Economic Research Initiative on the Uninsured Working Paper Series

Transcription

1 Economic Research Initiative on the Uninsured Working Paper Series EXTENDING HEALTH CARE COVERAGE TO THE LOW- INCOME POPULATION: THE INFLUENCE OF THE WISCONSIN BADGERCARE PROGRAM ON INSURANCE COVERAGE AND LABOR FORCE PARTICIPATION Barbara Wolfe University of Wisconsin-Madison Thomas Kaplan University of Wisconsin-Madison Robert Haveman University of Wisconsin-Madison CONFERENCE DRAFT Economic Research Initiative on the Uninsured University of Michigan 555 South Forest Street, 3 rd Floor Ann Arbor, MI Not to be distributed or copied without permission of the authors. July 2003

2 Extending Health Care Coverage to the Low-Income Population: The Influence of the Wisconsin BadgerCare Program on Insurance Coverage and Labor Force Participation I. Introduction Barbara Wolfe, Thomas Kaplan, and Robert Haveman 1 Institute for Research on Poverty University of Wisconsin Madison Soon after implementing the Wisconsin Works (W-2) welfare reform program in September 1997, Wisconsin also began its State Children s Health Insurance Program (SCHIP) program. The Wisconsin SCHIP program, called BadgerCare, started operation in July Unlike SCHIP programs in most other states, the BadgerCare health insurance program provides income-conditioned health insurance to both adults and children in low-income families with minor children. It also substantially expands the Medicaid program by providing Medicaidequivalent health care coverage for both children and adults with incomes below 185 percent of the federal poverty line (200 percent for those already enrolled). As such, the program forms a model that has been studied by several other states, and considered at the federal level. Supporters of BadgerCare promoted it for two key reasons: The program offered a reduction in the uninsured population, which is heavily concentrated among the low- income population, and The program offered health insurance support to low-income families, including welfare leavers, designed to assist them in finding and retaining employment in jobs that may not provide health insurance coverage. 2 In this paper, we analyze the extent to which BadgerCare (including its expansion of Medicaid coverage) was successful in meeting these objectives. In particular, we attempt to address the following two questions: 1. Does a public program with the characteristics of Wisconsin s BadgerCare reduce the proportion of the low income population without health care coverage? 1 The authors are listed in reverse alphabetical order; all contributed equally to this research. 2 Tommy Thompson, the governor of Wisconsin at the time, said repeatedly that BadgerCare was intended as a complement to W-2 as a source of health care support for W-2 participants as they moved off of cash assistance and into work. On the State of Wisconsin web site, the program is described as follows: BadgerCare seeks to eliminate barriers to successful employment by providing a transition for families from welfare to private insurance. BadgerCare is based on the premise that health care is essential for working families with children.

3 2 2. Does a public program with the characteristics of BadgerCare increase the employment and earnings of low-income single parents who are covered by the program? 3 Our goal is to provide an answer to these questions for the state of Wisconsin, thereby pointing the way to similar research on work-oriented welfare reform and accompanying health care programs being undertaken or proposed elsewhere. II. Background on Wisconsin s welfare reform the Wisconsin Works (W-2) Program In the mid-1990s, Wisconsin, in a very striking move, virtually eliminated the AFDC cash income support welfare program, replacing it with the Wisconsin Works (W-2) program. W-2 began in September While W-2 denies the state s poor population entitlement to cash income support, it does ensure that all working-age adults with incomes at or below 115 percent of the poverty line and who are parents of minor children have the opportunity to participate in work activities, and the state supports these activities with child care assistance and subsidized health insurance. The W-2 work activities are arrayed in four tiers. In the highest tier, the most work-ready have standard jobs and are eligible for many services (including help in finding a better job, child care subsidies, and health insurance) as long as their earnings remain below eligibility thresholds. The next tier (which generally enrolls fewer than 20 people in the state at any time) provides subsidized jobs for those not quite job-ready. Participants in this tier are also eligible for the noncash benefits. The third tier includes those not ready for private-sector jobs. They are placed in community service jobs funded by the W-2 agency. Participants are paid for their community service, but they are not eligible for the EITC, since the cash they receive is viewed as a grant and not as earnings. Those least able to work are in the fourth (known as transitional job ) tier. They are required to engage in productive activities, which may take a number of forms, including caring for a disabled child or participating in alcohol or drug abuse counseling. Participants in this tier who adhere to their case plan also receive a grant. No one is eligible for assistance without working or engaging in productive activities. All program participants have a Financial and Employment Planner (or FEP) who advises them of their work options and assists them to make the best choice among these options, and also helps them find child care, transportation, and other necessities. Counseling is provided and clients are told of the job-finding support offered by the program and the child care and health care subsidy arrangements that are available. Because these activities occur in job centers that often also contain offices for technical school staff and other programs (including those funded under the 3 This second question has not been studied since the inauguration of welfare reform systems that require and assist work. Some earlier research suggests that Medicaid coverage in the context of an AFDC system with adverse work incentives may reduce the labor force participation and employment of single women with children. However, this research may not be relevant in the policy regime that has developed in Wisconsin since the inauguration of its W-2 work-oriented welfare reform, that has been in effect since For an example of the earlier research, see Moffitt and Wolfe (1992).

