I thank Susan Black and Mike Clark for research assistance. Paul
Anglin, Michael Baye, William Custer, Daniel Hamermesh, and seminar
participants at the University of Kentucky and the W.E. Upjohn
Conference on Employee Benefits, Labor Costs, and Labor Markets
in Canada and the United States for useful comments. The W.E.
Upjohn Institute for Employment Research and the National Institutes
for Health provided research support.
This paper is to appear in W.E. Upjohn Institute volume entitled
Employee Benefits, Labor Cost, and Labor Markets in Canada
and the U.S
In this paper, I examine the impact of employer-provided health benefits on job turnover. Because many employer-provided plans extend coverage to a worker's entire family, the value of an employer's employment offer to a worker depends on whether the worker's spouse provides the family with health benefits. If a worker's spouse has an employer-provided health insurance for their family, the worker will value employment offers with and without health insurance benefits differently than a worker whose spouse does not have employer-provided health benefits. Importantly, this distortion arises from the reliance on employer-provided benefits and is independent of any pre-existing conditions clauses or issues concerning the portability of health plans. The estimates suggest that spouse-provided benefits substantially increase the likelihood of turnover.I. Introduction
One of the major differences between the U.S. and Canada labor markets lies in their treatment of health benefits. While Canada relies on government provision of health care, in the U.S. employers provide health insurance to most of the employed, and the government's role is primarily to provide health insurance to those over 64 years of age through the Medicare system and to the poor through the Medicaid system. Despite the recent calls for health care reform in the U.S., the reliance on employer-provided benefits appears to be a feature of the U.S. system for some time to come. The Clinton health care proposal of 1994 and the numerous Congressional alternatives rely on employer-provided health benefits.
In this paper, I examine the impact of employer-provided health benefits on job turnover. I focus on a peculiar aspect of employer-provided health benefits: because many employer-provided plans extend coverage to a worker's entire family, the value of an employer's employment offer to a worker depends on whether the worker's spouse provides the family with health benefits. If a worker's spouse has employer-provided health insurance for their family, the worker will value employment offers with and without health insurance benefits differently than a worker whose spouse does not have employer-provided health benefits. Importantly, this distortion arises from the reliance on employer-provided benefits and is independent of any pre-existing conditions clauses or issues concerning the portability of health plans. As I show in Section IV, this is potentially a large distortion. According to the April 1993 Supplement of CPS, among full-time workers at least 23% of the women and 12% of the men have coverage from their spouse.
II. Review of the Literature
There is no obvious reason why employers should provide health benefits. While health insurance is less expensive in groups, there is no particular reason the groups should be based on place of employment. Indeed, the initial growth in employer-provided health was the result of firms offering health insurance to their workers during World War II to avoid wage controls. As Long and Scott (1982) and Woodbury (1983) emphasize, the U.S. tax codes provide the major impetus for the employer provision of health and other benefits. The magnitudes of the tax savings are surprising. Consider a university in the Commonwealth of Kentucky that offers an insurance policy whose market value is $131 a month to a college professor who is in the 28% marginal tax rate for the federal income tax (family income is between $36,900 and $89,150 for married couples filing jointly). How much would it cost to increase the professor's after tax income by $131 in 1993? Assuming the professor's wages are not over the social security cap of $57,600 and taking into account Kentucky's 6% state income tax and the deductibility of state income taxes from the federal tax bill, the university would have to pay over $250.
As a result of the substantial tax savings associated with the exemption of health benefits from federal and state taxation, employers have become the major providers of health benefits in the U.S. The tax expenditures for the tax deductibility of employer health care premiums now exceed tax expenditures on the home mortgage deduction.
Economists have long recognized that the association of fringe benefits and the employment relationship may affect that relationship. Lazear (1979, 1981) argues that firms use defined benefit pensions to defer compensation in jobs with agency problems or in jobs with large investments in specific human capital. In jobs with agency problems, the deferred compensation deters the worker from shirking, while in jobs with specific human capital, the deferred compensation reduces job turnover. Thus, employers in the U.S. may use their pension plans to improve the efficiency of labor contracts, an option that many Canadian employers do not have. Lazear and the literature that his papers generate (e.g., Ippolito, 1985, Hutchens, 1987, and Dorsey, 1987) recognize that deferred compensation is not without its costs and may have to be implemented with other policies such as restrictions on hours and mandatory retirement to mitigate those costs.
