Janet Currie & Aaron Yelowitz - Health Insurance and Less Skilled Workers

Health Insurance and Less Skilled Workers

Janet Currie

UCLA and NBER

Aaron Yelowitz

UCLA and NBER

October, 1998

We thank participants in the "Labor Markets and Less Skilled Workers" conference for helpful comments. Paula Hamilton, Chen-ju Chen, and especially Jwa Hong Min provided excellent research assistance.

We use data from the March CPS, supplements to the CPS, and the SIPP to examine trends in private, employer-sponsored health insurance coverage among less skilled workers. Our first finding is that given the problems with the various data sources, it is surprisingly difficult to establish recent trends. However, regardless of the data source, it appears that the post-1970s decline in employer-sponsored coverage has slowed and may even have reversed in recent years. We examine three explanations for the trends in private health insurance coverage: crowdout, a general reduction in job quality, and an increase in the cost of health insurance. We conclude that all three are likely to have played some role in inducing the changes we observe.

Introduction

Most non-elderly Americans get their health insurance through either their own employment, or the employment of family members. Yet, the available evidence suggests that the rate of private health insurance coverage fell throughout the 1980s and early 1990s. Moreover, the coverage rate fell most sharply for families of the least skilled workers. For example, Table 1 shows figures that Farber and Levy (1998, Table 13) report for the percentage of private sector workers aged 20 to 65 who are covered by their own employer's insurance. In addition to documenting the decline in private health insurance coverage since 1979 however, these figures suggest that the decline in private health insurance has halted, or even started to reverse in recent years.

The literature raises two possible explanations for the decline in private health insurance coverage. One is that private health insurance has been "crowded out" by increases in the generosity of public health insurance coverage. A second hypothesis is that the decline reflects "bad jobs getting worse" (Farber and Levy, 1998). A third potential explanation that has received little attention is that the rising cost of medical care is behind the reduction in the prevalence of employer-sponsored health insurance. This last hypothesis offers a potential explanation for the recent turn-around in private health insurance coverage since the increase in health care costs also halted in early or mid-1990s.

This paper will discuss each of these possible explanations in turn. We begin with some theoretical considerations regarding the provision of employer-sponsored health insurance, and an overview of the data available for 1988 to 1997, the period we focus on in this article.

1. Theoretical Considerations

Before discussing the reasons that employer-sponsored health insurance has declined over time, it is helpful to ask why most Americans are covered by employer-sponsored policies to begin with. The main reason is likely to be that employers are able to offer insurance at a lower cost than employees can purchase it in the market. This situation is illustrated in Figure 1. By offering health insurance benefits, employers can move employees from the indifference curve U to a higher indifference curve U'. Figure 1 shows that given a cost advantage, employer-sponsored health insurance can make employees better off, even if it does not offer the optimal wage/benefits bundle for each employee.

There are several reasons for employers' cost advantage. First, a 1943 Internal Revenue Service ruling made compensation in the form of health insurance (and pensions) excludable from taxable income. In contrast, an employee who purchased an individual policy would be taxed on the income used to pay for it (although expenditures on insurance and medical expenses in excess of 7.5% of adjusted gross income are tax deductible). Gruber and Poterba (1996) calculate that the tax-induced reduction in the "price" of employer-provided health insurance averages about 27%.

A second factor creating a wedge between employee and employer costs is selection into the labor force. Poor health increases medical costs and reduces the probability of employment. Thus, the employed are likely to be healthier and lower cost than the unemployed. Moreover, large groups can reduce adverse selection and lower administrative expenses through pooling. These two factors can reduce the cost of providing health insurance in large firms relative to small firms by as much as 35% (Congressional Research Service, 1988).

This simple cost-based model of employer-sponsored health insurance suggests several reasons why not all workers will be covered by their own employer-sponsored insurance, and why less skilled workers will be the least likely to be insured:

a) If health insurance is a normal good then poor people will choose to consume less of it. What must be kept in mind is that in the event of medical catastrophe, indigent care exists even for those who are not insured. Thus, what health insurance buys is routine well care and better quality care in the event of emergencies. Lower income people may be willing to forgo these as luxuries. In terms of Figure 1, they are at an "all wages" corner solution.

b) For some workers, such as women who are covered by a spouse's plan, or those who have access to public insurance programs, the value of employer-sponsored medical insurance may be small or zero. These workers will also be at the "all wages" corner solution.

c) Given heterogeneity in tastes, there will be some workers who would like to purchase a different bundle of health insurance than the one that is offered by their employer. These workers may choose to consume no health insurance rather than purchasing a sub-optimal bundle.

d) Given a progressive tax schedule, the tax savings involved in receiving compensation in the form of benefits are smaller for low-income than for high-income workers. Moreover, changes in marginal tax rates over time will affect the tax advantage to be gained by purchasing health insurance through ones employer.

e) Small companies are less able to take advantage of risk pooling, and thus are less likely to offer insurance. In fact, in 1993 94.3% of companies with over 50 employees offered health insurance to at least some of their employees, compared to only 42.2% of companies with less than 50 employees (NCHS, 1997). Less-skilled people are more likely to work for small companies--73.7% of firms with fewer than 10 employees report that over half of their employees earned less than $5 per hour or less than $10,000 per year. The comparable figure for firms with over 50 employees was only 11.1% (NCHS, 1997).

The importance of these selection effects may vary over the business cycle if for example, those in ill-health are more likely to work during booms than in busts, or if small businesses are also more likely to spring up in boom times. Selection effects may also account for the fact that many firms impose waiting periods for health insurance on new employees, and/or exclude part-time workers from coverage. Seventy-four percent of establishments have minimum work hours requirements for health insurance eligibility and 70.6% have waiting periods for new employees. The average waiting period is 91 days (NCHS, 1997).

2. Data

Our analysis of the recent evolution of health insurance coverage will rely on data from three sources: The annual March Current Population Surveys (CPS) from 1988 to 1997; CPS Benefits Supplements that were conducted in May 1988, and April 1993 as well as the CPS Survey of Contingent Work Supplements conducted in February 1995 and February 1997; and the Survey of Income and Program Participation (SIPP) covering the years 1989 to 1995.

These data sets have various strengths and weaknesses. The March CPSs are one of the main sources of information about changes in health insurance over time, since they have included questions about health insurance coverage since 1980. However there are several issues that complicate analysis of these data. First, while the questions pertain to health insurance over the past 12 months, many analysts have concluded that people tend to answer them as if they referred to contemporaneous or more recent health insurance status. For example, Shore-Sheppard (1996) compares data from the 1988 and 1994 waves of the survey to information from the 1987 National Medical Expenditure Survey and the 1993 Benefits supplement and concludes that the March CPSs can be interpreted as point-in-time coverage rates as of a window between December and March.

