Trends in Health Insurance Coverage among Low-Skilled Women. March 3, Judith A. Levine University of Chicago

Transcription

1 Very preliminary; please do not cite or distribute Comments welcome Trends in Health Insurance Coverage among Low-Skilled Women March 3, 2004 Thomas DeLeire Harvard University and University of Chicago Judith A. Levine University of Chicago Helen Levy University of Chicago Paper Presented at the BU/Harvard/MIT Health Economics Seminar The authors would like to thank the Robert Wood Johnson Foundation s Economic Research Initiative on the Uninsured at the University of Michigan for financial support, Vanessa Coca, Patrick Wightman, and James Larson for excellent research assistance and Mark Duggan and seminar participants at the ERIU Ann Arbor conference for helpful comments.

2 Trends in Health Insurance Coverage among Low-Skilled Women Abstract We analyze the trends in health insurance coverage for women with less than a high school education during the 1990s and also examine how welfare reform affected their probability of insurance coverage. We compare the experiences of these women with trends for women with a high school and college education. We find that the probability of having health insurance declined for all of these groups during the 1990s, but the reasons for these declines differ across groups. Women with less than a high school education experienced a large decline in the probability of having health insurance as a result of declines in spousal coverage (due both to declines in marriage and to declines in spousal coverage) and declines in own-employer sponsored coverage for among workers. By contrast, women with a high school degree experienced a smaller decline in health insurance coverage during the 1990s, mostly because declines in spousal coverage were partially offset by increases in own-employer coverage. Women with a college education experienced only very small declines in the probability of health insurance coverage in the 1990s. Against this backdrop of large overall declines in health insurance coverage, welfare reform appears to have led to a modest increase in health insurance coverage for women with less than a high school degree, primarily by increasing the probability of having private health insurance. This modest increase was not large enough to offset the large secular declines in health insurance coverage among low-skilled women.

3 1. Introduction In this paper, we analyze the trends in health insurance coverage for women with less than a high school education during the 1990s and examine how welfare reform affected their probability of insurance coverage. As a comparison, we also examine trends in health insurance coverage for women with a high school degree but not a college degree, and women with a college education). We also analyze trends in the demographic and economic characteristics of these groups (such as race, ethnicity, marital status, and employment) to see how other underlying determinants of coverage may have changed over time. We find that the probability of having health insurance declined for all three groups during the 1990s, but the reasons for this decline differ across groups. For women with less than a high school degree, the probability of having any health insurance coverage declined by 8 percentage points between 1988 and The reason for this decline was a decline in spousal coverage a decline in marriage contributed 3.6 percentage points to this decline and a decline in spousal coverage among those married contributed 5.6 percentage points. A decline in the own-employer coverage among workers also contributed 3 percentage points to the decline. Health insurance coverage declined by only 2 percentage points for women with a high school education during this period as declined in marriage and spousal coverage were partially offset by increases in receipt of own-employer coverage. By contrast, women with a college education experienced a decline in the probability of having health insurance of only 0.45 percentage points. One potential causal or mitigating factor for the decline in health insurance coverage among low-skilled women might be welfare reform. The Personal Responsibility and Work Reconciliation Act of 1996 (PRWORA) was intended to increase employment, reduce welfare 1

4 dependence, and encourage poor women to lift themselves out of poverty. A comprehensive review of the literature on the impacts of welfare reform (Blank 2003) shows that since 1996, employment and earnings among poor women have increased (Moffitt 1999; Schoeni and Blank 2000), marriage has increased (Bitler, Gelbach, and Hoynes 2001; Schoeni and Blank 2000), and welfare rolls have dropped dramatically (Bell 2001; Blank 2001; Wallace and Blank 1999; Ziliak et al 2000). There is less evidence, however, on the association between welfare reform and other measures of economic well-being. In particular, there has been very little research on rates of health insurance coverage among the population likely to be affected by PRWORA: low-skilled women. About one-sixth of adult women in 1998 had no insurance coverage; among women with a high school education or less, the fraction was three-tenths (see Table 1). Lack of insurance may lead these women to delay necessary medical care and exposes them to the risk of potentially catastrophic medical expenses. Several studies have documented that lack of health insurance results in worse health, particularly for vulnerable low-income populations (Currie and Gruber 1996a, Currie and Gruber 1996b, Currie and Gruber 1997, Newhouse et al. 1993; this literature is reviewed by Levy and Meltzer 2004). The three main sources of insurance for women who do have coverage are their own employers, their spouses employers, and the Medicaid program. PRWORA through its effects on low-skilled women s eligibility for Medicaid and its effects on employment, income, and marriage rates may have increased or decreased the fraction of women who are uninsured. It may have increased coverage rates if more women are employed in jobs that provide insurance; on the other hand, it may have raised women s incomes to the point of losing Medicaid eligibility without moving them into jobs that provide health insurance. The probability of 2