Firms are not, however, perfectly able to tailor the parameters of their pension plans to meet contracting needs of an individual employee. As Scott, Berger, and Black (1989) and Scott, Berger, and Garen (forthcoming) emphasize, the Internal Revenue Service (IRS) of the U.S. government requires firms to offer fringe benefits in a manner that does not discriminate against the firm's low-wage employees. If the firm wishes to offer an executive a defined benefit pension plan that defers compensation, the firm must offer her secretary a similar plan. Thus, firms are not able to structure fringe benefit packages to match perfectly the optimal contract for each employee.
The requirement that fringe benefits be offered in a nondiscriminatory manner has a special bite in the provision of health benefits. While firms may tie pension benefits to the earnings of the worker, the firm must offer all full-time workers the same health benefits, which has the predictable consequence that high-wage firms will avoid hiring low-wage workers (Scott, Berger, and Black, 1989). Madrian (1994) identifies another possible distortion that employer-provided health benefits create: the possibility that workers will be locked into their jobs because they or family members have pre-existing conditions and would lose their medical coverage if they changed employers. Using the 1987 National Medical Expenditure Survey, she estimates that job lock reduces voluntary job turnover by 25% compared to a system of perfectly portable health insurance. Madrian's results are controversial. Holtz-Eakin (1994) found no evidence of job lock. If her results are correct, however, Madrian has identified a potentially important distortion in the U.S. labor market that employer-provided health insurance creates. Obviously, in Canada with perfectly portable health insurance, labor markets are free from such distortions.
Madrian argues that job lock arises from coverage gaps that pre-existing conditions clauses and length-of-service provisions create. If a worker must wait, say, for six months before being covered by a new employer's plan, then the worker may choose not to switch employers. Because this coverage gap is unrelated to the efficient allocation of labor, such a reduction in mobility is inefficient. She suggests that eliminating pre-existing condition clauses and increasing the portability of health insurance would largely eliminate the inefficient reduction in job turnover. In the next section, I offer a theoretical model that challenges this suggestion. I show that when dual earning couples consider employment offers, the value that they place on a job offer will depend on whether their spouse's health plan also covers them. As I demonstrate in the next section, this difference in valuation may explain the turnover pattern that Madrian uncovered.
III. Job Search with the Potential for Double Coverage of Health Benefits
In this section, I construct a simple model to examine the impact of the double coverage of health benefits on labor turnover. To abstract from other issues, I will assume that there are no pre-existing conditions provisions and no length-of-service provisions. If a worker finds employment at an alternative employer who is offering health insurance, the coverage begins immediately.
To begin, first consider a worker who has no spouse. The worker
is currently employed at a firm paying wage w0 and
a health plan indexed by the value h0. I assume that
all health plans may be indexed by a single value, h, and that
workers always strictly prefer plans with a greater h. Workers
without health coverage have a plan with the value of h0
= 0. Let the worker have a utility function u() that depends
on the level of wages, w0, and the level of health
benefits, h0, or
(1) .
The value of current employment, V0, forms the reservation utility for all subsequent employment offers. The worker has worked for the current employer for one period and will work at most one additional period for the employer. In Figure 1, I depict an indifference curve for the worker's utility function, which I depict as a convex function. If firms could individually tailor their fringe benefit package to the needs of a worker, the worker would simply pick the amount of health benefits he desires. If the worker had adequate coverage from another source, he could simply elect to take all compensation as wages. Unfortunately, IRS regulations preclude such a design.
Before beginning employment in the second period, the worker
entertains employment offers from other employers, which I assume
are exogenously determined. The worker's utility in the second
period is
(2) ,
where u(wa,ha) is the utility associated with the best alternative offer. In equation (2), the set of acceptable offers is simply all combinations of (w,h) that are above the indifference curve V0 depicted in Figure 1. The probability that a worker leaves his current employer, therefore, depends on the joint distribution of wages and health benefits offered.