4 3 federal Workforce Investment Act), the client is presented with other programs for which she may be eligible, and especially of the work supports (including child care subsidies) that may be available. Participants who are in the lower two tiers of W-2 (and therefore receive a grant for work done, rather than a wage payment) are provided the same level of support regardless of family size. They are also generally told that they should expect to work full-time, full-year and that their primary source of income should be from work. This is a major change in approach from the old AFDC system, in which the main task of the case worker was to see if the family met the eligibility rules of the program (single parent status, low assets, no other income source) and determine the amount of the monthly check that would be sent to the recipient. The larger the family, the bigger the check. There was no counseling, no planning, and little emphasis on work. Wiseman (1999) identified five features at the heart of W-2. The program focuses on adults, rather than children. It denies core cash assistance to most adults if they do not work full time. For those unable to find employment in the private sector, it provides jobs and pays a grant. Eligibility for other services (health care, child care, transportation, child support enforcement) is not tied to receipt of TANF cash benefits. 4 There is an emphasis on responsibility and reciprocity. Program provision and administration are separated: non-governmental organizations are sometimes responsible for running W-2 programs. Consistent with federal government constraints incorporated in TANF, W-2 provides a cash grant to any client for a maximum of five years, unless the W-2 agency requests and the state approves an extension. Teen parents are required to live with parents who will provide them support; they are mandated to attend high school. Similarly, W-2 adopts the TANF restrictions on providing services to legal immigrants, and aggressively seeks child support from noncustodial parents, passing this support directly through to the children s caregivers. III. The BadgerCare Program The U.S. government enacted the State Children s Health Insurance Program, or SCHIP, in the Balanced Budget Act of 1997, in an effort to increase health insurance coverage among children living in low-income families who were not eligible for Medicaid. Beginning in 1998, $4.275 billion was appropriated in matching funds to states under this program. States are 4 Indeed, only roughly 15 percent of the W-2 budget now goes to cash assistance grants for work done the remainder funds services such as child care, health care, transportation, and child support enforcement.

5 4 required to contribute their own funds in order to draw on the federal resources, but the federal government pays a larger share of SCHIP costs than it does for Medicaid. In FY 1999, fewer than 2 million children were enrolled in SCHIP at any point during the year; the number increased to 3.3 million in 2000 and to about 4.6 million in States have much discretion in designing their SCHIP program. They can expand Medicaid, set up a separate program or combine the two programs; they can set eligibility thresholds, 5 decide whether to use presumptive eligibility, set requirements for the duration of time that a child must be uninsured before becoming eligible, and establish application procedures. Four states, including Wisconsin, obtained waivers that enabled them to also extend benefits to parents. Wisconsin s SCHIP program is called BadgerCare; it was implemented in the fall of Unlike cash assistance programs under AFDC and TANF, in which eligibility is conferred upon families, eligibility for Medicaid and BadgerCare is determined separately for each individual in a family unit. In addition, because the federal government reimburses a higher share of medical costs under SCHIP than under Medicaid, federal law requires states to serve individuals eligible for both programs under Medicaid rather than SCHIP. Thus, in Wisconsin, those who are eligible for coverage under Medicaid are not able to enroll in BadgerCare. BadgerCare expanded eligibility levels for public health insurance coverage in Wisconsin in the following ways: 1. Children over age 5 can enter BadgerCare with a family income up to 185 percent of the federal poverty line; the previous limit was 100 percent for children aged 6-14 and 57 percent for children aged Adults who are not pregnant and have minor children can enter BadgerCare with a family income up to 185 percent of the poverty line; the previous limit varied by family size but was approximately 57 percent of the federal poverty line. 3. Because children under age 6 and pregnant women received Medicaid coverage up to 185 percent of the poverty line before BadgerCare, the new program did not extend initial eligibility thresholds for this population. However, BadgerCare allows all who have entered the program to remain on it until their incomes exceed 200 percent of the poverty line. 5 Program designers had in mind children in families up to 200 percent of the federal poverty line (FPL), but states were free to extend benefits up to 50 percentage points beyond the state s 1997 Medicaid eligibility levels. In 2000, eligibility levels ranged from 133 to 300 percent of the FPL (GAO, 2001). Additional criteria for eligibility include not being eligible for Medicaid and being younger than age 19 (unless a waiver is granted to extend benefits to adults).

