to model the income limit where Medicaid is lost, however. The Deficit Reduction Act of 1984 allowed Medicaid coverage for an additional nine to fifteen months to former AFDC recipients who were disqualified for AFDC (and would therefore lose Medicaid) when the benefit reduction rate increased from 67 percent to 100 percent.

B. Effects of the Expansions on Different Demographic Groups

            In targeting a benefit such as Medicaid, it is important to know which demographic groups respond to the expansions. This section attempts to ask whether GAIN% has a different impact based on marital status, educational attainment and age. First, I reestimate the models by marital status. A priori, there are several competing explanations on which type of single woman should have a larger response to the expansion. One hypothesis is that GAIN% should be weaker for ever married women than for never married since the former are more likely to have health insurance coverage through the absent father. Only 6.3 percent of never married women have health insurance coverage through an absent father, compared to 22 percent of separated women and 34 percent of divorced women [U.S. House of Representatives 1994]. Since ever married women are more likely to have health insurance coverage, then moving the Medicaid notch might be less important for them. On the other hand, being both never married and a mother could proxy for long-term welfare dependency. If this is the case, then one could reasonably expect the coefficient on GAIN% for never married women to be smaller because it takes more drastic changes in the budget constraint to remove them from welfare. Being never married could also proxy for lower earnings prospects and lower levels of non-labor income. This too could lead to a smaller effect because the never-married mother might not even be able to attain a level of earnings to remove herself from AFDC. Thus moving the Medicaid notch may not matter, because the notch was above the maximum earnings level she could attain from full time work.

            First, I stratify the sample by marital status, and reestimate the participation equations for the models that include STATE*TIME interactions. Stratifying seems reasonable, because there are several plausible stories for why the independent variables should affect labor force and AFDC participation differently based on marital status. For instance, being both black and divorced may signal something different about a woman's outside source of income than being black and never married, white and divorced or white and never married.

            Table VI shows that the results for labor force participation are dramatic: the fully phased-in expansions had strong effects on ever married women, and no effect on never married women. The labor force pool increased 1.6 percent, and the AFDC caseload fell by 4.6 percent for ever married women. On the other hand, the coefficient on GAIN% is insignificant and negative for never married women's labor force participation. In addition, the AFDC caseload fell by only 1.7 percent. If the interpretation of welfare dependency is correct, these findings suggest that the health insurance reform might be a useful avenue off of welfare for short-term participants, but would have little impact on long-term participants.

            The findings by stratifying on education and age are also consistent with the above explanation concerning welfare dependency. When I stratify on educational attainment (less than high school, some high school, completed high school and more than high school), I find that the expansions have little effect on those with less than a high school diploma. Low levels of education are likely to be correlated with long-term welfare dependence. The expansions have a large effect on those with a high school diploma, and little effect on those with more than a high school education. The results of stratifying by age groups (less than 25, ages 25 to 29, ages 30 to 34 and over age 34) show weaker effects of the expansions on the youngest women.

C. Difference-in-difference-in-differences Specification

            To gauge the importance of omitting KIDAGE*STATE interactions in the models presented in previous tables, table VII presents the results of a linear probability model that includes KIDAGE*STATE interactions. One source of bias by omitting KIDAGE*STATE interaction could be the effect of child care arrangements or child care prices on labor force and AFDC participation. Clearly child care affects the labor force participation of mothers with young children differently than mothers with older children, and this market could vary considerably by state. By including STATE*TIME and KIDAGE*TIME interactions, columns (1) and (3) are the analog to column (2) and (4) in table IV. I also include family size, number of children under age six, mother's age and its square, mother's education and its square, and dummy variable for black, central city, divorced and separated in all specifications. Columns (2) and (4) build upon these models by including KIDAGE*STATE interactions, which corresponds to the difference-in-difference-in-differences (DDD) estimator. Columns (1) and (3) tell much the same story as the previous probit models: moving the Medicaid income eligibility limit continues to have a significant positive effect on labor force participation and a significant negative effect on AFDC participation. Moving from column (1) to (2), and from (3) to (4), the coefficient on GAIN% gets stronger from including the KIDAGE*STATE interactions. In particular, the coefficient on labor force participation increases considerably.