A more serious problem is that the insurance questions have been overhauled twice, once in 1988 and once in 1995. Swartz (1997) provides a detailed discussion of the 1995 changes (as well as some discussion of the 1988 changes). Briefly, the wording of the questions changed, the ordering of the questions changed, and new questions about coverage by someone outside the household were added. In addition, the sampling frame of the CPS is changed every 10 years to reflect results from the most recent Census. This change also occurred in 1995. Swartz argues that the various changes to the questionnaire are likely to have caused more people to respond that they had private insurance coverage, Medicaid coverage, or military health care (CHAMPUS). Since the number of people without health insurance is calculated as a residual, these changes would have caused a reduction in the number of uninsured, other things being equal.

However, population growth between the 1980 and 1990 censuses occurred disproportionately in the south and southwest, where states have relatively ungenerous Medicaid eligibility criteria and relatively high proportions of employers who do not offer health insurance. Thus, the changes in the sampling frame that were introduced at the same time as the questionnaire changes would have been expected to reduce the number of people with Medicaid and private health insurance. Since there are a large number of military bases in the southwest, these changes would also have been expected to increase the fraction of the population with military health insurance.

Trends in health insurance coverage for the entire population calculated using data from the March CPSs are shown in the top part of Table 2. The first column shows the fraction of the population with health insurance coverage from any source. These figures indicate a very gradual increase in the fraction of people without insurance coverage. The next column shows the fraction with any private coverage, while the third shows the fraction with employer-based health insurance. The difference between these two columns reflects privately purchased insurance policies such as Blue Cross/Blue Shield. The next two columns show the fraction of the population covered by their own employer's health insurance and by a spouse's health insurance, respectively. Those who have employer-based coverage which is not their own or their spouses are virtually all children covered under parent's policies. Finally, the last row shows a 50% increase in the fraction of the population covered by Medicaid, the public health insurance program for low-income women and children.

The 1995 changes to the CPS would have been expected to affect the numbers calculated for 1994. These numbers indicate that between 1993 and 1994 the number of people with employer-sponsored health insurance actually rose 3.4 percentage points, reversing the 1988 to 1993 trend. Although it is not shown, Swartz comments that the CPS also showed increases in the number of people with military coverage, despite a decrease in the number of armed forces personnel. It is likely that these changes are due at least in part to the changes discussed above.

Further changes to the March CPS health insurance questions took effect in 1996, which affected the 1995 coverage numbers. These included a) the addition of separate questions for privately purchased, non-employer health insurance such as Blue Cross, b) questions designed to identify multiple, concurrent sources of coverage, and c) new questions about health insurance coverage in the current week. Although the addition of these questions represents a potentially large improvement in our knowledge of health insurance coverage, it could have changed respondent's answers to the new questions in unknown ways. Swartz notes for example, that (as shown in Table 2) the fraction of the population covered by Medicaid showed no growth between 1994 and 1995 even though administrative records show continuing growth in the caseload.

In view of the potential difficulties involved in establishing trends using the CPS data, we have also analyzed data from the SIPP. This survey is similar in terms of size and representativeness to the March CPS, and the questions have not changed since the 1990 panel. The SIPP is a panel survey in which a new panel is introduced each year. Each household in the SIPP is interviewed at four month intervals (known as "waves") for approximately 32 months. We use all the waves from the 1990, 1991, 1992, and 1993 SIPP panels which cover the period from October 1989 to October 1995. These 4 panels interviewed approximately 14,300, 14,000, 19,600, and 19,890 households, respectively. Regression models estimated below correct the standard errors for the fact that there are repeated observations on the same households.

The SIPP provides information on the economic, demographic, and social situations of surveyed household members. Although the SIPP asks about private health insurance coverage and Medicaid coverage in every month, it is well known that many respondents tend to give the same answer for every month within a 4 month interval (c.f. Blank and Ruggles, 1996). Thus we examine responses from January, April, July, and October.

Although the SIPP questions are not as comprehensive as latest March CPS questions, they are potentially more useful for detecting trends because they remained constant. The SIPP survey instrument (and data set) contain information about a) whether the respondent was the primary policy holder of a policy, or was covered by a policy in someone else's name, b) whether the coverage was through a current employer or union, former employer, or other source (such as the military), c) whether the health plan was an individual or family policy, and d) whether the respondent was covered by government programs such as Medicaid or Medicare.

These questions on insurance coverage were linked to the work history topical module (asked in waves 1 or 2 during our sample period). This module allows us to construct measures of industry, occupation, job tenure, firm size (we use firm size at "all locations"), and union coverage. Tenure and firm size are problematic in the March CPS, and the CPS supplements do not ask about union coverage in a consistent way.

The second half of Table 2 shows population trends in health insurance coverage calculated using the SIPP. Compared to the March CPS, the SIPP shows a somewhat more modest decline in rates of private health insurance coverage and employer provided health insurance coverage from 1989 to 1993. There is no sign of the upswing in coverage after 1993 that was evident in the CPS numbers, lending support to the idea that this upswing is an artifact of the changes in the CPS questionnaire. The SIPP shows persistently higher rates of private health insurance coverage than the CPS, although the series become closer after 1993. Thus, to the extent that the changes are thought to have improved the accuracy of the CPS, they suggest that the pre-1994 SIPP numbers are more accurate than the pre-1994 CPS numbers.

Since the focus of this paper is on "workers" we have also recalculated the figures shown in Table 2 for two groups: all adults age 25 to 64, and all adult workers aged 25 to 64. We focus on workers 25 to 64 in order to abstract from college students who may still be covered by their parent's health insurance. Given the periodicity of the data, workers are defined somewhat differently in the CPS and the SIPP. In the former, a worker is someone who has worked at least one week in the past year. In the latter, a worker is someone who has worked in the past month. The first part of Table 3 indicates that when we examine all adults, the rates of insurance coverage are quite similar in the CPS and the SIPP, especially in 1994 and 1995. However, rates of private health insurance coverage are consistently somewhat higher in the SIPP, while rates of Medicaid coverage are somewhat lower. This example provides yet another illustration of the importance of questionnaire design.

The second half of Table 3 shows that the definition of "worker" is also important. The CPS definition includes more people with weak labor force attachments, low probabilities of health insurance coverage, and high probabilities of being covered by Medicaid. Rates of health insurance coverage are 5 percentage points higher in the SIPP and rates of Medicaid coverage are almost 50% lower.

The main message of these tables however, is that one finds much less evidence of a decline in private health insurance coverage in the SIPP than in the March CPSs, and that there is little evidence of decline in either data set after 1993.

Additional problems with the March CPSs include a change in the definition of firm size (which as we saw above is closely related to the probability of health insurance coverage) which makes it difficult to compare waves of the survey before and after 1995, and the fact that it is difficult to determine job tenure.