6 5 4. Families with incomes above 150 percent of the poverty line pay a monthly premium equal to about 3 percent of their income. 6 (However, if other family members do not elect BadgerCare coverage, pregnant women and children below age 6 can receive Medicaid coverage up to 185 percent of the poverty line without paying a premium.) We hypothesize that a program which expands coverage so substantially and to adults as well as children should have significant effects on the probability of both health insurance coverage and increased work and earnings. Figure 1 shows enrollment in BadgerCare from its inception through A striking feature of the program is the much higher enrollment of adults than children. By the end of 2001, of the nearly 90,000 enrollees, 62,000 were adults. In the short time since it s inception, BadgerCare has grown to become a major component of the public health coverage in Wisconsin. Figure 2 shows enrollment in all public family-based (requiring the presence of a minor child for eligibility) health insurance programs in the state. 7 The time period is from 1995 through 2001, the period of our analysis. In July of 1999, just before the introduction of BadgerCare, 218,000 people were enrolled in publicly subsidized family-based coverage. By the end of 2001, 367,000 people were enrolled in such coverage, including 90,000 BadgerCare enrollees. As of December 2001, BadgerCare accounted for nearly a quarter of all individuals covered by publicly subsidized family-based health insurance in the state of Wisconsin. IV. Our Research Approach In our research, we analyze the effect of BadgerCare program coverage on: 1) the probability that low- income single mother families are covered by public health insurance, 2) the probability that low-income single mothers work, and 3) the level of their earnings. We study three cohorts of mother-only families who were receiving cash assistance under the Wisconsin AFDC and TANF programs in September 1995, 1997, and 1999, and who left welfare during the subsequent period. These leaver families have been tracked on a quarterly basis in a series of administrative databases from the two years before leaving welfare through the end of Through these data sets, we observe the levels of employment, earnings and health care coverage for these women both before the introduction of BadgerCare and the period after BadgerCare has been implemented. In estimating the effect of BadgerCare on the insurance and labor market outcomes of interest, we measure insurance coverage and work/earnings patterns of these mothers from the period before BadgerCare to the period after BadgerCare. Since coverage of children is clearly an 6 This is likely to rise to 5 percent under the state budget now being considered in the Wisconsin legislature. 7 This excludes SSI-related individuals, such as those in long-term nursing facilities.

7 6 important aspect of any program evaluation targeted at this group, we also track coverage of children of the women in our sample. Identifying the causal effect of the implementation of BadgerCare on these outcomes is difficult as other relevant changes in policy and in economic conditions occurred over the same period that we observe the patterns of insurance coverage and work for these leavers. We seek to tease out reliable estimates of the effect of BadgerCare by adopting a variety of approaches. In the most straightforward estimation, we present regression-based estimates that control for both individual characteristics (such as prior work experience, welfare experience, duration since leaving cash assistance) and contemporaneous economic conditions (such as the county unemployment rate in the relevant quarter) that are likely to also affect health care coverage and the probability of work and level of earnings. Following this, we present random and fixed effects models seeking to explain public health insurance coverage and labor market performance. V. Our Data The analysis reported here is based on administrative data from the state of Wisconsin. We merged data from (1) the Client Assistance for Re-employment and Economic Support (CARES) system, which includes information collected in administering AFDC, W-2, and related means-tested programs, (2) the Computer Reporting Network (CRN) system, the precursor of CARES, providing earlier AFDC administrative data useful for constructing an AFDC history for each case, and (3) the Unemployment Insurance (UI) system, which includes information on quarterly earnings and employers. These data include all women receiving assistance under the AFDC-Regular or W-2 programs in September of 1995, 1997, and 1999 who are listed as the case head and who do not have the father of any of the children also listed on the case. We select from these participants those women who exited cash assistance within three months of our initial observation and remained off the welfare caseload for at least two consecutive months. (Our samples include those who returned to welfare within the next calendar year as well as those who stayed off.) Table 1 shows the characteristics of each of our three cohorts of leavers as of the quarter of their exit, namely: the 49,605 AFDC recipients in September 1995, and the 8,042 women who left AFDC during the last quarter of 1995, the 20,608 recipients of cash assistance in September 1997, and the 8,162 women who left AFDC or W-2 during the last quarter of 1997, and the 7,363 recipients of cash assistance in September 1999, and the 2,997 women who left cash assistance under W-2 during the last quarter of 1999.