Now consider a worker with a spouse. Let hs denote
the value of the worker's coverage under his spouse's health plan.
If the worker has no such coverage, then hs=0. The
worker's utility from employment in the first period is
(3) .
Again, before beginning employment in the second period, the worker
entertains offers from alternative employers. The utility from
second period employment is
(4)
The value of the right-hand side of equation (3) and (4) depends on the value of hs. Spouse-provided health care benefits, therefore, alter the value of current employment and thus alter the value of alternative offers.
Figure 2 illustrates how the coverage by a spouse's plan affects
the worker's job mobility decision. In Panel A, I consider the
case where hs< h0, or the worker's own
plan is more generous than his spouse's plan. The indifference
curve V0 denotes a worker's indifference curve if hs=0,
with the point (w0,h0) denoting the worker's
current contract. From equation (3), clearly spouse-provided
coverage (h0>hs>0) does not alter
the value of current employment, but it may affect the value of
alternative offers. To see why, consider the point (ws,hs),
where ws is implicitly defined as
(5) .
The wage ws leaves the worker indifferent to his current position and the job offering ws and consuming his spouse's health insurance. Any job that pays a wage greater than ws will be strictly preferred to his current position. Thus, the area under the indifference curve V0 and above the wage ws, which I denote as A in Panel A, becomes a part of the set of acceptable offers. Spouse-provided coverage, therefore, unambiguously increases the likelihood of turnover whenever hs<h0. Unlike the analysis of Madrian, this result does not depend on the lack of portability of benefits but is the direct result of the increase in the acceptable offer set that double coverage provides.
In Panel B, I consider the case in which hs>h0, where the spouse's benefits are more generous than the worker's own. Again, the indifference curve V0 corresponds to the worker without coverage by his spouse's benefits, or hs = 0. When a worker's spouse provides access to more generous benefits, the worker's utility increases. The indifference curve V0' depicts the worker's indifference curve when hs>ho. In comparing the values of current employment of workers with and without spouse-provided coverage, there are two regions of interest. First, the area under the indifference curve V0' and above V0, which I denote as region B in Panel B, represents offers that would be acceptable to workers without spouse-provided coverage but that are not acceptable to workers with spouse-provided coverage. Thus, one effect of spouse-provided coverage, when hs>h0, is to reduce this portion of the acceptable offer set. The second region of interest, however, offsets this result. The region that lies above w0 and below the indifference curve V0, which I denote as region C in Panel B, represents an area of offers that are acceptable to the workers with spouse-provided coverage but are unacceptable to workers without spouse-provided coverage. As the worker does not use his own health benefits, any job that offers a wage greater than w0 is strictly preferred to his current situation regardless of the level of health benefits associated with the job. For worker's with hs>h0, therefore, spouse-provided coverage has an ambiguous impact on turnover probabilities.
My analysis has abstracted from the search decision of the worker's spouse. When allowing for joint search decisions, the worker's valuation of his current job and alternative offers depends not only on his spouse's current position but also her best alternative offer. While the impact of the spouse-provided coverage on a worker's turnover probabilities is ambiguous, the impact on efficiency is unambiguous. Having a worker's valuation of an employment offer depend on his spouse's health insurance plan, only limits the efficient allocation of labor.
Of course, my analysis has not considered the possible responses of firms. One obvious response to double coverage is to offer employees the ability to select other benefits or cash in the place of health care benefits. The Revenue Act of 1978 permitted establishment of such cafeteria plans. The economic rational for offering such plans is obvious: by allowing employees who already have other sources of coverage to select from other benefits or cash payments, firms may reduce their turnover.
Another way in which firms may counter the problem of dual coverage is to attempt to specialize in the hiring of workers of one type of coverage or another. For instance, a firm may seek to hire only workers with access to alternative forms of health care coverage by offering jobs with higher wages and no health benefits. Another firm may seek to specialize in the hiring of workers who wish to provide coverage to their entire family by offering low wages but a generous health plan with family coverage. See Dye and Antle (1984) for a model of such a separating equilibrium applied to fringe benefits.