5 insurance coverage among women is no lower just above the poverty threshold than just below it (see Table 2). Similarly, an increase in marriage may result in more private insurance for women if their husbands have jobs that provide family coverage, or it may simply disqualify these women from public coverage. In addition, welfare reform decoupled Medicaid eligibility from welfare enrollment. Although eligibility for Medicaid and other public health insurance expanded during the 1990s, many welfare recipients may have been unaware of this or may have faced new barriers to enrollment as a result of the changes in welfare (Card and Shore-Sheppard 2001). Against this backdrop of declining coverage, welfare reform appears to have led to an increase in health insurance coverage for women with less than a high school education which partially offset the large overarching declines in health insurance coverage. For women with less than a high school education, welfare waivers are associated with a 2.9 percentage point decline in the probability of being uninsured and a 2.4 percentage point increase in the probability of having own-employer provided coverage. For women with a high school or college education, welfare reform is not associated with a change in the probability of having health insurance coverage. These findings suggest that welfare reform partially offset the large declines in insurance coverage that occurred for low-skilled women between 1988 and Further research is needed to understand the underlying economic factors driving the across-the-board decline in insurance coverage during the economic expansion of the 1990s. 2. Background In the 1990s, the landscape for both private and public health insurance was changing dramatically. Low educated women s health insurance coverage during this period may also 3

6 have been changing as a result of welfare reform. For example, research shows that labor markets for low-skilled workers have been deteriorating since the late 1970s (Blank 1997) and that over the same period there have been increases in temporary and part-time work (Tilly 1995). Moreover, women s labor force participation had been increasing and marriage rates had been decreasing. Many studies have documented the gradual erosion of employer-sponsored insurance coverage (for example, Farber and Levy 2000, Cooper and Schone 1997). Each of these factors could independently have changed the health insurance coverage rates of low skilled women. At the same time, expansions in eligibility for Medicaid and other public health insurance programs have been occurring since 1979 (see Currie and Gruber 1996a, Currie and Gruber 1996b, and Aizer and Grogger 2000 for excellent overviews of these expansions). These Medicaid expansions may themselves have contributed to declines in private coverage: the socalled crowd out effect (Cutler and Gruber 1996, Cutler and Gruber 1997, Ham and Shore- Sheppard 2000, Dubay and Kenney 1997). Thus, evaluating the impact of welfare reform on the problem of the uninsured requires a thorough understanding of the underlying trends in private and public coverage that were occurring during this period. There is a large literature on the impact of welfare reform (primarily the studies cited above) and also a large literature on trends in health insurance coverage (for example, in addition to the papers already cited, Fronstin and Snider 1996/97). There is, however, very little research in the intersection of these two areas. In fact, we are aware of only three papers analyzing health insurance coverage for low-skilled individuals likely to have been affected by welfare reform: Farber and Levy (2000), Currie and Yelowitz (1999), and Kaestner and Kaushal (2003). Farber and Levy (2000) examine trends in coverage by health insurance from 1979 to 1997 using data from the Current Population Survey (CPS). They find that coverage by employer- 4

7 sponsored health insurance declined over this period, and that the erosion of coverage from 1988 to 1997 was due to declines in takeup among high tenure workers and to declines in eligibility of low-tenure workers rather than to declines in employer offering. Finally, they find as large a decline in takeup among college-educated workers as among workers with a high school diploma or less. Their analysis includes only workers, does not consider public coverage, and does not examine women s rates of coverage separately from men s. Currie and Yelowitz (1999) find trends similar to those documented by Farber and Levy (2000). In addition, they examine trends in any health insurance coverage, coverage by a private employer, and Medicaid coverage for single mothers. They perform this analysis using both the March CPS and the Survey of Income and Program Participation (SIPP). Among all adult single mothers, they find a decline in private insurance coverage, but no decline in employer-provided coverage or increase in Medicaid coverage in the March CPS. The SIPP, however, shows declines in both private coverage and employer-provided coverage and increases in Medicaid coverage. Among single mothers who work, they find declines in private coverage in both surveys, declines in employer-provided coverage only in the CPS, and increases in Medicaid coverage in both surveys. The much-heralded success of welfare reform is based primarily on evidence on declines in the welfare caseloads and increases in employment, earnings, and marriage among former recipients. However, true success of welfare reform depends not simply on one-time exits from welfare into employment or marriage, but stability in either or both of these states. While longterm effects of welfare reform on employment patterns are as yet unknown, past work on welfare dynamics under the Aid to Families with Dependent Children (AFDC) regime suggests that there is reason to be concerned about the stability of employment and marital relationships for women 5