The CPS supplements offer a third source of information. The supplements ask about employer-provided health insurance in the survey week. They ask first whether a person's employer offered insurance to anyone in the firm, and then whether the employee is covered by that health insurance. If the employee is not covered, inquiries are made about the reasons. Employees are also asked whether the employer pays for some or all of the insurance. In addition, employees are asked about other benefits such as pensions. In fact, the questions about pension coverage are very similar to those about health insurance, asking about offers, coverage, and reasons for lack of coverage. These supplements also include information about tenure on the job, though they suffer from the same inconsistency in the firm size questions as the March CPSs.

A potential drawback to the use of the benefits supplements is that the 1988 supplement differs slightly from the 1993 supplement, which in turn is quite different from the 1995 and 1997 supplements. In particular, the 1988 and 1993 supplements first asked whether a person's employer offered health insurance, then whether the person was covered (and if not why not), and finally about whether the person had health insurance from other sources. Beginning in 1995, the sequence of questions was changed so that employees were first asked whether they had any health insurance, and then whether it was through their employer. If they did not have insurance through their employer, they were asked whether the employer offered insurance, whether they were eligible, and why they were not covered. Question wording also varied from year to year. It is not clear what the net effect of these changes is likely to have been, but it does cast some doubt on the use of the Supplements for trend analyses.

3. Crowdout

The model discussed above suggests that people will be less likely to purchase health insurance through their employers when the "non-private health insurance" alternative becomes more attractive. In the late 1980s and early 1990s, the generosity of the public health insurance coverage offered to lower income children and pregnant women under the Medicaid program increased greatly. Moreover, the period of greatest expansion of the Medicaid program corresponds with the period of most rapid decline of employer-based health insurance coverage. Thus, it is natural to suspect that the two phenomena are related and that increases in eligibility for public insurance under the Medicaid program may have "crowded out" private health insurance coverage.

The Medicaid expansions have been extensively discussed elsewhere (c.f. Yelowitz, 1995; Currie and Gruber, 1996a; Currie and Gruber, 1996b; Cutler and Gruber, 1996). Briefly, a series of federal laws first gave states the option, and then required them to raise the income-eligibility thresholds for Medicaid coverage for pregnant women, and various age-groups of children. Because states started with very different levels of generosity to begin with and took up these federal options at different rates, there was a great deal of variation in income cutoffs both across states and within states over time which could be used to identify the effects of the expansions. By April 1990, states were required to cover children up to age six in families with incomes up to 133% of the federal poverty line. Moreover, effective July 1991, states were required to cover all children under age 19 (born after Sept. 30, 1983) whose family incomes were less than 100% of poverty. By 1992, states were also required to cover all pregnant women (from the date of verification of pregnancy) with incomes less than 133% of poverty. Many states have also chosen to extend coverage of these groups further, using state-only funds.

As Tables 2 and 3 showed, Medicaid coverage has increased while the prevalence of employer-sponsored health insurance coverage has fallen. While these figures are suggestive, they do not prove that the relationship between increases in Medicaid coverage and decreases in private health insurance coverage was causal. We have already observed that the declining trend in private health insurance coverage predates the Medicaid expansions. Shore-Sheppard (1996) observes that there were increases in reported Medicaid coverage, and decreases in private health insurance coverage even among single, childless males, a group that one would not expect to have been greatly affected by the Medicaid expansions to pregnant women and children.

Nevertheless, most observers agree that crowdout exists, although the magnitude of the measured effect has been the subject of much debate (c.f. Cutler and Gruber, 1996, 1997; Shore-Sheppard, 1996, 1997; Dubay and Kenney, 1997; Yazici and Kaestner, 1998). The measured effect of crowdout depends on several factors:

a) How crowdout is defined. Cutler and Gruber (1996) conclude that 3.5 million people gained public coverage and 1.7 million lost private health insurance coverage as a direct result of Medicaid expansions that occurred between 1987 and 1992. Dubay and Kenney calculate the reduction in private insurance coverage as a share of the total increase in Medicaid enrollments was 22%. This number is lower than Cutler and Gruber's estimate because much of the increase in Medicaid coverage over the period was among people who would have been eligible even in the absence of the Medicaid expansions. Shore-Sheppard (1996) asks what fraction of the total decline in private insurance coverage over the 1987 to 1992 period resulted from the Medicaid expansions? Since employer-sponsored insurance coverage was declining even among those who were ineligible for the expansions, this figure is only 15%. All of these studies were based on data from the March CPSs.

b) What period crowdout is measured over. In a revision of her earlier work, Shore-Sheppard (1997) finds that adding the years 1994 to 1996 to her time period doubles her estimate of the extent of crowdout from 15 to 30%. One should expect estimates of crowdout to be sensitive to the sample period for several reasons. First, as the generosity of public insurance increases, the composition of newly eligible households changes. Covering the poorest households will not cause crowdout because most of these families do not have the option of purchasing private employer-sponsored health insurance to begin with. At the other end of the spectrum, relatively well-off families with insurance that they believe is superior to Medicaid will be unlikely to make the switch.

A second related issue is that families who do not know that they are eligible for Medicaid will not drop private health insurance coverage in order to take public coverage. The evidence suggests that although in 1994 and 1995, 39% of births were paid for by Medicaid, many women did not take advantage of the free prenatal care provided by the program (NGA, 1997; Ellwood and Kenney, 1995). A possible reason is that they did not learn of their eligibility until they arrived at the hospital to deliver.

As a second example, recent Census estimates indicate that as many as one million children may have already lost Medicaid benefits as a result of their parents leaving the welfare roles. In Wisconsin and two other states with aggressive programs to get people off cash welfare, Medicaid enrollments have dropped by 40 to 50% among those who have been forced off the roles. This is despite the fact that under the new Medicaid rules, most of these children remain eligible. The problem seems to be that neither welfare recipients nor their case workers know about the Medicaid expansions (Rubin, 1997). Knowledge about increases in eligibility is likely to increase over time with consequent increases in the possibilities for crowdout.

c) The Data Source. Most of the work on crowdout to date has been conducted using the March CPSs. Given that both the levels and the trends in health insurance coverage are sensitive to the way these questions are asked, it is likely that reliance on different data sets will generate somewhat different answers.

4. Bad Jobs Getting Worse

While crowdout is important, it obviously cannot account for the entire downward trend in private employer-sponsored health insurance coverage. In this section, we consider an alternative hypothesis, which is that the decline reflects "Bad Jobs Getting Worse". The literature on wage inequality suggests one method of operationalizing the concept of a "bad job". This literature finds that among both men and women, wages for the least skilled workers have been falling, while those for the most skilled workers have been increasing, leading to growing wage inequality over the past 25 years (c.f. Juhn, Murphy, and Pierce, 1993; Bernstein and Mishel, 1997). Moreover, although wives earnings tend to reduce income inequality, family income inequality has also been increasing over time with increases in female headship and higher returns to college education playing key roles (Cancian and Reed, 1997; Bradbury, 1996).