In the next section, I provide an overview of employer-provided health benefits in the U.S. with data from the April 1993 Supplement to the CPS. I demonstrate that neither the use of cafeteria plans nor sorting strategies on the part of firms have solved the problem of double coverage. I show that a significant portion of the population has double coverage, that a surprising number of people turn down coverage, and that among those that turn down coverage, most do so without explicit compensation.
IV. Coverage, Double Coverage, and Refusal of Employer-Provided Health Benefits
In this section, I present an overview of employer-provided health benefits from the April 1993 Supplement to the CPS. The supplement provides detailed information about employee benefits. I limit my sample to workers between the ages of 18 and 64 for all the tables. In addition, I report most statistics for full-time workers, which I define to be those who usually work at least 35 hours a week and those who work at least 47 weeks a year.
In Table 1, I present means for health insurance coverage and employer-provided coverage for full-time workers broken out by gender. Nearly 90% of the male workers and 90% of the female workers have health insurance from some source. For female workers, 88.0% report that they are at a firm that offers health insurance to at least some workers at the firm, and 88.5% of males respond similarly. Firms can place some restrictions on who may qualify for insurance. Often times, temporary, part-time, or leased employees may not be eligible for health benefits. Also, many firms require length-of- service requirements that a worker must complete before they qualify for health benefits. To see who is and is not eligible for health benefits, I identify workers as eligible for health benefits if they report that their firm offers health insurance to some of its workers and either report that they receive those benefits or explicitly state that they declined those benefits. Using this definition, 83.7% of female workers and 85.0% of male workers report that they are eligible for benefits.
Looking at the coverage rate of employer-provided health plans, 79.5% of all men but only 72.5% of women report that they have employer-provided health benefits. Thus, gender differences in wage understate the true compensation difference. 18.2% of women and 10.1% of men do not receive health insurance from their employers, but do receive it from another source. The differentials between the eligibility rates and the coverage rates suggest that many workers refuse health insurance coverage, and, indeed, 11.2% of all women and 5.6% of all men decline coverage from their employers. Among full-time workers, 22.0% of all women and 10.7% of all men report that they have health insurance under their spouse's plan.
The CPS Supplement also gives us an opportunity to examine another issue: the health insurance coverage of the self-employed. Folklore suggests that the spouses of the self-employed provide the health coverage for the family. In Table 2, I examine this issue by comparing the rate at which the spouses of the self-employed provide health insurance to their spouses compared to the rate at which the spouses of wage and salary workers provided health insurance to their spouses. In Panel A, we see no evidence for this folklore. The husbands' provision of health insurance to their wives is independent of their wives' self-employment status, which is surprising. In contrast, from Panel B we see that wives are more likely to provide self-employed husbands with health insurance than are wives of wage and salary workers. Women with self-employed spouses are 2/3 more likely to provide their husbands with health insurance than are women whose spouses are not self-employed.
The model I presented in the previous section suggests that employees whose spouses also have employer-provided coverage may value job offers differently than employees whose spouses do not have such coverage. For dual coverage to have an important effect on labor market transitions, however, there must be a sizable portion of the working population that may have double coverage. To determine what fraction of dual earning couples have dual health coverage, I matched husbands' and wives' responses to the April Supplement for those households in which both members are full-time, full-year workers. In Table 3, I present evidence about the possibility of double coverage. For males, 80.3% of the men from dual earning households are eligible for health insurance from their employers and their spouses are also eligible for family benefits. Thus, over 80% of these males could be covered by their wife's plan, and 38.5% of these men have wives who elect to provide family benefits. Similar stories arise for men whose employers offer family coverage. 80.6% of men who are eligible for family coverage have wives whose employers offer family plans. Interestingly, 38.0% of men from dual earning households who are eligible for family health plans have wives who provide family health plans, which represents a sizable segment of the married, dual earning families. Workers with spouses that have their own employer-provided health benefits may value family health benefits differently than workers whose spouses do not have employer-provided health benefits. 84.9% of these male workers have spouses who are eligible for employer-provided health benefits, and 62.4% have spouses who receive employer-provided health benefits.