8 7 The rate of exit is much higher for the second and third cohorts than for the first: 16 percent of women participating in AFDC in September 1995 left the program in the next three months (8,042 leavers out of the 49,605 total participants); in 1997, 40 percent of those receiving cash assistance in September 1997 left within three months; in 1999, 41 percent of cash assistance recipients in September 1999 left within three months. The women in the 1995 cohort left cash assistance before Wisconsin implemented key statewide work-focused welfare reforms in September Thus, the 1995 cohort left a welfare program that had undergone early reforms, but that had yet to implement the more demanding work-focused requirements of Pay-for Performance on a statewide basis. The 1997 cohort left cash assistance after the Wisconsin program had been transformed by waiver-based reforms and during the initial implementation of W-2. The final cohort, those who left cash assistance in 1999, were exiting a W-2 program that retained its emphasis on work, but which had added substantial work supports in the form of child care and family health insurance. Given earlier reforms and substantial declines in the caseload, we would expect women receiving benefits in 1997 and 1999 to have greater barriers to work on average than those receiving benefits in The data in Table 1 are generally consistent with this expectation: the proportion of women with low education and experience, large numbers of children, and other barriers to work are more highly represented in the later cohorts than in the 1995 cohort. The case heads in the later cohorts are generally younger than in the 1995 cohort. Perhaps owing to their relative youth, the 1999 cohort contained a lower percentage of cases that had been on welfare during most of the previous two years. In interpreting outcomes for the three cohorts, it is important to consider differences in the characteristics of women receiving and leaving welfare in each period, and our measures capture a substantial array of differences in the circumstances and characteristics of welfare participants and leavers over the three periods. However, the potential for other unobserved characteristics of welfare participants and leavers or the labor markets into which they enter must be considered in interpreting our results. 8 With these demographic and income data, we identify all families from among the population of leavers in each of the three cohorts (1995, 1997, 1999) that are eligible for BadgerCare/Medicaid benefits, using an income-based algorithm: that is, we calculate Medicaid 8 As noted above, in order to study the effect of this program expansion, we merged employer-reported wages from the Wage Record File maintained by the Division of Unemployment Insurance in the Wisconsin Department of Workforce Development through December, This file also provides information on whether each firm reporting earnings offers insurance to any of its employees. We also create a variable to measure the probability a woman has private health insurance coverage utilizing another data set the Wisconsin Family Health Survey, a continuously operated telephone survey of Wisconsin households concerning the health care conditions and coverage of household members. We combine the 1998 and 1999 surveys and estimate the probability that single women with children have coverage in those surveys using variables available in both our administrative data set and the Family Health survey. The results of this regression suggest that single women with children who are older, whose youngest child is older, or who have other adults in the house are more likely to have private coverage, while those who are black, or have only 1 or 2 children, or have a child on SSI are less likely to have private coverage.

9 8 eligibility for each household member based on the poverty-related criteria for eligibility. Household earnings are calculated as the total earnings reported in the UI database with deductions of $90/month for work expenses and $30/month plus 1/3 of the remainder earnings. VI. Some Descriptive Results on Public Health Insurance Coverage and Labor Market Outcomes In this section, we present the overall time trends in the levels of the relevant variables describing public health insurance coverage, labor force participation and earnings. It is these variables that our estimation models seek to explain; they will be presented next. Public Health Care Coverage Figure 3 shows the trends in the levels of eligibility for and take-up of public health insurance coverage from the date of exit from cash assistance for each of these three cohorts of women. We emphasize the changes occurring at the time of the introduction of BadgerCare, in late In Panel A, we show the trend in both eligibility for public health insurance and the takeup of coverage for the 1995 cohort. When this group left welfare in1995, nearly all of the women were eligible for Medicaid, and about 80 percent of them accepted Medicaid benefits. Both the percentage of mothers eligible and the percentage with Medicaid coverage fell over the nearly four years from the time they exited welfare until the beginning of BadgerCare. At the eve of BadgerCare implementation, only 63 percent of this group were eligible for Medicaid, and less than 20 percent were enrolled in the Medicaid program. As they made their way in the world of work, their labor market success reduced their eligibility. The proportion of the eligible taking up benefits dropped to about 30 percent. After BadgerCare was introduced in the third quarter of 1999, both of these rates rose substantially. The rate of eligibility again exceeded 90 percent as the BadgerCare program extended coverage to a larger proportion of these women than did the Medicaid program. The percent of leavers accepting public health benefits also began to rise. By the end of 2001, nearly 30 percent of these women were enrolled in a public health insurance program, over one-half of which were enrolled in BadgerCare. For this group of women, the descriptive evidence strongly suggests that BadgerCare contributed in an important way to increasing both their public health care coverage, and their enrollment in the BC program. Panel B shows the same eligibility and take-up outcomes for the 1997 cohort. Relative to the 1995 cohort, a higher proportion of the 1997 cohort of leavers were eligible and covered by public insurance just prior to the introduction of BadgerCare; given their characteristics, this is not surprising. Again, however, the proportion of these women eligible for public health care coverage notched up from about 80 percent to 95 percent just after the start of BadgerCare. Enrollment in a public health insurance program also began to increase beginning with the third quarter of 1999, so that by the end of 2001 about 55 percent of the women were enrolled in a