8 leaving welfare. Pavetti (1993) finds that 65% of those who leave welfare for paid work eventually return. Bane and Ellwood (1992) also document dramatic differences between the length of single welfare spells and time spent on welfare over the life-course. Returns to welfare will likely decline under Temporary Assistance to Needy Families (TANF) given the new limits on lifetime welfare use, but that fact does not mean employment will necessarily become more stable. Burtless (2000) predicted that while most former welfare recipients will be able to find some type of employment, many will find low-wage jobs that end quickly. The likelihood of job instability for welfare leavers raises immediate concerns about lack of health insurance coverage. Health insurance coverage is one of the hallmarks of a good (i.e. stable) job. Farber and Levy (2000) document that part-time workers, workers with low levels of education and workers with less than one year of experience on the job are much less likely than full-time workers in long-term jobs or workers with higher levels of education to be offered health insurance. If unskilled women s cycling between spells of welfare and employment is replaced by their cycling between different low-wage jobs, reform is unlikely to result in increased health insurance coverage rates. Similarly, spousal coverage is a realistic option only for women who remain married to men who are in stable jobs. If their own or their spouses employment is unstable or of low quality, former welfare recipients are likely to be uninsured, even if their incomes rise above the poverty threshold. Moreover, increased earnings or marriage as a result of welfare reform could contribute to the ranks of the uninsured by reducing the number of women eligible for public coverage. 6

9 3. Methods Our empirical analysis has three components: first, an analysis of long-term trends in health insurance coverage for women by levels of education between 1988 and 2000, and second, an analysis of the impact of welfare reform on the coverage rates of these groups. We are primarily interested in trends in health insurance coverage for a group likely to have been affected by changes in the economic and policy environment in the 1990s, women with less than a high school degree. We contrast the experiences of these women with those of women with more education because the insurance coverage rates of women with more education, especially college educated women, likely is less affected by changes in the economy or by policy changes such as welfare reform. The individuals in our analysis can obtain coverage from a number of different sources: from their own employers, from their spouse s employer, directly from insurance companies, or from public programs such as Medicaid. We define these sources of coverage so that they are mutually exclusive and sum to the overall rate of insurance coverage. That is, C = OWN + SP + IND + PUB (1) where C is the rate of health insurance coverage (from any source); OWN is own-employer provided coverage rate; SP is the rate of coverage through a spouse s policy; IND is the independently purchased private non-group coverage rate; and PUB is the public health insurance coverage rate. We begin our analysis by calculating the trends in the components of coverage (own employer, spouse, non-group, and public) in order to determine which of these components is 7

10 responsible for any change in the rate of insurance coverage. The changes in each of the components of health insurance coverage rates must sum to the change in the rate of health insurance coverage, so this exercise will reveal how much of the change in health insurance coverage is the result of changes in employer-sponsored coverage, spousal coverage, non-group coverage, and public coverage. We would also like to know how much of the change in private insurance coverage between 1988 and 2000 is explained by changes in employment and marriage rates and how much is due to changes in the underlying trend in own-employer coverage conditional upon employment status, and the trend in spousal coverage conditional upon being married. In order to quantify the extent to which trends in employment and marriage rates are responsible for the changes in coverage, we use a simple decomposition, described in the following equation: ( ) ( ) MAR ( SP MAR) ( OWN EMP) EMP ( OWN NONEMP) (1 EMP) ( SP MAR) MAR C = EMP OWN EMP OWN NONEMP IND + PUB (2) where? C is the change in the rate of health insurance coverage from 1988 to 2000; OWN EMP is the rate of own-employer provided coverage conditional upon being employed in 2000;? OWN EMP is the change in the rate of own-employer provided coverage conditional upon being employed from 1988 to 2000; EMP is the fraction of the sample working in 1988; OWN NONEMP is the rate of own-employer provided coverage conditional upon not being employed in 2000;? OWN NONEMP is the change in the rate of own-employer provided coverage conditional upon not being employed from 1988 to 2000;? EMP is the change in the employment rate from 1988 to 2000; 8

