Background and previous literature

employees must contribute directly for coverage. Third, did the Medicaid expan- sions affect the percentage of workers accepting offers of employer-sponsored insurance?

2. Background and previous literature

2.1. The Medicaid program Medicaid is a joint state–federal program providing health insurance to three groups of Americans: low-income aged and disabled people; people who qualify Ž . because of large medical expenses ‘‘the medically needy’’ ; and low-income families with children. Historically, Medicaid eligibility among the third and largest group was tied to eligibility for the cash welfare program, Aid to Families Ž . with Dependent Children AFDC . Because AFDC eligibility depended partly on family structure, and because the program’s income threshold was well below the federal poverty line in most states, up through the late 1980s Medicaid covered Ž less than half the families with incomes below the poverty line Coughlin et al., . 1994 . Starting in the mid-1980s, a series of federal laws diminished the link between the two programs by extending Medicaid eligibility to poor families who previously had been ‘‘categorically ineligible’’ for AFDC by reason of family structure, and by increasing the Medicaid income threshold above the AFDC threshold. These expansions increased both Medicaid eligibility and coverage Ž . substantially Coughlin et al., 1994; Shore-Sheppard, 1997 . The Medicaid expansions occurred against a backdrop of declining private Ž insurance, particularly for lower-income workers Kronick, 1991; Acs, 1995; Long . and Rodgers, 1995; Olson, 1995; Fronstin, 1997 . Over the late 1980s and early 1990s, the increase in Medicaid coverage and the decline in private insurance were roughly offsetting. One interpretation of these two trends is that the percentage of Americans without insurance would have increased were it not for the Medicaid expansions. However, it is also possible that the expansions contributed to the decline in private insurance by inducing a movement from private coverage to Medicaid by the newly eligible. 2.2. The economics of employer-proÕided health insurance To the extent that Medicaid is viewed as a substitute for private insurance, the expansions likely reduced the demand for private insurance among low wage workers eligible for the public program. Since nearly 90 of Americans with private insurance are covered by employer-sponsored policies, it is important to understand how such a shift in demand among workers might affect the decisions of employers. Employer-provided health insurance is similar in a number of important ways Ž . to local public goods provided by communities Goldstein and Pauly, 1976 . Employers that choose to offer insurance are typically constrained to offer a limited number of benefit options, which they will choose based on the prefer- ences of their employees. This creates an incentive for workers to sort among firms according to their demand for health insurance, cash wages and other benefits and amenities. 1 When sorting is complete, workers with a low demand Ž for insurance e.g., lower income workers or those with an alternative source of . coverage will choose to work for non-insuring firms, while those with a greater Ž . preference for coverage e.g., higher income workers will accept lower wages to Ž work for an insuring firm. When sorting is incomplete due, say, to production . complementarities between low- and high-skill workers , employers’ health benefit decisions will involve weighing the preferences of different worker constituencies, taking account of their value to the firm, their outside employment opportunities, and their willingness to forego wages in return for health benefits. The public good model provides a useful framework for considering the potential for the Medicaid expansions to reduce insurance offers by firms. To the extent that sorting is complete, firms with the greatest proportion of workers made eligible by the Medicaid expansions will be those that were least likely to offer insurance prior to the expansions, while firms that were most likely to offer insurance will have very few Medicaid-eligible workers even after the expansions. In cases where sorting is incomplete, employers will have responded to the expansions by weighing the interests of workers who are newly eligible for Medicaid against those of higher income workers who remain ineligible. If enough employees become eligible for Medicaid, a firm may drop coverage. However, if workers remaining ineligible outnumber newly eligible workers, or if employers’ benefit decisions are more responsive to the preferences of more highly paid employees, then employers will be unlikely to drop health insurance in response to changes in Medicaid eligibility rules. Ž . Ž . Work by Dranove and Spier 1996 and Levy 1997 suggests an alternative type of response by insuring firms that employ some workers who are eligible for Medicaid and others who are not. In Dranove and Spier’s model, employers require employee premium contributions to encourage workers with an alternative source of insurance to decline coverage through the firm. While its main emphasis is on the effect of coverage through a spouse, their model is easily extended to account for workers who have the option of Medicaid coverage, with the empirical prediction that an increase in the number of workers who are eligible for Medicaid Ž . will increase the prevalence of employee premium contributions. Levy 1997 presents a slightly different model with similar empirical predictions. An implica- 1 Ž . Ž . Scott et al. 1989 and Dye and Antle 1984 also present models in which workers sort among firms according to their demand for fringe benefits. tion of both studies is that in examining the effect of the Medicaid expansions on employer health insurance decisions, it is important to look not only at the decision to offer insurance, but also how premium payments are split between employers and employees, and the decision by employees to accept coverage that is offered to them. Recent research has pointed to the importance of declining take-up as an explanation for the decline in employer-sponsored insurance over the past decade Ž . or so. For example, Cooper and Schone 1997 compare take-up rates from the 1987 National Medical Expenditure Survey with rates from the 1996 Medical Expenditure Panel Survey and find an increase in the proportion of workers who Ž . do not accept offered coverage. Similarly, Farber and Levy 1998 examine various supplements to the CPS from 1979 to 1997 and conclude that reductions in health insurance coverage have come from both declining take-up rates for long-term, full-time employees and from reductions in eligibility for short-term and part-time employees. 