In Panel B of Table 3, I report similar statistics for full-time female employees. 84.6% of women in dual earning households who are offered health insurance have spouses who are eligible for family plans, and 58.6% have spouses who provide family health benefits. Thus, women are more likely to have access to health benefits from multiple sources than are men. Of women who are eligible to provide family health benefits, 84.9% of their spouses are eligible for family health benefits and 58.0% provide such benefits. Finally, of women in dual earning households who are eligible for family health benefits, 87.7% are married to men who are eligible for health benefits and 76.5 are married to men who have employer-provided benefits.
When employers only partially pay for health benefits, employees have an incentive not to accept health benefits when they receive coverage from their spouses' plans. The refusal of health benefits is not uncommon; from Table 1, 11.2% of all female workers and 5.6% of all male workers decline employer-provided health benefits. In Table 4, I examine the incidence of a worker from a dual earning household refusing employer-provided health benefits by whether or not the worker's spouse is eligible for family health benefits. From Panel A, 3.1 % of male workers whose spouses are not eligible for family health benefits refuse coverage, but 12.9% of workers whose spouses are eligible for family health benefits refuse coverage. Thus, among men, workers who have spouses who are eligible for family health coverage are over 4 times more likely to refuse employer-provided health benefits than are men whose wives are not eligible for family health benefits. From Panel B, the impact for females is even more dramatic. Only 4.1% of women whose spouses are not eligible for family health benefits refuse employer-provided benefits, but 26.7% of women whose spouses are eligible for family health benefits refuse employer-provided benefits. Thus, women whose husbands have access to family health benefits are 6 times more likely to refuse health benefits than women whose husbands do not have access to family health benefits.
When husband and wife search for employment and employers offer health insurance coverage for the whole family, my theory predicts that the husband's and wife's health care coverage decision should be negatively correlated. Thus, controlling for other factors that affect the demand for health insurance coverage, we should see the likelihood of a worker choosing employer-provided health insurance declining when his spouse has selected employer-provided health insurance. To test this hypothesis, I estimate a bivariate probit model that allows for correlation between the husband's and wife's decisions. I limit the sample to couples where both are full-time, full-year workers. For covariates, I use a vector of race dummies (whites are the excluded category), a vector of education variables (high school graduates are the excluded category), the number of children in the household less than 18 years old, a quadratic in the worker's age, a quadratic in the worker's tenure at the firm, and a dummy variable indicating that the worker's tenure is less than a year. The method of estimation is full information, maximum likelihood. The starting values were taken from probits on the individual equations, and the starting value for the correlation coefficient, , is zero.
In Table 5, I present the result of the estimations. The estimated coefficients on the covariates provide few surprises. Workers of both genders have strong tenure affects. It seems unlikely that length-of-service requirements would account for the strong tenure-health benefits relationship so the strong relationship may simply reflect the fact that matches that offer health benefits tend to survive while those that do not offer health insurance do not survive, a point that Mortensen (1989) and Garen (1988) make in examining the wage-tenure relationship. Workers with at least a BA degree are more likely to have health insurance than less educated workers. For women, a larger number of children reduces the likelihood of having employer-provided health insurance, while for men the relationship is not statistically significant. Interestingly, hispanic wives are more likely but hispanic husbands are less likely to have employer-provided health insurance than similar whites. Similarly, black wives are more likely to have employer-provided health insurance than are white wives.
Controlling for the worker's own characteristics, there is a strong, negative correlation between husbands' and wives' health care decisions. The estimated correlation coefficient is -0.35 and the z-statistic is -9.03. Thus, the data overwhelmingly reject the hypothesis that the health care decisions of dual earning couples are independent and accept the hypothesis, which my theory implies, that the decisions are negatively correlated. Husbands and wives appear to coordinate their search activities, presumably looking for other forms of compensation when their spouse provides health benefits. Thus, within households, there is some evidence that workers do indeed trade off health benefits for other forms of compensation.