As discussed above, if health insurance is a normal good, people will demand less of it when they are poorer and more of it when they are richer. Therefore, trends in wage and income inequality suggest that one might expect to see reductions in private health insurance coverage among less skilled workers, but increases in health insurance coverage among more skilled workers, other things being equal. Instead, the figures in Table 1 (which were computed using the CPS Supplements) showed that among workers, the decline in rates of own employer-provided health insurance coverage between 1988 and 1997 was almost as great among college graduates as among high school dropouts.

If changes in coverage were driven solely by income effects, then one might also expect to see similar patterns for other types of benefits which are purchased through employers. In Table 4, we compare trends in own-employer-sponsored health insurance coverage to trends in pension coverage for workers with at least some college education and those without. We focus on this comparison for two reasons: First, along with health insurance, pension coverage is one of the costliest and most common components of benefits packages. Second, people obtain pension coverage through employers for some of the same reasons that they obtain health coverage--favorable tax treatment, and risk pooling.

Following Farber and Levy (1998), Table 4 also shows whether employees work for a firm that offers health insurance, whether the employees are eligible for an employer-provided plan, and whether the employer offers a pension plan to any employees. (We do not delve into eligibility for offered pension plans, since this is governed by vesting provisions).

We extend Farber and Levy's analysis by examining men and women separately. Previous research (c.f. Currie and Chaykowski, 1995 and Currie, 1997) indicates that gender is an important determinant of benefits coverage. As Table 4 shows, there are large gender differences in benefit "offers" and even greater differences in propensities to take up benefits. Women also make up the bulk of the part-time workforce, suggesting that it would be useful to distinguish between the genders when analyzing the effects of part-time status.

Table 4 confirms that there have been modest declines in health insurance coverage among both men and women, and that these declines are slightly larger among less skilled workers. The declines in coverage among less skilled workers appear to be due to changes in both "takeup" and eligibility, while among more skilled workers the changes primarily reflect reductions in takeup. In contrast to these trends in health insurance coverage, there have been increases in the fraction of workers in establishments that offer pension coverage (except among low-skilled men), and in the fraction of workers covered. These gains have been particularly pronounced among college-educated workers. These trends suggest that changes in health insurance coverage are not primarily driven by income effects (although changes in pension coverage may be).

Farber and Levy (1998) interpret "bad jobs" not as jobs held by less skilled workers but as either part-time or low-tenure jobs. They break down the overall decline in employer-sponsored health insurance coverage into 12 components: First they define four groups of workers: "old" full-time, "new" full-time, old part-time, and new part-time. Old workers are those who have been in their jobs for over a year, while full-time refers to those who usually work more than 35 hours per week. For each group of workers, they calculate the share of the decline associated with changes in the fraction of workers in establishments that offer insurance to some workers; changes in the fraction of workers in such establishments who are eligible for coverage; and changes in the fraction of these workers who take up coverage. The employment-share weighted sum of these components over the groups is equal to the overall decline in insurance coverage.

Using this technique, and the fact that they find virtually no change in the fraction of workers who are low-tenure or part-time over the sample period, they calculate that half of the decline in own-employer-sponsored health insurance coverage is due to changes in takeup among old full-time workers. Most of the rest is due to changes in eligibility for insurance among part-time and new workers, although these reductions in eligibility appear to be partially offset by increases in the fraction of such workers in firms that offer insurance.

The decomposition suggested by Farber and Levy does not allow us to test the statistical significance of the hypothesized changes in the effects of worker characteristics on insurance coverage. Table 5 offers a different look at the effects of low tenure and full-time status. Part 1 of this table shows coefficient estimates from regressions of private health insurance variables on demographic characteristics, whether the worker has tenure less than one year, whether the worker is fulltime, and industry and occupation dummies.

Estimates are shown for each of our four gender/education groups. Data from the 1988 and 1997 supplements have been pooled, and interactions are included between the dependent variables and a dummy variable for 1997. This specification allows us to test whether the coefficients on full-time and low tenure vary over time. Since "bad jobs" may also have been getting worse in terms of other benefits, we also include coefficients from regressions with pension coverage as the dependent variable.

Table 5 confirms that as Farber and Levy suggest, people who are working part-time and/or have tenure less than one year are much less likely to work in places that offer health insurance coverage. They are also less likely to be eligible for coverage and to have private health insurance coverage. It is remarkable that low tenure has almost as great a negative effect on probability of health insurance coverage as it has on pension coverage.

There is little evidence in Table 5 that the penalty associated with being a new worker has changed over time. None of the estimated coefficients on interactions with "low tenure" are statistically significant. There have been changes in the importance of full-time employment however.

Among less educated men, there is a significant positive interaction between full-time and the 1997 dummy for both health insurance coverage and pension coverage. Among less-educated women, the advantage of being full-time in terms of health insurance coverage has fallen over time. The relative improvement in the position of less-educated part-time women appears to be associated with an increased probability of working at a firm that offers health insurance coverage. Among more highly educated women, there have been increases in the probability of being eligible for health insurance coverage that are associated with full-time status. But these changes in eligibility do not seem to have translated into any change in the probability of coverage among full-time relative to part-time college-educated women workers.

Because of our concerns about conducting trend analyses using the CPS supplements, we have extended this analysis using the SIPP. Part 5 of Table 5 shows estimates from linear probability models in which own-employer health insurance coverage is a function of the variables described above, as well as an indicator for firm size less than 100 and union coverage. Being in a large firm and having union coverage can be viewed as additional indicators of a "good job". These variables are interacted with a dummy variable equal to one if the year is 1993 or greater.

The main effects of low tenure and fulltime status are qualitatively similar to those reported above, although the effects of low tenure are much weaker. Being in a larger firm and having union coverage have large positive effects on the probability of health insurance coverage. However, very few of the interactions are statistically significant. The effect of low tenure decreased slightly for more educated men, while the positive effect of union coverage increased among less educated men and women.

Part 6 of Table 5 shows a similar analysis of the probability of Medicaid coverage using SIPP data. These estimates show that the probability of Medicaid coverage is higher for part-time, low tenure, non-union workers. Firm size has a significant effect for less-educated women. The interactions indicate that full-time status had a less negative effect on coverage among less educated male and female workers over time, while the effect of being a low tenure worker grew among less educated workers and female workers with over 12 years of education. These patterns suggest that among both men and women, more low tenure and full-time workers were becoming covered by Medicaid over time. A possible explanation is that the Medicaid expansions to women and children were accompanied by other (here-to-fore unresearched) measures that made Medicaid coverage more accessible to men.

We also use the regressions underlying Table 5 to test for whether the coefficients on marital status, the number of children, and the presence of children of different age groups in the household have changed over time in a manner consistent with the crowdout hypothesis described above. The coefficients from regressions with coverage as the dependent variable are shown in Table 6. As in Table 5, the first part of the table shows estimates from regressions based on the CPS supplements, while the second part shows estimates based on the SIPP.