11 SP MAR is the rate of coverage through a spouse s group policy conditional upon being married in 2000;? SP MAR is the change in rate of coverage through a spouse s group policy conditional upon being married from 1988 to 2000; MAR is the fraction of the sample married in 1988;? MAR is the change in the fraction married from 1988 to 2000;? IND is the change in the independently purchased private non-group coverage rate from 1988 to 2000; and? PUB is the change in the public health insurance coverage rate from 1988 to That is, the change in coverage rates between 1988 and 2000 can be decomposed into the changes in employment, marriage, own-employer provided coverage conditional upon employment, own-employer provided coverage conditional upon not being employed (for example COBRA coverage), spousal coverage conditional upon being married, nongroup coverage and public coverage. Finally, we wish to determine how much of the trends in being uninsured can be explained by controlling for demographics. To do so, we estimate the following linear regressions: UNINSist = Xistβ + STATEs+ YEARt + eist (3) where i indexes individuals, s indexes states, and t indexes years; X is a set of individual characteristics that vary by individual, state, and year including age, age squared, employment status, marriage, race/ethnicity, and presence of children in the household; STATE represents a vector of state dummy variables; and YEAR represents a vector of year dummy variables. The vector of coefficients on the state dummy variables represents the trend in the probability of being without health insurance coverage controlling for demographic characteristics. In order to determine the effect of welfare reform on coverage, we estimate the following linear regression model for being uninsured: UNINSist = Xist a+ ( WAIVER) st + ( TANF / waiver) st + ( TANF / nowaiver) st + STATEs+ YEARt + uist (4) 9

12 where equation 4 is identical to equation 3 with the addition of: WAIVER is an indicator of whether a has in place a welfare waiver in year t and has not yet enacted TANF (please see Table 3 for a summary of when states first implemented TANF or welfare waivers); TANF/waiver is an indicator of whether a state that ever had a waiver had implemented TANF as of year t; and TANF/no waiver is an indicator of whether a state that never had a waiver had implemented TANF as of year t. The coefficients on the waiver and TANF dummy variables measure the effect of welfare reform on the probability of being uninsured for low-skilled women. We also want to know whether any effect of welfare reform on the probability of being uninsured operated through changes in private or public insurance. 1 Therefore we also estimate a set of linear probability models similar to equation 4 but with any private insurance and any public insurance as dependent variables. It seems likely that if welfare reform had an effect on health insurance coverage, this effect would have operated through employment in the sense that welfare reform encouraged employment, which is an important determinant of insurance coverage. This means that the regression just described, which controls for employment, may lead us to underestimate the true effect of welfare reform on insurance coverage. In order to test for this possibility, we estimate equation 4 both with and without controls for employment status. 4. The Data The data for the analyses conducted in this paper come from the annual March income supplements to the Current Population Survey (CPS). The annual March income supplements provide information on demographic characteristics, employment, income, and public and 1 We also plan to estimate specifications that control for the changing generosity of Medicaid during this period. 10

13 private health insurance coverage. Unlike the employment and earnings questions in the basic monthly CPS, which refer to employment in the week before the one in which the survey takes place and usual earnings on the job held during that week, the March supplement questions pertain to employment, earnings and income during the entire calendar year before the year in which the survey takes place. For example, the March 1992 supplement contains information on the longest job held by the respondent in 1991: the number of weeks worked, usual hours worked on this job, total earnings, and industry and occupation codes. Similarly, the health insurance questions in the March supplement ask whether the respondent had coverage from a particular source (for example, through her own employer or from Medicaid) at any time during the previous calendar year. The employment and health insurance questions in the March supplement therefore refer to the same reference period: the calendar year before the year of the survey. Having information on a full year s employment allows us to differentiate between workers with strong and weak attachments to the labor force in a way that is not possible in the basic monthly CPS. Specifically, we are able to categorize every adult in the sample as either a nonworker; a full-time, full-year worker; a part-time, full-year worker; a full-time, part-year worker, or a part-time, part-year worker. Since the strength of a worker s attachment to the labor force is such a critical determinant of insurance coverage, this detail is a great advantage of using the March supplements. One limitation of the March supplement data for our analysis is that information on marital status, education and the presence of children in the household refer to the survey date, rather than to the prior calendar year. Therefore there will be some temporal mismatch between our information on employment and health insurance coverage and our information on (for example) marriage. There is also the possibility, as suggested by Swartz (1986), that survey 11

14 respondents answer the health insurance questions as if they were asked about coverage at the time of the survey, rather than coverage in the previous calendar year. In addition to our baseline results which assume that respondents answer the health insurance questions correctly (referring to the previous calendar year), we will present a set of results as a specification check that assume respondents answer the health insurance questions as if they were asked about coverage at the time of the survey. We also include an analysis of health insurance offering, eligibility, and takeup among workers using the February 1995, 1997, 1999 and 2001 Contingent and Alternative Employment Arrangement Supplements (which we will refer to from now on as the Contingent Work Supplements). In 1995, 1997 and 1999, these supplements were administered to all basic monthly survey respondents who were working at the time of the survey; in 2001 the supplement was administered to three-quarters of all working respondents. 2 The Contingent Work Supplements ask whether the worker is in a firm that offers health insurance to any of its workers; if so, whether the worker is herself eligible for coverage; and if so, whether the worker is in fact covered by her employer s plan. The data from these supplements are not directly comparable to data from the March supplements; the health insurance questions are quite different and are asked only of workers. But the Contingent Work supplements allow us to look at the determinants of coverage through one s own employer in a way that the March supplements do not, so we use them for an analysis of trends in health insurance offering, eligibility, and takeup (enrollment conditional on eligibility) for different groups of workers. 2 Each CPS household is interviewed eight times over the course of sixteen months. The 2001 February supplement was not given to respondents in households being interviewed by the CPS for either the fourth or the eighth time (the outgoing rotation groups ). 12