V. Does Spouse-Provided Health Insurance Affect Turnover Probabilities?
The analysis in Section III suggests that coverage under a spouse's health insurance plan alters the worker's likelihood of accepting an offer. If the spouse's plan is less generous than the worker's own health insurance plan, then coverage by the spouse unambiguously increases the likelihood that a worker will accept another offer. In equilibrium, therefore, we should see such workers more likely to change jobs than workers without spouse-provided coverage. When the spouse's plan is more generous than the worker's own plan, there is an ambiguity, but it remains possible that spouse-provided coverage would result in higher turnover rates.
Unfortunately, the CPS is a less than ideal data set to use to examine job transitions. Because the CPS is a short panel and provides few details about a worker's employers, it is often impossible to spot job-to-job transitions. In the April 1993 Supplement, however, workers are asked directly if they have less than one year tenure, and answers to this question allow me to identify those individual's who have changed jobs in the last year. It is not possible, however to determine whether the transition was a result of a quit, a layoff, or a dismissal.
The CPS provides only workers' current health insurance and not their coverage at the time of their job transitions, which causes a potentially serious problem. If workers who have recently had an involuntary job transition (layoff or dismissal) are likely to enroll in their spouse's health care plan, then there is a correlation between current health care coverage under a spouse's plan and turnover that is unrelated to any search story. In addition, the CPS provides no information about the generosity of workers' health care plans nor of their spouses' plans. As the generosity of the two plans affects the likelihood of turnover in my model, this data limitation is particularly serious. Finally, the CPS provides no information about tenure on the previous job. As virtually all research has found that hazard functions for employment spells exhibit duration dependence (e.g., Farber, 1994), the failure to include tenure in a turnover equation may cause a specification bias.
With these caveats in mind, I can examine the relationship between
job transitions and health insurance coverage provided by a workers'
spouse with the equation:
where Xi is a vector of controls, is the corresponding vector of parameters, Si is an indicator variable that is equal to one if the worker is covered by his spouse's plan and zero otherwise, is the corresponding parameter, ui is the error term that I assume is identically and independently logistically distributed, and F() is a cumulative logistic distribution function.
Because males and females may have much different patterns of turnover, I run separate equations for male and female workers. In addition to controls for whether or not the spouse is employed or self-employed, I use the same control variables as those I use in Table 5, except of course I use no controls for tenure. In columns (1) of Table 6 and 7, I present the estimates for equation (6) for male and female workers. I limit my sample to workers who are married, full-time, full-year workers who have at least two years of potential experience, where potential experience is defined to be age minus years of schooling minus six. This restriction should exclude most school-to-work transitions, which presumably occur regardless of the spouse's provision of health benefits.
A common feature of the results from both samples is that having an employed spouse substantially reduces the likelihood workers change jobs. (This result remains regardless of whether I control for coverage by the spouse's health insurance plan.) Spouse-provided coverage has a large impact on the likelihood of turnover for male workers; evaluated at the mean, spouse-provided coverage increases the likelihood of a male worker changing jobs from about 0.10 to 0.16. For females, the impact is smaller, but is still large; evaluated at the mean, spouse-provided coverage increases the likelihood of a female worker changing jobs from about 0.10 to 0.14.
My estimates for males are somewhat higher than those of Madrian (1994), who found that not having other health insurance coverage lowered male job transitions by about 26%. Importantly, Madrian is able to control for whether the job transition was voluntary, and I am unable to do so. To guard against the possibility that spouse-provided coverage is somehow indicative of an involuntary transition from the last job, I re-estimate the equation limiting my sample to those workers who report that they are eligible for employer-provided health insurance. For this sample, workers who made job transitions at least have the option of taking their employer-provided plan. While clearly this does not preclude a worker from having been laid-off or dismissed from his past position, this does eliminate any workers who have spouse-provided benefits because they have no alternative source of health care. With this sample restriction, the coefficients on the spouse-provided coverage are reasonably stable. Evaluated at the means, spouse-provided coverage increases the likelihood of a male worker changing jobs from 0.07 to 0.11 and the likelihood of a female worker changing jobs from 0.07 to 0.12.