The first part of the table contains one suggestive finding for less educated women: In 1988, these women were 14% more likely to have health insurance coverage through their employers if they had an infant in the household. By 1997, however, this effect had been entirely wiped out. This finding is echoed in the models estimated using SIPP data, although the size of the effects is much smaller. Given that infants whose deliveries are paid for by the Medicaid program are covered for one year after delivery, and that over 40% of births are now paid for by Medicaid, we might expect the strongest crowding out for infants of less skilled workers.

In the CPS supplements, the negative effect of marital status on the probability of health insurance coverage became more negative over time for all four groups, but the coefficients are larger for the more educated than for the less educated. Hence, this finding is more suggestive of households economizing by eliminating duplicative coverage than of crowdout. In the SIPP, the effects of marriage are qualitatively similar, but the interaction terms are not statistically significant except for college-educated men.

4. Changes in the Price of Health Insurance

The simplest economic explanation for a decline in the number of people purchasing a product is that its price has gone up. Cutler and Sheiner (1997, page 1) note that "After decades of double-digit increases, health insurance cost growth has essentially ground to a halt". As we have seen, the long decline in rates of private health insurance coverage also seems to have leveled off, which suggests that the break in the two trends is related.

One might argue however, that the decline in private health insurance coverage has been very gradual relative to the rapid run-up in health care costs. This may be due to the fact that health care costs increase both the costs of insurance, and the value of insurance. Moreover, the value of health insurance is likely to increase most rapidly for those who have assets to lose in the event of a health shock, suggesting that the poor may be most likely to respond to increases in health care costs by dropping health insurance coverage.

It is difficult to get the price data necessary to estimate the demand for health insurance. Studies such as the RAND Health Insurance experiment focus on the demand for health care where the treatment is the type of insurance policy. Moreover, the employee's choice of insurance is complicated by the fact that it is only one element of a bundle of goods that is chosen when he or she accepts employment at one firm rather than another.

One option we explored was using state-level variation in the costs of health care and in the fraction of firms offering health insurance to try to identify the effects of costs. National Center for Health Statistics (1997) reports that the fraction of firms offering health insurance varies widely from state to state. The rate approaches 55 to 60% in states such as Delaware and Pennsylvania, but is closer to 30% in states like Mississippi and Arkansas. State-level data about expenditures on medical care in 1985, 1990, and 1992 is available from Levit et al. (1997).

We examine the relationship between state-to-state variations in medical expenditures (measured using personal health care expenditures as a percent of gross state product) and in the probability of private health insurance coverage, eligibility, and offers in Table 7. Because some of the variation in rates of private health insurance coverage may be due to demographic or compositional differences between the states, the dependent variable we focus on is a set of state dummies from regressions similar to those reported in Tables 5 and 6 above (similar except of course that they included state dummies). We estimated separate models for each year of the three CPS supplements and then pooled the state dummies from each year to yield 153 observations per outcome (the District of Columbia is included as a state). Because the higher rates of private health insurance coverage could lead to higher medical expenditures (and because of data limitations), our independent variable is health care expenditures lagged three years.

The results indicate that there are negative correlations between state-level health care expenditures, eligibility and coverage. However, there is no relationship between the probability that health insurance is offered and expenditures. Moreover, adding year dummies to these models reduces all correlations to statistical insignificance suggesting that it is the time trend in the expenditure data that is correlated with employer-provided health insurance, rather than the cross-state variation in these expenditures.

A second way to get at the issue of increasing costs of private health insurance is to ask whether employees appear to be paying more now for health insurance than they used to. Unfortunately, neither the CPS or the SIPP asks questions about the cost of health insurance. The CPS does ask whether the employer pays some or all of the cost of health insurance, but approximately 95% of employees with employer-sponsored insurance report that the employer does so in all years of the survey.

We attempted to get at this issue by estimating the cost of health insurance in terms of foregone wages. In principle, employees "pay" for benefits by accepting lower wages than those who do not receive benefits. In practice, it has proven extremely difficult to find evidence of these "compensating differentials" in pay (see Currie and Madrian, 1998 for a discussion). The fundamental problem is that "good jobs" tend to have high salaries, superior benefits packages, and more skilled employees.

In order to isolate the effect of benefits provision on wages, it is necessary to find instruments that affect benefits provision without having any independent effect on wages. The discussion of crowdout above suggests that the Medicaid expansions could provide a plausible set of such instruments, at least for some groups such as less-educated women workers. We have estimated the effects of the Medicaid expansions on the private coverage rates of workers in our sample. Our measure of the Medicaid expansions was the income limit within a state for children in an age group (such as the limit for children aged 1 to 4) interacted with a dummy variable indicating whether the household had children in that age group.

If crowdout of the parent's own employer provided health insurance exists, then these interactions should be negative. They were not statistically significant, however. These insignificant "first stage" estimates may be due to the fact that we focus only on workers, who are less likely to take up Medicaid than nonworkers. Also, our measure of the Medicaid expansions refers to changes in the children's program, but the employer coverage refers to the parent. Only in the case when the parent completely drops private coverage would we be able to detect effects. If the parent simply switched from a "family" policy to an "individual" policy, we would be unable to detect this crowding out. In summary, we were unable to estimate compensating wage differentials because the instruments were not significant in the first stage.

6. Discussion and Conclusion

We began this research with the belief that the decline in private-employer sponsored health insurance was a continuing problem, especially among less skilled workers. But, our analysis paints a much more ambiguous picture. Rates of employer-sponsored health insurance coverage are sensitive to the way that insurance questions are posed, to the way that "workers" are defined, and to the age range of workers examined. Regardless of these problems, however, we find that in recent years the decline in private employer-sponsored health insurance coverage has slowed, and may even have reversed.

Neither crowdout, nor a deterioration in the quality of jobs available to the less skilled seems likely to fully explain recent time-series trends in health insurance coverage. A simple explanation that has been rather overlooked, probably because of data limitations, is that rising health care costs have driven much of the reduction in private health insurance coverage. It is likely that all three explanations contributed to a decline in employer-provided health care coverage from its level at the end of the 1970s.

Three factors suggest however, the decline in employer-sponsored health insurance coverage could accelerate again in future. First, although health care costs stopped rising in the early 1990s, this may prove to be a mere hiatus. Cutler and Sheiner (1997) point out that much of the cost-savings from managed care and hospital reorganization has already been realized, and that technological change is the underlying force driving health care costs. Although the value of health insurance also increases with rising health care costs, a future run-up in such costs could drive many families to the point where the cost of insurance becomes prohibitive. Further research on the link between health insurance costs and coverage is certainly warranted.