VII. Concluding Remarks

            This paper has shown that decoupling Medicaid eligibility from AFDC eligibility significantly affects labor supply and welfare participation. Using recent Medicaid health insurance expansions for children that severed the link to AFDC, I show that a substantial decreasing in the AFDC caseload could occur from completely severing the link between AFDC and Medicaid eligibility. The recent law changes reduced the probability of AFDC participation by 1.2 percentage points. The results also show that means-testing Medicaid $15,000 or $20,000 instead of the AFDC limit (which was $7,440 in the average state in 1991) could result more substantial outflows from AFDC. While my findings would suggest that the AFDC caseload should have decreased, the caseload actually increased. At least two factors are responsible for this. First, worsening economic conditions at the state- and national- levels increased the welfare caseload. Second, not all families were eligible for the reforms. Moreover, even among those who were eligible, the budget constraint might not have been drastically changed. In the aggregate numbers, these factors mask the importance of health insurance in the decision to leave welfare.

            These findings have consequences for the cost of health care reform. Expanding eligibility for Medicaid could result in reduced expenditure for current welfare recipients by encouraging them to exit AFDC. These exits could reduce current AFDC expenditure and result in some growth in the taxable base due to increased hours of work. Several notes of caution are appropriate, however. While reductions in the welfare rolls may be possible from severing the link between Medicaid and AFDC, such an expansion presents new work disincentives for households initially off of welfare (for both single- and dual- earner families). This could discourage work and reduce the taxable base. More importantly, the government would be responsible for paying the health care costs for newly eligible children.

            The reforms in Medicaid during the 1980s had consequences on three other outcomes as well. First, the reforms severed the link to another AFDC requirement, family structure. Yelowitz [1995] finds that severing this link encourages marriage. He also finds outflows from marriage, however, by changing the single woman's budget constraint. Second, the expansion of health insurance for both pregnant women and for children could affect child health. Currie and Gruber [1994a, 1994b] find that a 20 percentage point increase in eligibility for women of child bearing age was associated with a decrease in infant mortality of 7 percent. Among children, extending Medicaid was associated with large increases in care for children delivered at physician's offices. Third, the reforms might have had an effect on the demand for private insurance. Cutler and Gruber [1995] find evidence that public insurance crowds out private insurance.

APPENDIX A: Legislative Changes in the 1980sSixth Omnibus Budget Reconciliation Act, 1986 (SOBRA 86): Permitted States to extend Medicaid coverage to children under age 2 with incomes below 100 percent of the Federal poverty line effective April, 1987. Beginning July, 1988, States could increase the age level by one in each fiscal year until all children under age five were included.

Omnibus Budget Reconciliation Act, 1987 (OBRA 1987): Effective July, 1988, States could immediately cover children under age 5 (rather than phasing in coverage) who were born after September, 1983. Effective October, 1988, States can expand coverage to children under age 8. Allowed states to extend Medicaid eligibility for infants up to 185 percent of the federal poverty level.

Medicare Catastrophic Coverage Act, 1988 (MCCA 88): Required states to cover infants on a phased-in schedule: to 75 percent of the federal poverty level, effective July, 1989 and to 100 percent effective July, 1990.

Family Support Act, 1988 (FSA 88): Effective April, 1990, required states to continue Medicaid coverage for 12 months for families who received AFDC in three of the previous six months, but became ineligible for assistance because of increased earnings. Families whose incomes exceeded 185 percent of the federal poverty level would not qualify. Families incomes between 100 and 185 percent of the poverty guidelines could be charged a premium during the second six months.

Omnibus Budget Reconciliation Act, 1989 (OBRA 89): Required states to extend Medicaid coverage to all children under age six with family incomes up to 133 percent of the federal poverty level. Effective April, 1990.

Omnibus Budget Reconciliation Act, 1990 (OBRA 90): Starting July, 1991, States are required to cover all children under age 19, who were born after September, 1983, to 100 percent of the Federal poverty level.

APPENDIX B

Construction of AFDC% Variable

                  This appendix explains the construction of AFDC%. This example follows the U.S. House of Representatives [1993]. Under AFDC, each State establishes a "need standard" and a "payment standard" (which may equal to or lower than the need standard); these standards are adjusted by family size. To receive AFDC payments, a family must pass two income tests, a gross income test and a countable income test. Families with gross incomes that exceed 185 percent of the State's need standard are ineligible for AFDC. Benefits are generally computed by subtracting countable income (i.e. gross income less certain amounts known as disregards) from the payment standard. The maximum benefit, which is the amount paid to a family with no other income, may be lower than the payment standard; as of January 1, 1992 this was true for 9 states.

The gross income test is

(A1) Gross Income ≤ (1.85*Need Standard).

To be eligible for an actual payment, the family's counted income also be below the State's payment standard. Counted ("net") income test:

(A2) Net Income = (Gross Income-Deductions) ≤ Payment Standard,

which implies,

(A3) Gross Income ≤ (Payment Standard+Deductions).