15 5. Results We begin our discussion of results with a set of descriptive tables. In Table 1, we show demographic characteristics and health insurance coverage rates pooled across all years ( ) for all women and for women by educational attainment (less than a high school degree, high school degree but no college degree, college degree). A large percentage (30.7) of women with less than a high school education does not have health insurance coverage. In contrast, only 16.2 percent of all women overall, 15.7 percent of women with a high school diploma, and 7.9 percent of female college graduates are without health insurance coverage. Only 17.5 percent of women with less than a high school education have own-employer provided coverage compared with 38 percent for female high school graduates and 55.8 percent for female college graduates. A substantial fraction (16.3 percent) of women with less than a high school education are covered through a spouse s employer-provided policy compared with 26 percent for both female high school and college graduates. Roughly 10 percent of women obtain coverage through independently purchased non-group policies a fraction that does not change much with women s education level. A large fraction of low-skilled women 25.5 percent receive public health insurance coverage. This fraction is much lower for women with a high school degree (8.5 percent) and for women with a college degree (2.1 percent). Lowskilled women have different demographic characteristics than women with higher levels of education. For example, women with less than a high school education are less likely to be employed and less likely to be married than are women with more education. In addition, family income for women with less than a high school education is substantially below that of any of the other groups. These differences likely contribute to the lower rates of insurance coverage for low-skilled women. 13

16 In Table 4, we report the fraction uninsured and the fraction receiving coverage from each of the four sources (own-employer, spouse, private non-group, and public) for women by education level in each year from 1988 to For women overall, the fraction uninsured increased each year between 1988 and 1998 (from to 0.182) and fell in 1999 and 2000 (to 0.164). For women with less than a high school education, the fraction uninsured increased even more dramatically from 1988 to 1998 (increasing from to 0.352) and barely decreased in 1999 and 2000 (to only 0.348). The trends in the fraction uninsured for female high school graduates mirror the trends for women overall increasing from in 1988 to in 1998 and falling to by Relative to the large increases in the fraction uninsured for women with lower levels of education, the fraction of female college graduates uninsured changes little over this time period (rising from to in 1998 and falling to in 2000). How did the sources of insurance coverage change for low-skilled women? From 1988 to 1993, the fraction of low-skilled women receiving coverage from their own-employer decreased from to (2.9 percentage points) and the fraction receiving coverage from a spouse s employment decreased from to (9.3 percentage points). At the same time, the fraction on Medicaid increased from to (5.8 percentage points) and the fraction with private non-group coverage increased from to (4 percentage points). Because of these offsetting trends, the fraction of low-skilled women without health insurance increase by only 2.6 percentage points from 1988 to Beginning in 1993, the trends in the sources of coverage changed for these women. From 1993 to 2000, public coverage fell by 7.3 percentage points and the fraction with private non-group coverage fell by 1.3 percentage points. A 2.8 percentage point increase in own-employer coverage mitigated these declines somewhat, but the fraction uninsured increased by another 5.5 percentage points from 1993 to

17 The patters for women with a high school education are similar to those for low-skilled women. Declines in spousal coverage were partially offset by increases in public coverage from 1998 to From 1993 to 2000, declines in public and non-group coverage were completely offset by increases in own-employer coverage. Unlike for low-skilled women, own-employer coverage increased steadily over the 1988 to 2000 periods for women with a high school degree. Even though rates of being without health insurance changed little over the 1988 to 2000 period for female college graduates, there was a compositional change among the sources of health insurance coverage for these women. Own-employer coverage increased by 4.7 percentage points (from to 0.575) during this period while spousal and private non-group coverage fell. Table 5 presents trends in the demographic characteristics shown in Table 1 for all women and for women by education level. For women overall, there was an increase in employment and a decline in marriage between 1988 and These changes were relatively large for lowskilled women: married rates fell be more than 11 percentage points during this period and employment rates increased by more than 6 percentage points. For college educated women, by contrast, marriage rates increased and employment rates decreased in the 1988 to 2000 period. The fact that both marriage and employment are changing over time for low-skilled workers underscores the need to take account of them in our analysis of health insurance coverage. These trends might be expected to have offsetting effects on the probability of insurance coverage for women, since both marriage and employment are both important and positive determinants of coverage. The results of our decomposition of the change in health insurance coverage over the study years for women by education level are reported in Table 6. The two columns of the 15