2 2.3. PreÕious eÕidence on crowd-out Several studies address the question of whether the recent increase in Medicaid eligibility contributed to the decline in private health insurance coverage. Since all of these studies are based on household survey data, the role of employers has been considered only indirectly, if at all. Ž . Using 1989 and 1994 data from the March CPS, Dubay and Kenney 1997 compare the change in private coverage for children to the change for men, who Ž were theoretically unaffected by the expansion of Medicaid though Medicaid . coverage for men did rise over this period . They conclude that Medicaid eligibility expansions led to a slight reduction in private insurance, and that 17 Ž . Ž . 14 of the total increase in Medicaid enrolment of children pregnant women Ž . was associated with a decline in private coverage. Shore-Sheppard 1997 also uses the data from the CPS, for the years 1988 to 1996, to estimate the impact of the expansions on coverage of the newly eligible children. She finds that state– age–income cells with larger fractions of newly eligible children experienced a larger loss of private coverage. In addition, she finds that children who were eligible at the beginning of the period experienced a more extensive loss of private coverage than did children who became eligible during the later expansions. Ž . The largest estimate of crowding out comes from Cutler and Gruber 1996 . Using March CPS data from 1988 to 1993, they conclude that roughly half of the increase in Medicaid coverage was associated with a reduction on private insur- 2 It is important to note that while tangentially relevant, these studies do not directly bear on the question of Medicaid crowd-out. By the same token, while our analysis exploits variation over time in Medicaid eligibility, it is not our objective to explain the recent decline in insurance coverage. ance. They use additional data from the 1988 and 1993 CPS Benefit Supplements to explore the reasons behind their findings. Specifically, they examine the effect of Medicaid eligibility on whether a worker is offered health insurance benefits, and whether a worker takes coverage conditional on an offer. They find no effect of Medicaid eligibility on the first outcome, and a significant negative effect on the second. From these results they conclude that crowding out arose mainly from workers declining coverage offered to them and dropping coverage for their dependents, rather than a decline in offers of coverage by firms. 3 In addition to these studies using repeated cross-section data from the CPS, three more recent studies use panel data to examine the relationship between Medicaid and private insurance coverage for children. Using data from the Ž . Ž . National Longitudinal Survey of Youth NLSY , Yazici and Kaestner 1998 compare insurance transitions between 1988 and 1992 for various ‘‘treatment’’ and ‘‘control’’ groups, defined based on income and family circumstances. Pairwise comparisons of different treatment and control groups produce alternative estimates of the percent of the increase in Medicaid enrolment that came from private insurance. These estimates range from 5.3 to 23.9. Thorpe and Florence Ž . 1998r1999 also use NLSY data. They do not model the effect of Medicaid eligibility per se, and therefore do not provide a direct estimate of crowd-out that is comparable to those of the CPS-based studies. Rather, they provide descriptive evidence on changes in insurance coverage for children who are newly enrolled in Medicaid and their parents. They find that only a small fraction of parents with children enrolled in Medicaid maintained private coverage for themselves, and that for most parents who lost employer-sponsored coverage at the same time their children were taking up Medicaid coverage, the change in insurance status coincided with a job loss. Ž . Finally, Blumberg et al. 1999 use data from the 1990 Survey of Income and Ž . Program Participation SIPP , which covers a 32-month period beginning in late 1989. Stratifying their sample by insurance status at the beginning of the panel, they compare the end-of-panel insurance coverage of children who most likely gained eligibility as a result of the expansions with that for a comparison group who, because of their age and family income, remained ineligible over the entire period. They find that among children with private insurance at the start of the panel, those in the Medicaid-eligible ‘‘target’’ group were 1.2 percentage points less likely to have private insurance, and 5.3 percentage points more likely to have Medicaid at the end of the panel than children in the comparison group. Taking the ratio of these two effects, they conclude that 23 of the transitions from private insurance to Medicaid can be attributed to the expansions. Contrary to expecta- 3 The studies using the CPS differ not only in terms of the empirical methods used, but in the way Ž . the different authors define the concept of crowd-out. See Cutler and Gruber 1997 for a discussion of the key conceptual issues. tions, they find that among children who began the period uninsured, those in the Ž . target group were slightly and insignificantly more likely to have private insurance 28 months later, suggesting that expanded Medicaid eligibility did not impede transitions from uninsurance to private coverage. Finally, they find that among children who were initially uninsured, those in the target group were 7.9 percentage points more likely to gain Medicaid coverage, and 8.6 percentage points less likely to remain uninsured than those in the comparison group. While the data sources and analytical techniques are quite different, the studies using longitudinal data share an important shortcoming with the analysis we present below. Compared to a repeated cross-section of CPS data, the NLSY, the SIPP and our employer survey data provide much smaller samples. For example, Ž . the main analysis of children by Cutler and Gruber 1996 is based on a sample that is over 100 times as large as the NLSY sample used by Yazici and Kaestner Ž . Ž . 1998 : 266,421 observations vs. 2244 observations. Blumberg et al. 1999 use samples ranging from 902 to 2587 observations; as discussed in the next section, the largest sample available from our employer survey data is 3062 observations. Thus, in addition to any differences in the empirical specifications used, there are important differences across existing studies in terms of the precision with which policy effects can be estimated. As a result, in many cases, because of large confidence intervals what appear to be widely divergent results are not statistically distinguishable from one another.

3. Estimation strategy: measuring the impact of Medicaid using firm data