AFDC% is then determined by minimum of equations (A1) and (A3):

(A4) AFDC%=min{1.85*Need Standard, Payment Standard+Deductions}/POV

where POV stands for the poverty limit in dollars. Equation (A3) is further complicated because the treatment of deductions has changed over time, and varies by the number of months that a recipient has been off of welfare. To illustrate this, the following table taken from the U.S. House of Representatives [1993], illustrates the calculation on an AFDC payment during different regimes. To calculate the Medicaid notch, we would find the level of gross earnings such that net countable income is zero.

Calculation of Monthly AFDC Benefits for a Worker with Low Earnings Under DEFRA and Current Law

Consider the column containing information on the first 4 months after the Family Support Act is enacted. To calculate the income level where Medicaid eligibility is lost (based on the payment standard), we find the level of net countable income that equals the payment standard.

(A5) (Gross Income-Initial Disregards) - 1/3*(Gross Income-Initial Disregards)

               - Childcare = Payment Standard

Solving this in terms of Gross Income gives us:

(A6) Gross Income = 3/2*(Payment Standard+Childcare) + Initial Disregards.

              For the $680 payment standard, and $100 childcare costs, the Medicaid notch is $1,290 per month based on the payment standard. If the need standard is greater than $697 (=1290/1.85) then the payment standard binds. Otherwise the Medicaid notch is simply given by equation (A1). This corresponds to equation (2) in the text. One could perform this same exercise for first 4 months after the Deficit Reduction Act.

              The need and payment standards vary by state, time and family size. The information on the need and payment standards (as of July 1 of each year) was obtained from documents provided by the National Governors Association for the years 1988 to 1991.

UNIVERSITY OF CALIFORNIA, LOS ANGELES

REFERENCES

Blank, Rebecca, "The Effect of Medical Need and Medicaid on AFDC Participation," Journal of Human Resources, XXIV(1989), 54-87.

Currie, Janet, and Jonathan Gruber, "Saving Babies: The Efficacy and Cost of Recent Expansions of Medicaid Eligibility for Pregnant Women," NBER Working Paper No. 4644, 1994a.

Currie, Janet, and Jonathan Gruber, "Health Insurance Eligibility, Utilization of Medical Care, and Child Health," Mimeo, University of California, Los Angeles, 1994b.

Cutler, David, and Jonathan Gruber, "Does Public Insurance Crowd Out Private Insurance?" Mimeo, Harvard University, 1995.

Ellwood, David T., and Mary Jo Bane. "The Impact of AFDC on Family Structure and Living Arrangements." Research in Labor Economics, Ronald G. Ehrenberg, Ed. Vol. 7 Greenwich, CT: JAI Press, Inc., 1985.

Gruber, Jonathan, "The Incidence of Mandated Maternity Benefits," American Economic Review, LXXXIV(1994), 622-641.

Gruber, Jonathan, and James Poterba, "Tax Incentives and the Decision to Purchase Health Insurance: Evidence from the Self-Employed," Quarterly Journal of Economics, CIX(August 1994), 701-733.

Intergovernmental Health Policy Project, "Major Changes in State Medicaid and Indigent Care Programs." Debra J. Lipson, Rhona S. Fisher and Constance Thomas, Eds., The George Washington University, various editions.

Krueger, Alan and Stephen Pischke, "The Effect of Social Security on Labor Supply: A Cohort Analysis of the Notch Generation," Journal of Labor Economics, X(1992), 412-437.

Moffitt, Robert, "Incentive Effects of the U.S. Welfare System: A Review," Journal of Economic Literature, XV(1992), 1-61.

Moffitt, Robert, and Barbara Wolfe, "The Effect of the Medicaid Program on Welfare Participation and Labor Supply," Review of Economics and Statistics, LXXIV(1992), 615-626.

Short, Pamela, Joel Cantor, and Alan Monheit, "The Dynamics of Medicaid Enrollment," Inquiry, XXV(1988), 504-516.

U.S. House of Representatives. Medicaid Source Book: Background Data and Analysis (A 1993 Update). Washington: Government Printing Office, January 1993.

U.S. House of Representatives. Background Material and Data on Programs within the Jurisdiction of the Committee on Ways and Means. Washington: Government Printing Office, various years.

Winkler, Anne, "The Incentive Effects of Medicaid on Women's Labor Supply," Journal Human Resources, XXVI(1991), 308-337.

Yelowitz, Aaron, "Will Extending Medicaid to Two Parent Families Encourage Marriage?" Mimeo, University of California, Los Angeles, 1995.

TABLE I

State Medicaid Eligibility Thresholds for Children

TABLE II

Summary Statistics

TABLE III

TABLE IV

TABLE V

TABLE VI

TABLE VII