10 9 public health insurance program, up from about 52 percent prior to the start of BadgerCare. About 40 percent of those eligible were enrolled in BadgerCare. For these two early cohorts, then, the introduction of BadgerCare appears to have contributed to an increase in both the share of these women who are eligible for public health insurance and the share of eligibles who are enrolled in either Medicaid or BadgerCare. The patterns for the 1999 cohort are shown in Panel C. BadgerCare was an immediate option upon leaving cash assistance for this group. For this cohort, the rate of eligiblity for public health insurance remained above 90 percent for the entire period since leaving, and shows virtually no downward trend. By the last quarter of 2001,over 70 percent of this group were enrolled in a public health insurance program; over one-quarter of these are enrolled in BadgerCare. Because the SCHIP program was designed to increase coverage of children, their eligibility and coverage is also relevant. Since eligibility for children was more generous for children than for their mothers under Medicaid, we expect the patterns in eligibility and take-up to be more muted than those for mothers. Figure 4 shows eligibility and coverage for the children in the leavers families, and shows just this pattern. Panel A presents estimates for the 1995 cohort, panel B for the 1997 cohort, and panel C the 1999 cohort. For the 1995 cohort, prior to BadgerCare s introduction, more than 80 percent of the children were eligible for public coverage through Medicaid, so that the potential gain in eligibility from BadgerCare was far smaller than for their mothers. Eligibility remained at over 80 percent prior to BadgerCare, and eroded steadily over this period. While about 75 percent of children were enrolled in Medicaid at the time that their mothers left welfare, this percentage declined persistently to about 33 percent on the eve of BadgerCare. After BadgerCare was introduced, the share of children eligible for public health insurance increased substantially, and the percent of children enrolled in a public health insurance plan stopped its downward slide and increased to 40 percent by the end of However, less than 8 percent of the enrolled children were in BadgerCare. More than 90 percent of the children in the 1997 cohort were eligible for Medicaid prior to the introduction of BadgerCare, which again limited the potential gains in eligibility and enrollment from the new program. Again, the downward slide in the percent of the children enrolled in public health care plan was halted after the introduction of BadgerCare. By the end of 2001, BadgerCare enrollment had grown to just under 6 percent of the children in the cohort, again far below their mothers rate of BadgerCare enrollment but still sizeable. Finally, for the 1999 cohort of children, nearly 100 percent were eligible for Medicaid coverage and the eligibility rate remained in the high 90 percent range over the period of observation. This rate is higher than that of the prior cohorts, and reflects the failure of this group of women to attain income levels that would bring them above Medicaid eligibility levels. While BadgerCare enrollment increased from the first year to the second, by 2001 less than 3 percent of the children of families in this cohort of leavers were covered by the new program.

11 10 Labor Market Outcomes As we have noted, a central objective of BadgerCare is to support the work activities of leavers from cash assistance. Are the work incentives associated with the introduction of BadgerCare consistent with this expectation? Work Incentives With and Without BadgerCare Before we explore some descriptive statistics on labor market outcomes, we discuss the changes in work incentives that have accompanied the introduction of the BadgerCare program. Figure 5 compares the work incentives when the BadgerCare program is in place with those that exist when Medicaid is the only source of health insurance. The y-axis of the diagram is the total income of family, in which income is defined to be the sum of annual cash income receipts (e.g., earnings, cash public assistance), the annual value of in-kind benefits (e.g., Food Stamps, public health insurance), and annual net tax liability (which may be positive or negative). The x-axis records the returns from labor market activity, in the form of earnings. If there are no public cash or in-kind benefit programs, and no taxes, all families would fall on the 45-degree line annual earnings would equal annual total income. This diagram only becomes meaningful when the parameters of public programs and taxes are known. These parameters, in turn, depend on the size and structure of the family for which the diagram is drawn. For the purpose of understanding the incentive effects of those health insurance programs targeted on the traditional welfare population, consider a family unit composed of an unmarried woman with two children, one older and the other younger. Three income-earnings patterns are shown in the diagram. The first line above the 45- degree line shows the relationship between the total income from earnings, non-health public benefits and net taxes, and the mother s own earnings. The line is drawn to roughly describe the relationship between income and earnings for this family aside from public insurance benefits. The intercept of this line with the y-axis indicates the total level of income available to the family should the mother not work at all, so that earnings are zero; a sizable income guarantee exists. In the context of Wisconsin, a woman who is not working is required to be productive which might mean taking care of a disabled child or parent or working in a public works job for hours per week. The line indicates that income increases only partially with increases in earnings, reflecting the implicit marginal tax rate created by income-conditioned public transfer programs. The slope of the line at any earnings level indicates the amount by which income increases for each dollar of additional earnings the marginal tax rate. Hence, a lower slope indicates a greater marginal tax rate. The second line provides the same information when the benefits of the Medicaid program are added to other forms of income. Throughout the lower earnings range, this line is parallel to that of the previous line, indicating that Medicaid benefits are simply added to other forms of income throughout this range. However, as earnings increase this total benefit line becomes steeper, dropping first when the mother s income exceeds her eligibility level (approximately 50