Second, the increase in wage inequality that began in the 1970s is continuing into the 1990s with the result that there are more relatively "low wage" workers than ever. Moreover, if time limits on welfare are effective, they will push many unskilled women into the work force, again increasing the number of less skilled workers (see Moffitt's discussion in this volume). Although past patterns in benefits coverage do not appear to have been driven primarily by income effects, the "bad jobs getting worse" phenomena could become more important in future.

Third, crowdout is likely to become more important over time, as people become aware of the public insurance option. In addition to outreach campaigns, administrative changes designed to make Medicaid more accessible have also been undertaken recently in many states. However, little is known about their effects. Growing knowledge about the Medicaid alternative may interact with rising health care costs and the falling relative wages of less-skilled workers to increase crowdout.

In view of the attention that has been paid to the Medicaid expansions to pregnant women and children, the fact that Medicaid enrollments have been rising for men as well as women is surprising and remains something of a puzzle. A possible explanation is that states have made less heralded changes to their programs which have made it easier for men as well as women and children to receive benefits. This issue deserves further investigation.

It is striking that private employer-sponsored health insurance coverage has declined while the fraction of workers in establishments that offer health insurance coverage has not. The underlying message for policy makers may be that many poor people will decline health coverage insurance even at the subsidized rate that employers typically offer. In this case, legislation requiring employers to offer health insurance to all employees will not result in universal coverage and governments are likely to remain the suppliers of last resort for indigent medical care.

References

Bernstein, Jared and Lawrence Mishel. "Has Wage Inequality Stopped Growing?," Monthly Labor Review, 120 #12 (Dec., 1997) 3-16.

Blank, Rebecca and Patricia Ruggles. "When do Women Use Aid to Families with Dependent Children and Food Stamps? The Dynamics of Eligibility Versus Participation," Journal of Human Resources, 31 #1 (Winter, 1996) 57-89.

Bradbury, Katherine. "The Growing Inequality of Family Income: Changing Families and Changing Wages," New England Economic Review, (July/Aug. 1996) 55-82.

Cancian, Maria and Deborah Reed. "Assessing the Effects of Wives Earnings on Family Income Inequality," Review of Economics and Statistics, 79 #4 (Nov. 97) 665-669.

Congressional Research Service. Costs and Effects of Extending Health Insurance Coverage, (Washington D.C.: Government Printing Office, 1988).

Currie, Janet. "Gender Gaps in Benefits Coverage," in

The Handbook of Human Resource Management, Daniel Mitchell and Mahmood Zaidi (eds.) (Greenwich CT: JAI Press, 1997).

Currie, Janet and Richard Chaykowski. "Male Jobs, Female Jobs, and Gender Gaps in Benefits Coverage in Canada," Research in Labor Economics, 14 (1995) 171-210.

Currie, Janet and Jonathan Gruber. "Health Insurance Eligibility, Utilization of Medical Care, and Child Health," The Quarterly Journal of Economics, 111 #2 (May 1996a) 431-466.

Currie, Janet and Jonathan Gruber. "Saving Babies: The Efficacy and Cost of Recent Changes in the Medicaid Eligibility of Pregnant Women," Journal of Political Economy, 104 #6 (December, 1996) 1263-1246.

Currie, Janet and Brigitte Madrian. "Health, Health Insurance, and the Labor Market," in Handbook of Labor Economics, Orley Ashenfelter and David Card (eds.) (New York: North Holland, 1998).

Cutler, David and Jonathan Gruber. "Does Public Insurance Crowd Out Public Insurance?," The Quarterly Journal of Economics, 111 #2 (May 1996) 391-430.

Cutler, David and Jonathan Gruber. "Medicaid and Private Insurance: Evidence and Implications," Health Affairs, 16 #1 (Jan/Feb. 1997) 194-200.

Cutler, David and Louise Sheiner. "Managed Care and the Growth of Medical Expenditures," NBER Working Paper #6140, Cambridge MA (August 1997).

Dubay, Lisa and Genevieve Kenney. "Did Medicaid Expansions for Pregnant Women Crowd out Private Coverage?," Health Affairs, 16 #1 (Jan/Feb. 1997) 185-193.

Ellwood, Marilyn R. and Genevieve Kenney. "Medicaid and Pregnant Women: Who is Being Enrolled and When?" Health Care Financing Review, 17 #2 (Winter, 1995) 7-28.

Farber, Henry and Helen Levy. "Recent Trends in Employer-Sponsored Health Insurance Coverage: Are Bad Jobs Getting Worse?," Working Paper #6709, National Bureau of Economic Research (August, 1998).

Gruber, Jonathan and James Poterba. "Tax Subsidies to Employer-Provided Health Insurance," in Martin Feldstein and James Poterba (eds.) Empirical Foundations of Household Taxation, (Chicago: University of Chicago Press, 1996) 135-164.

Juhn, Chin-hui, Kevin Murphy, and Brooks Pierce. "Wage Inequality and the Rise in Return to Skill," Journal of Political Economy, 101 #3 (June 1993) 410-442.

Levit, Katharine R., Helen C. Lazenby, Cathy A. Cowan, Darleen K. Won et al. "State Health Expenditure Accounts: Building Blocks for State Health Spending Analysis," Health Care Financing Review, 17 #1 (Fall, 1995) 201-254.

National Center for Health Statistics. Employer-Sponsored Health Insurance, State and National Estimates, (Hyattsville MD: U.S. Department of Health and Human Services, Centers for Disease Control and Prevention, 1997).

Rubin, Alissa. "Poor Children Falling Out of Medicaid's Safety Net," Los Angeles Times, (November 18, 1997) A1.

Shore-Sheppard, Lara. "Stemming the Tide? The Effect of Expanding Medicaid Eligibility on Health Insurance Coverage," Working paper #361, Princeton University Industrial Relations Section (April 1996).

Shore-Sheppard, Lara. "Stemming the Tide? The Effect of Expanding Medicaid Eligibility on Health Insurance Coverage," Dept. of Economics, University of Pittsburgh xerox, November 1997.

Swartz, Katherine. "Changes in the 1995 Current Population Survey and Estimates of Health Insurance Coverage," Inquiry, 34 (Spring, 1997) 70-79.

Yazici, Esel and Robert Kaestner, "Medicaid Expansions and The Crowding Out of Private Health Insurance," Working Paper #6527, National Bureau of Economic Research (April, 1998).

Yelowitz, Aaron. "The Medicaid Notch, Labor Supply, and Welfare Participation: Evidence from Eligibility Expansions," The Quarterly Journal of Economics, CX #4 (November, 1995) 909-940.

Table 1: Percent Private Sector Workers Covered

by Own Employer's Insurance

Source: Farber and Levy, 1998

All College Some High < High

Graduates College School School

May 1979 71.9 80.6 71.3 71.4 67.3

May 1988 69.1 81.9 68.0 67.2 57.8

April 1993 64.7 77.4 63.8 62.7 47.1

Feb. 1997 64.5 76.0 63.2 61.6 50.2

Note: These numbers were calculated using the CPS Supplements.