18 table report the probabilities in 1988 and 2000 that a women (a) has any source of health insurance coverage; (b) is working; (c) is married; (d) has own-employer coverage conditional upon working; (e) has own-employer coverage conditional on not working; (f) has spousal coverage conditional on being married; (g) has private non-group coverage; and (h) has public coverage. The third column reports the change in each probability and the fourth column reports the percentage point contribution of each change to the change in the fraction with health insurance coverage (as determined by equation 2). As the third column in the table reveals, coverage declined for women overall and for each group of women by education level. The largest occurred for women with less than a high school education (8.09 percentage points) followed by female high school graduates (2.38 percentage points) and female college graduates (0.45 percentage points). The factors responsible for these declines in health insurance coverage, however, for women of different education levels. For women with less than a high school education, the 8.09 percentage point decline in coverage is due primarily to declines in spousal coverage among those married (5.62 percentage points), declines in marriage (3.55 percentage points), and declines in own-employer coverage among workers (3.05 percentage points). These declines were partially offset by increases in non-group coverage and increases in employment. Women with a high school education had a smaller decline in coverage (2.38 percentage points) and the sources of this decline were similar to those for women with less than a high school education, with one exception: unlike women with less than a high school education, women with a high school education experienced an increase in own-employer coverage among workers. 16

19 Women with a college degree experienced a mere 0.45 percentage point decline in coverage from 1988 to Important shifts in the sources of coverage occurred during this period for this group, however. Spousal coverage for married women and private non-group coverage rates declined, while own-employer coverage among workers increased. An alternative approach to examining the determinants of the decline in health insurance coverage from 1988 to 2000 is to conduct multivariate analyses of the determinants of the probability of not having health insurance for women by education level. That is, we estimate equation 3 as a linear probability model. Each model controls for age, age squared, marital status, race and ethnicity, the presence of children, employment status, education (when all women are pooled), a set of state dummy variables, and a set of survey year dummy variables. We estimate the model separately for each groups; results are reported in Table 7. Consistent with other studies of the correlates of insurance coverage, we find that individuals who are older, married, and have children are more likely to have insurance; non-whites and Hispanics are less likely to, all else equal. Relative to nonworkers, full-time, full-year workers are less likely to be uninsured while part-time and part-year workers are more likely to be uninsured. This finding suggests that part-time or part-year employment may increase income enough to make individuals ineligible for public coverage but may not offer the employment benefits common in full-time, full-year employment. We can determine how much of the trends in health insurance coverage are the result of changing demographics by comparing the coefficients on the year dummies from Table 7 for each group with the actual rates of not having health insurance. We plot these coefficients and rates in Figures 1 through 4. The actual and adjusted fraction of women without health insurance coverage are almost identical for women with a high school or college education, but not for 17

20 women with less than a high school education. For women with less than a high school education, controlling for changes in demographic characteristics explains roughly half of the increase in the fraction uninsured. For female high school and college graduates, changes in demographic characteristics can explain almost none of the trends in the probability of being uninsured. Before turning to our analysis of the association between welfare and health insurance coverage, it is worth thinking about what the possible effects of reform could be on the health insurance coverage of low-skilled women. First, welfare reform might reduce the insurance coverage of low-skilled women if as caseloads are reduced former welfare recipients become ineligible for public coverage but cannot obtain coverage through employers, spouses employers, or independent coverage. Second, among low-skilled women who were not receiving welfare, welfare reform could increase coverage by encouraging work and marriage formation or it could reduce coverage if competition for jobs offering health insurance increases in response to reduced welfare caseloads. Overall, whether welfare reform could possibly have led to increases in the fraction of low-skilled women with health insurance depends both upon the signs and sizes of any effects on former welfare recipients and on non-recipients and on the relative size of the two groups. According to data from the March CPS, between 10 and 12 percent of women with less than a high school degree received AFDC in the years before welfare reform (see Table 8). Moreover, in the years prior to welfare reform, 95% of low-skilled women on AFDC received health insurance coverage through Medicaid (and the remaining 5% through some form of private coverage with none reporting being uninsured). Therefore, even if we would expect large and negative effects of welfare reform on the probability that low-skilled AFDC recipients have 18