12 11 percent of the poverty line), and dropping again when the older child loses coverage (at 100 percent of the poverty line). These Medicaid program notches provide a substantial incentive for women to avoid earnings increases that would result in income high enough to trigger the loss of own or older child insurance coverage. The incentives implicit in the Medicaid program severely discourage additional work effort. By allowing adults and older children to remain eligible for public health insurance at far higher levels of income (up to 200 percent of the poverty line), the BadgerCare program eliminates these precipitous decreases in incomes as earnings increase. This is shown in the top line on the diagram, where total income is greater at each level of earnings beyond the point where the woman lost her coverage under the Medicaid only option, and where no precipitous discontinuities or notches exist. The slightly higher slope of this line when income exceeds 150 percent of the poverty line indicates the required premium of 3 percent of income in excess of this level. Two perspectives for thinking about work incentives are reflected in this diagram. The first perspective concerns the marginal incentives to increase work and earnings at any level of earnings. The extent of this incentive is captured by the slope of each segment of the line which is the marginal tax rate the rate at which benefits are reduced when earnings increase. Inspecting the diagram indicates that the set of benefits with the BadgerCare program in place has the same marginal tax rates as does the benefit package with only Medicaid, until the income level of $9,300 or the end of eligibility for the mother. The only segment of the BadgerCare line that is steeper than the Medicaid line is the final segment shown where the family is required to pay up to 3 percent of their income toward the BadgerCare premium. More importantly, the benefit package with the BadgerCare program in place eliminates all of the Medicaid notches, which impose enormous disincentives to increase work effort at the point that they occur. From this perspective, the benefit package with BadgerCare in place has far stronger incentives to work than does the arrangement with only the Medicaid program. The second perspective takes a broader view of what constitutes the marginal work choice of these low skill women. For example, consider a woman who is working part time at a job that pays earnings that yield income of less than 50 percent of the poverty line. Assume that she now has the opportunity to be employed in a job that pays, say, twice that amount this job becomes her marginal work choice. To determine whether or not to take this job, the woman compares her total income with the new job with that in her current job; the size of the gain indicates her incentive to take the job. From inspecting the diagram, it is clear that for any job change that moves the woman from an earnings level below where a notch exists to an earnings level above a notch, the set of benefits with BadgerCare in effect yields a substantial increase in income, relative to the increase available under the benefit regime with only the Medicaid program. This comparison, too, indicates that the addition of BadgerCare to the existing package of benefits strongly encourages additional work effort on the part of those women who are eligible for benefits. In essence, these single mothers will be more likely to work, and more likely to work more hours if employed, if the probability of losing health insurance coverage by doing so is reduced; it is just this effect that BadgerCare creates.

13 12 This conclusion, however, must be tempered by the economist s concern that the possibility of exogenous income from programs such as BadgerCare will reduce the need to work and earn in order to attain a given level of well-being; the recipient will be able to purchase more of all desired goods, including leisure, and hence may choose less time working and lower earnings. However, this concern fails to reflect the fact that such an unrestricted increase in income also enables the recipient to buy a number of things that both encourage and facilitate work and earnings, especially when the fallback position of reliance on means-tested benefits is no longer available. These work-facilitating purchases often entail the taking of risks, and include items such as: arranging higher quality or more reliable child care, enabling the accumulation of savings necessary to purchase an automobile or otherwise reduce travel difficulties, moving to a location where more jobs are available or where living is more workconducive, or moving to jobs that pay a higher wage rate, or have more work-time available. Closely related to the increased ability to buy such work-enhancing items is the more difficult-to-measure, but potentially important, impact of the increased resources available in strengthening the belief on the part of the mother that the goal of self-sufficiency is actually attainable. Further, the protection afforded by BadgerCare coverage may also provide a sufficient cushion to enable the mother to handle financial or other difficulties without disrupting work and to make investments that will increase self-sufficiency. Finally, BadgerCare may introduce a demand side effect: firms may be more willing to hire women and pay them more if they have public health insurance and are not expected to enroll in firm-based health care benefit programs. If these effects dominate the standard labor supply effect, as they are likely to do in the absence of available means-tested benefits, the increase in available public health insurance will encourage mothers to enter, and sustain, their participation in the labor force, and to secure financial self-sufficiency and independence. Given these considerations, we hypothesize that the introduction of BadgerCare will complement the reduction of means-tested, non-work-related benefits in increasing mothers work and earnings.