Table 2: Trends in Health Insurance Coverage

in the March CPS and in the SIPP

Source: March CPS

Type Coverage: Employer Own Spouse

Any Private Provided Employer Employer Medicaid

1987 87.1 75.5 62.2 31.6 11.4 7.9

1988 86.6 74.7 62.0 31.8 11.3 8.0

1989 86.4 74.6 61.8 31.8 11.1 8.0

1990 86.1 73.2 60.6 31.3 11.0 9.0

1991 85.9 72.1 59.8 30.9 11.0 9.7

1992 85.0 71.1 58.5 30.0 10.8 10.0

1993 84.7 70.2 57.1 30.7 9.4 11.0

______________________________________________________________________

1994 84.8 70.3 60.5 32.0 10.0 12.1

1995 84.6 70.3 60.6 32.1 10.0 12.1

1996 84.4 70.2 60.7 32.1 10.1 11.8

Source: SIPP

Type Coverage: Employer Own Spouse

Any Private Provided Employer Employer Medicaid

1989 86.5 76.1 65.2 32.0 12.2 6.7

1990 87.0 75.7 64.8 32.3 11.9 7.8

1991 87.0 74.4 64.0 31.9 11.8 8.8

1992 86.4 73.3 63.0 30.9 11.7 9.5

1993 85.8 71.9 62.0 30.4 11.5 10.5

______________________________________________________________________

1994 86.0 71.7 62.1 30.6 11.5 11.0

1995 86.5 72.0 62.7 31.1 11.6 11.4

Notes: The lines indicate the date of the change in the March CPS questionnaires. The 1995 changes would have been expected to affect the rates for 1994.

Table 3: Trends in Health Insurance Coverage Among Adults and Workers

Source: March CPS, All Adults 25-64

Type Coverage: Employer Own Spouse

Private Provided Employer Employer Medicaid

1987 79.4 70.6 51.2 19.3 5.0

1988 78.5 70.3 51.3 19.0 5.0

1989 78.4 69.8 51.1 18.7 5.1

1990 77.2 68.6 50.2 18.4 5.7

1991 76.3 68.2 49.8 18.4 6.1

1992 74.9 66.5 48.5 18.1 6.4

1993 74.3 65.4 49.8 15.6 7.0

1994 74.9 68.6 51.3 17.0 7.0

1995 74.7 68.8 51.4 17.2 7.0

1996 74.9 69.0 51.6 17.1 7.1

Source: SIPP, All Adults 25-64

Type Coverage: Employer Own Spouse

Private Provided Employer Employer Medicaid

1989 79.2 71.6 50.9 19.2 4.3

1990 79.1 71.5 51.2 18.9 4.7

1991 78.1 70.8 50.9 18.5 5.2

1992 77.0 70.0 50.0 18.6 5.6

1993 76.0 69.3 49.4 18.5 6.1

1994 75.9 69.6 49.7 18.4 6.6

1995 76.6 70.6 50.6 18.6 6.7

Source: March CPS, Workers 25-64

Type Coverage: Employer Own Spouse

Private Provided Employer Employer Medicaid

1987 84.0 78.0 67.8 14.6 .8

1988 82.8 77.3 63.3 14.0 2.2

1989 83.0 77.4 63.0 14.4 2.4

1990 81.6 76.1 62.0 14.1 2.8

1991 80.9 75.6 61.2 14.4 2.9

1992 79.6 74.0 59.6 14.3 3.1

1993 79.3 73.0 60.8 12.2 3.5

1994 79.6 75.6 61.4 14.0 3.7

1995 79.9 76.1 61.8 14.0 3.6

1996 80.0 76.4 62.3 13.9 3.8

Source: SIPP, Workers 25-64

Type Coverage: Employer Own Spouse

Private Provided Employer Employer Medicaid

1989 87.6 83.6 67.8 14.6 .8

1990 86.8 83.1 67.8 14.1 1.2

1991 86.4 83.0 68.1 13.9 1.5

1992 85.6 82.3 66.8 14.4 1.6

1993 84.9 81.7 66.0 14.5 1.8

1994 84.6 81.5 65.7 14.7 2.1

1995 85.3 82.4 66.5 14.9 2.1

Notes: See Table 2.

Table 4: Own-Employer Benefits Coverage Among

Private Sector Workers, 25-64

All <= 12 Years Ed. At Least Some College

Men Women Men Women

Health Offered

1988 .83 .83 .76 .90 .85

1993 .82 .78 .75 .90 .84

1997 .84 .81 .76 .91 .86

Eligible for HI

1988 .80 .81 .69 .88 .80

1993 .78 .75 .69 .88 .78

1997 .79 .77 .68 .88 .79

Health Coverage

1988 .71 .75 .57 .83 .66

1993 .67 .68 .54 .80 .63

1997 .69 .72 .55 .81 .65

Pension Offered

1988 .64 .63 .56 .72 .66

1993 .65 .58 .56 .75 .70

1997 .67 .61 .56 .76 .72

Pension Coverage

1988 .51 .54 .41 .60 .46

1993 .52 .49 .41 .62 .51

1997 .55 .51 .42 .66 .55

Notes: Source is the CPS Supplements. Means from 1995 are not shown as they are generally very similar to 1997. Means of eligibility and coverage are not conditional on being offered the benefit. The sample excludes non-workers and those in the military and public sectors. All means are weighted using the supplement weights.

Table 5

Coefficients on "Full-time" and "Low Tenure" from Regressions of

Own-Employer Health Insurance Offers, Eligibility, and Coverage

A: Source=CPS Supplements

<= 12 Years Ed. At Least Some College

Women Men Women Men

1. Dependent Variable=Health Offered

Full-time .216 .192 .144 .175

(.015) (.025) (.054) (.021)

Full-time x 1997 -.044 .007 .009 -.027

(.020) (.032) (.018) (.025)

Low Tenure -.099 -.134 -.095 -.059

(.018) (.016) (.017) (.014)

Low Tenure x 1997 -.012 .003 -.001 -.025

(.022) (.021) (.019) (.016)

R-squared .166 .145 .130 .098

# Obs. 9,124 9,935 9,639 10,783

2. Dependent Variable=Eligible for Health Insurance

Full-time .316 .265 .248 .299

(.016) (.021) (.017) (.024)

Full-time x 1997 .009 .052 .094 .000

(.021) (.033) (.020) (.029)

Low Tenure -.216 -.221 -.173 -.153

(.018) (.017) (.019) (.015)

Low Tenure x 1997 -.023 -.032 -.018 -.004

(.023) (.022) (.022) (.018)

R-squared .253 .201 .241 .167

# Obs. 9,068 9,897 9,591 10,748

3. Dependent Variable=Covered by Employer's Health Insurance

Full-time .355 .278 .342 .355

(.017) (.028) (.020) (.029)