21 health insurance coverage, a small positive effect of welfare reform on the coverage rates for low-skilled women not on welfare could be the dominant effect since the latter group represents 90 percent of low-skilled women. Therefore, it would not be surprising to find either a positive or a negative effect of welfare reform on the insurance coverage rates for low-skilled women. Our analyses of the effects of welfare waivers and the implementation of TANF on the probability of being uninsured, having private health insurance coverage, and of having public health insurance coverage are presented in Tables 9 through 12. The first three rows of each table report the coefficients on the indicators for the individual being surveyed in a year and state with a welfare waiver in place but not TANF, TANF with waivers ever having been implemented, and TANF with waivers never having been implemented. For each outcome, we show the results of two models, the first including a set of employment controls and the second not including them. Table 9 displays the results from our welfare reform models for women with less than a high school education. For these women, welfare waivers are associated with a 2.3 percentage point decline in the probability of being uninsured. Waivers are associated with a 1.9 percentage point increase in the probability of having private insurance and a 0.4 percentage point increase (not statistically different from zero) in the probability of having public health insurance. The coefficients on TANF, in states who ever had a welfare waiver, are similar in magnitude and sign as those for welfare waivers (but are not statistically significant) suggesting that TANF continued doing what waivers had begun. The coefficients on TANF in states that never implemented a welfare waiver are all very small and are not statistically different from zero. Controlling for employment matters little for the effects of either waivers or TANF. 19

22 Table 10 reports the results for women with a high school degree. Waivers are not associated with any change the probability of having insurance coverage. In states that ever had implemented a welfare waiver, TANF is associated with a 1.0 percentage point decline in the probability of being uninsured and a 1.4 percentage point increase in the probability of having private health insurance coverage. Table 11 reports the estimates of equation 4 for women with a college degree. For these women, as expected, waivers and TANF had negligible effects on insurance coverage, although TANF in states that never implemented waivers is associated with a 1.0 percentage point increase in the probability of being without health insurance coverage. Table 12 reports the results of two specification checks we conducted for the subsample of women with less than a high school education. First, as we have already said, we assume in our analyses that the health insurance questions pertain to coverage in the previous calendar year (as the survey instrument specifies). However, it is possible that individuals instead respond with their health insurance status at the time of the survey. While this matters little for our analysis of trends, it could affect our analysis of welfare waivers and TANF because we are using state-year variation in these policy variables. To see how much this matters, we estimate the effect of waivers and TANF on the probability of being without health insurance coverage in columns 1 and 2 as if individuals are responding to the health insurance questions in the CPS using previous year information (as they are supposed to do) and in columns 3 and 4 as if they are responding using current year information. Second, we examine whether changes in the sample matter. Our main results (Tables 1 through 11) use all individuals surveyed. Since some individuals surveyed in any given year were also surveyed the previous year, and some will be surveyed again the following year, one might be concerned that including all individuals in our analyses might yield incorrect standard 20

23 errors. To see how much this might matter, we include only individuals surveyed in months 1 through 4 in columns 1 and 3 and those surveyed in months 5 though 8 in columns 2 and 4. Columns 1 and 3 differ very little in their estimated effects of TANF and welfare waivers on the probability of being without health insurance coverage for women with less than a high school education. This suggests that the interpretation of our results is not dependent upon whether intervals surveyed are responding using current year or previous year information. Columns 1 and 2 are very similar, but columns 3 and 4 yield strikingly different results. In column 4, the estimated effects of waivers and TANF are very small and are not statistically different from zero. While we expected that the inclusion of all individuals in our sample might affect our standard errors, we cannot think of a reason why splitting the sample into two random halves should affect the estimated coefficients on the waiver and TANF variables. The analysis just presented uses the March income supplements to analyze trends in different types of coverage for different groups from 1988 to The Contingent Work supplements allow us to focus on a single component of this analysis coverage for workers from their own employer and analyze whether changes in this component between 1995 and 2001 were driven by changes in offering, eligibility, or takeup. One might argue that employer-sponsored coverage is the most important single type of insurance on which to focus in order to understand trends in the probability of not having any insurance. It is the modal source of coverage for nonelderly adults and is often considered the bedrock of the insurance market. We have already documented its importance in explaining trends in the overall coverage rate between 1988 and Moreover, it is likely helping to drive the trends in spousal coverage for married female high school dropouts that are documented in table 6. Therefore, understanding the trends in offering, eligibility and takeup that drive rates of 21