14 Some Labor Market Patterns 13 A first look at the labor supply implications of the introduction of BadgerCare involve a set of before/after comparisons for the three cohorts. Figure 6 shows the fourth quarter patterns in the proportion of each cohort that is employed. For all three cohorts, the employment rate is about 70 percent as these women terminate the receipt of cash assistance. As the time from exit lengthens, the employment rate erodes for all of the groups. For all of the cohorts, the employment rate in 2001 drops noticeably, consistent with the recession of that year when the state unemployment rose to 5 percent. The effect of the recession appears to have been most severe for those with the least post-exit time in the labor market, the 1999 cohort of leavers. The patterns shown in Figure 6 do not indicate any substantial change in labor force participation accompanying the introduction of the BadgerCare program. 9 Figure 7 shows quarterly earnings for those leavers in each of the three cohorts who work. Among the workers in the 1995 cohort, quarterly earnings immediately after exit were about $2700, and rose rather continuously over the subsequent years, reaching an average of over $4000 by The immediate post-exit earnings of workers in the later two cohorts were lower than those of workers in the first cohort, and averaged about $2200. They too trended upward as the time from exit increased. For all of the cohorts, earnings grew slowly during the recession of Earnings appear to jump for workers in the 4 th quarter of 1999, after the introduction of BadgerCare in July of that year. However, this pattern may be simply a calendar year effect, reflecting both an increase demand for temporary and part-time employment prior to the holidays, and a greater willingness to work in anticipation of the desire to purchase holiday gifts. For each cohort, a spike in earnings appears in the 4 th quarter of each year. VII. Modeling the Effect of BadgerCare on Public Health Insurance Coverage and Labor Market Outcomes Our simple before/after comparisons suggest that BadgerCare increased the public health insurance coverage of leavers in each of the 1995 and 1997 cohorts. We saw little employment and earnings changes from before to after the introduction of BadgerCare. However, these patterns do not account for the possibility that a variety of other factors may have been changing contemporaneously with the introduction of BadgerCare; any reliable estimation of the effect of BadgerCare on these outcomes must account for the effect of these other factors. In this section, we present a series of estimates designed to reveal the independent effect of BadgerCare on both health insurance coverage and labor market outcomes. 9 The deterioration in the state s labor market, as captured by the unemployment rate, began in early This is true as well for the state s major city, Milwaukee. 10 Note, however, that the women included in these trends are an increasingly selected group among all of the women in each cohort, as the rate of employment eroded over time for each cohort; see Figure 6.

15 14 An Estimation Framework for Modeling the Effect of BadgerCare Consider the following equation: y = X β + θ + u it it i it where i indexes individuals and t indexes time period. y it is the dependent variable such as public health insurance coverage, employment, and earnings of individual i at time t. X it contains observable variables that change across i but not t, variables that change across t but not i, and variables that change across i and t. For example, the education level of the mothers varies across individuals, but not t; BadgerCare availability does not vary across individual, but does vary with time; the county unemployment rate varies across individuals (who live in different counties) and across time. θi is individual unobserved heterogeneity such as ability. The error term uit can be decomposed into the effect of time since exit, calendar time and a random error, that is: u it = 1 µ t + δ 2 δ BC + η t it Since BC t enters as a dummy variable equal to one beginning in the 4 th quarter of 1999 (the first full quarter after BadgerCare was introduced), it s coefficient reflects both the effect of calendar time and Badger Care. Assume the absence of unobserved heterogeneity, θ i, such that all individual differences are captured by observable characteristics,. In this case, an estimate that pools all of the X it observations assumes that uit is iid both over time and across individuals. In this case, each observation of each individual in each time period contributes to the regression as though it comes from different person. In our estimation, we first we run the regression with all available data, including 25 quarters of observations for the 1995 cohort, 17 quarters for the 1997 cohort, and 9 quarters for the 1999 cohort. In a second step, we estimate the model with 9 quarters for the members of each cohort to keep the number of observations for each cohort the same. The variation for identification of this equation comes from differences between individuals across time. It may be the case that idiosyncratic heterogeneity of individuals is not captured by the observable characteristics, and may effect behavior. To take into account this possibility, we include θ i in the equation. If θi is orthogonal to X it, µ t and BC t, then individual heterogeneity, a random variable, has explanatory power. The underlying assumption is E( η X, µ, BC, θ )=0 and E( θ i X it, µ t, BC t )=E( θ i )=0 it it t t i

16 15 On the other hand, if θ i is not orthogonal to, but is correlated in some way with X it, µ t, and BC t, then θi is treated as a parameter to be estimated. The fixed effect regression violates the second assumption above. By differencing (or time demeaning), the equation can remove individual specificθ i and capture the effect of changes in other explanatory variables, including BadgerCare. The variation for the identification is from the change in the value of the explanatory variables over time. The Effect of BadgerCare on Health Insurance Coverage In order to better identify a tie between the implementation of BadgerCare and public health care coverage, we first estimate a probit equation in which having public health coverage is given a value of one. We introduce into the estimation a large number of independent variables designed to reflect factors that are potentially related to public health care coverage, including: race, education, age and number of children, number of other adults in household, prior duration on cash assistance, earnings history while on welfare (for up to 8 quarters), current area of residence, current unemployment rate, proportion of female headed families in the woman s neighborhood, the presence of any child in the family on SSI, the number of quarters since the woman exited cash assistance, the estimated probability the woman has private coverage (based on the estimate described above and reported in Appendix II). In addition, to capture the independent effect of BadgerCare on this outcome, we include a dummy variable for whether or not BadgerCare was available. To reflect possible other differences between the cohorts, we also include a dummy variable indicating to which cohort an