Full-time x 1997 -.047 .083 .023 -.032

(.022) (.036) (.024) (.034)

Low Tenure -.236 -.260 -.212 -.178

(.020) (.019) (.022) (.019)

Low Tenure x 1997 -.010 -.036 .000 -.024

(.026) (.025) (.025) (.022)

R-squared .261 .212 .272 .155

# Obs. 8,818 9,623 9,439 10,596

4. Dependent Variable=Pension Coverage

Full-time .227 .094 .227 .169

(.018) (.033) (.022) (.037)

Full-time x 1997 -.024 .137 .017 .062

(.023) (.042) (.026) (.043)

Low Tenure -.295 -.325 -.334 -.338

(.021) (.022) (.024) (.024)

Low Tenure x 1997 -.019 -.009 -.038 -.003

(.027) (.028) (.028) (.028)

R-squared .226 .214 .218 .179

# Obs. 8,739 9,552 9,315 10,426

Table 5, continued

B: Source=SIPP

5. Dependent Variable=Covered by Employer's Health Insurance

Full-time .274 .243 .343 .296

(.007) (.010) (.009) (.012)

Full-time x 1993+ .011 .022 -.003 -.006

(.009) (.013) (.011) (.016)

Low Tenure -.097 -.120 -.081 -.097

(.007) (.008) (.008) (.008)

Low Tenure x 1993+ .013 -.003 -.002 -.034

(.011) (.014) (.014) (.016)

Firmsize < 100 -.171 -.131 -.161 -.117

(.008) (.007) (.008) (.007)

Firmsize < 100 .001 -.000 -.008 -.010

x 1993+ (.010) (.009) (.011) (.010)

Union .201 .211 .158 .148

(.009) (.005) (.011) (.007)

Union x 1993+ .023 .025 .018 .014

(.012) (.007) (.015) (.008)

R-squared .306 .250 .297 .174

#Obs 146,218 160,552 124,671 147,309

6. Dependent Variable=Medicaid

Full-time -.032 -.032 -.011 -.011

(.003) (.005) (.002) (.003)

Full-time x 1993+ -.008 -.022 -.003 -.003

(.004) (.007) (.003) (.004)

Low Tenure .032 .010 .014 .004

(.004) (.002) (.002) (.001)

Low Tenure x 1993+ .027 .017 .022 .007

(.007) (.006) (.006) (.004)

Firmsize < 100 .007 .001 .002 .000

(.003) (.002) (.002) (.001)

Firmsize < 100 -.004 .002 .005 .001

x 1993+ (.004) (.002) (.002) (.001)

Union -.011 -.006 -.006 -.002

(.003) (.001) (.003) (.001)

Union x 1993+ -.002 -.003 -.002 .001

(.005) (.002) (.003) (.002)

R-squared .088 .034 .055 .011

#Obs 146,218 160,552 124,671 147,309

Notes: Source is the CPS Supplements for May 1988 and Feb. 1997, and the SIPP (all years). Models were estimated separately for each group indicated in the column headings. Models also included demographic variables, industry, and occupation as described in the text. The sample consists of workers aged 25-64 and excludes those in the military, those in the public sector, and those with missing data. Standard errors in parentheses.

Table 6

Coefficients on Family Structure Variables from Regressions of

Employer-Provided Health Coverage

A: Source=CPS Supplements

<= 12 Years Ed. At Least Some College

Women Men Women Men

Married -.104 .039 -.133 -.017

(.017) (.017) (.018) (.018)

Married x 1997 -.043 -.045 -.063 -.062

(.021) (.022) (.022) (.020)

# Children -.028 -.010 -.021 .012

(.017) (.014) (.019) (.013)

# Children x 1997 .019 .015 -.021 .009

(.022) (.018) (.022) (.016)

Any child <1 .138 .035 -.024 .020

(.054) (.035) (.047) (.030)

Any child <1 x 97 -.141 -.042 .081 -.025

(.071) (.045) (.055) (.037)

Any child 1-4 .036 .012 -.024 -.011

(.031) (.025) (.033) (.023)

Any child 1-4 x 97 -.037 -.062 .053 .005

(.040) (.032) (.038) (.028)

Any child 5-10 .014 .028 -.016 -.008

(.028) (.024) (.032) (.024)

Any child 5-10 x 97 -.021 -.033 .052 -.017

(.037) (.031) (.037) (.029)

Any child 11+ .008 .014 -.042 -.009

(.029) (.024) (.033) (.024)

Any child 11+ x 97 -.045 -.030 .033 -.007

(.037) (.031) (.039) (.029)

R-squared .261 .212 .272 .155

# Obs. 8818 9623 9439 10596

B: Source=SIPP

Married -.158 .025 -.193 -.013

(.007) (.007) (.007) (.007)

Married x 1993+ .009 -.012 .005 -.027

(.009) (.009) (.010) (.009)

# Children -.030 -.019 -.033 -.014

(.012) (.011) (.014) (.011)

# Children x 1993+ .013 -.003 -.009 .003

(.017) (.015) (.019) (.015)

Any child <1 .025 -.017 .022 .007

(.013) (.035) (.014) (.010)

Any child <1 x 93+ -.031 .002 .016 -.019

(.018) (.015) (.020) (.014)

Any child 1-4 -.011 .012 -.006 .002

(.010) (.025) (.012) (.009)

Any child 1-4 x 93+ -.001 -.027 .008 .017

(.013) (.009) (.016) (.013)

Table 6, continued

Any child 5-10 -.024 .001 -.023 .006

(.009) (.008) (.010) (.008)

Any child 5-10 x 93+ .008 -.011 .001 .009

(.012) (.011) (.014) (.011)

Any child 11+ -.038 -.004 -.028 -.002

(.010) (.009) (.012) (.010)

Any child 11+ x 93+ -.011 .013 -.010 -.001

(.014) (.013) (.016) (.013)

R-squared .306 .250 .297 .174

# Obs. 146,218 160,552 124,671 147,309

Notes: See Table 5.

Table 7

The Relationship Between the State-Level Prevalence of Health

Insurance and Personal Health Care Expenditures

(as a Percent of State Product)

Dependent Variable: Coverage Eligibility Offers

Intercept .478 .577 .637

(.023) (.025) (.024)

Expenditures -.044 -.084 -.022

(.021) (.023) (.022)

R-squared .029 .078 .005

Dependent Variable: Coverage Eligibility Offers

Intercept .477 .537 .628

(.020) (.021) (.021)

Expenditures -.015 .007 .022

(.020) (.022) (.022)

Dummy for 1993 -.076 -.100 -.079

(.009) (.010) (.010)

Dummy for 1995 -.014 -.071 -.026

(.010) (.011) (.011)

R-squared .387 .450 .338

Notes: Source of the expenditure data is Levit et al. (1997). Data is from the May 1988, April 1993, and February 1995 CPS Supplements. Standard errors in parentheses.