24 employer-sponsored coverage for workers should significantly increase our understanding of trends in lack of any kind of insurance. The years for which we have data consistent data on offering, eligibility and takeup (1995 through 2001) happen to be years in which employer-sponsored coverage among low-skilled workers was fairly stable, so that the analysis of offering, eligibility and takeup in these years is unfortunately less useful than it would be during a period when more change occurred. Table 13 presents a decomposition of trends in overall insurance coverage from 1995 to 2000 using the March CPS supplements (that is, it is the same as table 6 except using 1995 instead of 1988 as a base year). As Table 13 shows, own-employer coverage among workers contributed less than half a percentage point to the four percentage point decline in the overall coverage rate for women with less than a high school education between 1995 and By contrast, declines in own-employer coverage among workers explained three percentage points of the eight percentage point decline in overall coverage from 1988 to Similarly, for employed men with less than a high school degree, there was a dramatic decline in own-employer coverage between 1988 and 2001 (nine percentage points), but less than half a percentage point decline between 1995 and Since there have been relatively small changes in own-employer coverage among workers during this period, it seems likely that there have also been few changes in its components: offering, eligibility and takeup. With this caveat in mind, we turn to the analysis of the Contingent Work supplements. We begin by comparing the population represented by these supplements to that represented by the March income supplements. The first and most obvious difference is that the Contingent Work supplements include only workers, who comprise about two-thirds of the unweighted basic monthly sample, as shown in table 14. Of course, as we have already documented, the 22

25 probability of working varies with education: less than half of all female high school dropouts in the 1995, 1997, 1999, and 2001 February supplements are workers, compared with 71 percent of female high school graduates, 80 percent of female college graduates, and 66 percent of male high school graduates. Thus the restriction to workers means we are looking at a differentially smaller subset of the population depending on which group we consider. Second, not all workers complete the Contingent Work supplements. The second panel of table 16 shows the fraction of workers with complete supplement data by year and demographic group. On average, in 1995, 1997, and 1999, about three-quarters of workers have complete supplement data; in February 2001, when as we have noted only three-quarters of workers were eligible for the supplement, the overall fraction of workers with complete supplement data is 57 percent. Within each year, more highly educated workers are somewhat more likely than workers with less education to have complete supplement data (for example, in percent of working female high school dropouts have complete supplement data compared to 79.8 percent of working female college graduates). Table 15 shows trends in own-employer coverage, offering, eligibility and takeup for each group of workers over time. The final column of table 15 calculates the contribution of each component to the overall change in the coverage rate according to the formula: OWN EMP= O E2001T2001+ E O1995T2001+ T O1995E1995 where 23

26 OWN EMP = change in fraction of workers with own-employer coverage, O year = fraction of workers who are in firms offering coverage in each year (1995 or 2001) E year = fraction of workers eligible for coverage, conditional on being in firms offering coverage in each year (1995 or 2001) T year = fraction of workers who take up coverage, conditional on eligibility in each year (1995 or 2001). The three terms on the right hand side correspond to the contributions from changes in offering, eligibility and takeup, respectively. As is evident in table 15, own-employer coverage increased slightly on average for both male and female workers between 1995 and Overall, own-employer coverage increased by almost one percentage point for all male workers and by 1.9 percentage points for female workers. These average increases mask the fact that own-employer coverage declined during this period for the lowest-skilled workers: by 1.1 percentage points among female high school dropouts and by 2.2 percentage points among male dropouts. The declines for both male and female high school dropouts were driven almost entirely by declines in takeup among workers eligible for coverage. Male dropouts experienced small declines (less than one percentage point) in offering and eligibility, while female dropouts experienced a small increase in offering and a small decrease in eligibility (both less than one-half of one percentage point). 6. Conclusion Our paper speaks to several related but distinct policy concerns. First, the full impact of welfare reform cannot be assessed unless we know what has happened to health insurance, a critical measure of economic well-being. Second, our study provides insight into the growing 24

27 problem of lack of insurance for adult women, an often overlooked but important group. The health, insurance status and economic well-being of all women, regardless of parenthood, employment, or marriage status are a public health and policy concern. We find that from 1988 to 2000, the probability of being uninsured increased dramatically for women with less than a high school education, increased modestly for women with a high school degree, and increased slightly for women with a college degree. The reasons for these declines differ across groups, however. These findings can be summarized as follows: For women with less than a high school education, the probability of having any health insurance coverage declined by 8.09 percentage points. More than half of this decline 5.62 percentage points is attributable to the decline in the probability of having spousal coverage conditional upon being married. The remainder of the decline is equally due to declines in marriage and declines in own-employer coverage conditional upon working. For women with a high school degree, the probability of having any health insurance coverage declined 2.38 percentage points. As for women with less than a high school education, most of this decline can be attributed to a decline in spousal coverage conditional upon being married. Welfare reform appears have increased private health insurance coverage for women with less than a high school education and for women with a high school education. Welfare reform appears to have led to modest gains in health insurance coverage for less educated women, but these modest gains must be measured against the backdrop of the large declines in health insurance coverage that occurred from 1988 to