17 16 observation belongs. We estimate this model across all three cohorts where each woman is entered each quarter. Table 2 presents the results of this estimate. The coefficients on the control variables are, in general, as expected. The estimated probability of having private coverage (estprob) is positively and significantly related to having public health insurance coverage, as is the contemporaneous unemployment rate. The coefficient on the variable describing whether or not BadgerCare was available is positive and highly statistically significant. The coefficient is 0.23, while the standard error is less than 0.01, suggesting that after controlling for other factors likely to be associated with eligibility for, and take-up of, public coverage, BadgerCare plays a important role. The coefficients on the dummy variables for the 1995 and 1997 cohorts are negative and significant, indicating that the probability of having public health insurance is lower for these cohorts than for the 1999 cohort. The coefficient on the interaction term suggests that BadgerCare was more positively associated with public health insurance coverage for the 1995 cohort than for the other cohorts. Finally, to account for the possibility that the iid assumption is not correct, we have estimated both fixed and random effects logistic regression models. These estimates are run separately for mothers in the 1995 and 1997 cohorts. Tables 3 and 4 present the results of these estimations for the 1995 and 1997 cohorts. 11 Table 3 shows the random effects estimates for the 1995 and 1997 cohorts. The coefficient on the BadgerCare variable is positive and significant for the 1995 cohort; the coefficient is 1.16, and the standard error is.03. However, for the 1997 cohort the sign on the BadgerCare variable becomes negative, and it too is significant. Table 4 shows the fixed effects models for 1995 and 1997, in which changes in the timevarying variables are included in the estimate. The 1995 coefficient on the BadgerCare variable is very similar to that in the random effects model, 1.17 and again the standard error is.03, indicating statistical significance. Again, the pattern is reversed for the 1997 cohort. 12 These results suggest that for the 1995 cohort of welfare leavers the availability of BadgerCare is statistically significantly associated with an increase in the probability of having public health insurance coverage. The results for the 1997 cohort are less clear: a positive effect is indicated in the simple probit estimates; however, the random and fixed effects estimates suggest a negative relationship. Because of the nonlinear nature of the estimates, it is difficult to perceive the quantitative importance of BadgerCare availability on the probability of public health insurance coverage from the estimated coefficients that have been presented. In Table 5, we present the results of a 11 We also run a similar model using only 9 quarters for each individual in order to have a balanced sample. These results are consistent with the reported model and are available from the authors upon request. 12 The Hausman tests run on these estimates provide evidence that the fixed effects models are not consistent and hence, the random effects model is preferred. We report both as the results are robust.

18 17 simulation of the effect of BadgerCare availability on the probability of this insurance outcome, holding other variables in the estimates at their mean levels. Simulated effects are presented for the pooled results without individual fixed effects, and (eventually) for the fixed and random effects estimates. When BadgerCare is available and other values are at their mean, the probability of having public health insurance is.52 for the 1995 cohort and.66 for the 1997 cohort. In contrast, when BadgerCare is not available, the predicted probability of having public health insurance coverage is.42 for the 1995 cohort and.58 for the 1997 cohort. Stated another way, the simulated availability of BadgerCare raised the probability of having public health insurance coverage by 10 percentage points for the 1995 cohort and 8 percentage points for the 1997 cohort. The with/without BadgerCare differences were generally similar for subsamples arrayed by education (less than 12 years, 12 years, and more than 12 years of education), suggesting that the effect of BadgerCare on the probability of having public health insurance was approximately equal for those with greater and lesser education. The Effect of BadgerCare on Employment In section VI., we presented our analysis of the labor supply effects of BadgerCare availability, leading us to hypothesize that the BadgerCare program would encourage work effort among those who are potentially eligible for its benefits. In the following paragraphs we describe our estimates of the employment effects of BadgerCare; the next subsection describes the effects on earnings. In Tables 6-9, we present our estimates of the effects of BadgerCare availability on the decision of women to work. Table 6 presents the basic probit estimates of the effect of BadgerCare availability on the employment of mothers. Again, the coefficients on the control variables are, in general, as expected. The coefficients on the dummy variables for the 1995 and 1997 cohorts are negative and significant, indicating that being employed is lower in these two cohorts than in the 1999 cohort. The coefficient on the variable describing the effect of BadgerCare availability on employment is positive and statistically significant for the 1995 cohort, suggesting that BadgerCare availability encourages labor supply for these welfare leavers, but not for the overall sample of leavers. 13 Tables 7 and 8 show the results of the fixed and random effects estimates of the effect of BadgerCare availability on employment for the 1995 and 1997 cohorts. Table 7 indicates that random effects coefficient on the BadgerCare variable is positive for the 1995 cohort, but not 13 We also estimated this equation using only the 9 quarters since exit for each mother, in order to secure a balanced panel. The coefficient on the BadgerCare dummy variable increases, and remains statistically significant. Results are available from the authors. An F test indicates that there is no statistical significance between the two sets of results.