Thus, the CPS data seem to support the conclusion that spouse-provided coverage does encourage job transitions, and the results are largely consistent with those of Madrian (1994) for workers with dual coverage. Her interpretation, however, is that workers without dual coverage are possibly "locked-out" of jobs that offer insurance with pre-existing-conditions clauses or length-of-service requirements. Health care reform that eliminates pre-existing-conditions clauses and length-of-service requirements and requires employers to offer health insurance would virtually eliminate job-lock. Unless the employer mandate also eliminates variations in the type of employer-provided coverage, my analysis suggests that the turnover that spouse-provided coverage creates is likely to persist. Ideally, therefore, we would like to be able to distinguish my search explanation from her job-lock explanation and be able to decompose the turnover effect into a search component and a job-lock component.
This is likely to prove a difficult task. Gruber and Madrian (1994) and Holtz-Eakin (1994) contend that most job-lock appears to be a short run problem, presumably arising more from the length-of-service requirements than from pre-existing conditions. Individuals without a pre-existing condition, however, have the option of purchasing insurance from the private market, or, as Gruber and Madrian emphasize, some workers may purchase health care from their previous employers to bridge the gap in coverage that length-of-service provisions create. This solution to a coverage gap is expensive: the worker loses the tax exemption of health care insurance premiums, and, if purchasing health insurance from the private market, non-group policies are often more expensive. Yet for these workers, a solution does exist, and a sufficiently generous offer will induce the worker to change jobs. Because this solution is expensive and because workers with spouse-provided coverage avoid these costs, workers differ in their valuation of offers from alternative employers, which of course is the essence of my search explanation for the turnover effect from spouse-provided coverage. In my view, distinguishing between these two explanations would be difficult.
VI. Policy Implications
My results support the findings of Madrian (1994) and Gruber and Madrian (1994) that employer-provided health insurance does affect the turnover propensities of workers. Indeed, the magnitude of my results for male workers is somewhat larger than Madrian's estimate, and I find that female workers are similarly affected. While I have offered no formal welfare analysis of this effect, it is difficult to believe that a policy that makes a worker's turnover propensity dependent on the health care policy of his spouse would improve the efficiency of labor markets.
Why have employer-provided health insurance? Friedman (1993) argues that many firms initially offered health care as a fringe benefit; as a means of avoiding the wage-price controls of World War II. As the IRS did not initially count fringe benefits as a part of taxable income, the tax system encouraged firms to offer health care and Congress eventually codified the tax exemption. As health benefits are income elastic (Woodbury and Huang, 1991), the tax exemption favors those with high earnings. Therefore, equity concerns suggest that a change is in order as well. When efficiency and equity concerns agree, one hopes that economists would find the course of action uncontroversial.
The political appeal of continuing the employer-provision of health benefits or the expansion of the system through mandates seems to arise because the costs remain hidden from consumers. Gruber (1994) and Gruber and Krueger (1992) suggest that most if not all costs of mandated benefits are passed through to the workers as lower wages, but if the mandated program is sufficiently small, these wage-pass-throughs may be difficult for workers to perceive. Moreover, the tax expenditure that arises from the exemption of employer-provided health insurance is not readily apparent. Those of us who are beneficiaries of the tax expenditure probably do not appreciate the largesse of the US government, at least not until the exemption is threatened.
Unfortunately, any elimination of the tax subsidy of health insurance
benefits would not be invisible. Consider a reform along the
lines that Diamond (1992) suggests, but one without any tax subsidy
for middle-class families. In such a plan, employer-provided
health insurance is replaced with a system of mandatory coverage
where, at least for most middle class households, consumers pay
the full cost of their health insurance. Those workers who previously
had employer-provided health insurance should receive a nice increase
in compensation. Under Diamond's proposal, regional "HealthFeds"
negotiate several different policies with insurance companies,
and consumers within the region choose among the approved policies.
When consumers begin looking at the prices of the various policies,
however, they will notice that, even if firms increased their
compensation by the exact cost of the previously provided health
insurance, the increase in their compensation is not enough to
allow them to purchase an insurance plan of comparable quality
to their employer-provided plan. Because the tax subsidy is eliminated,
the income and substitution effects presumably would move most
consumers to purchase less generous insurance plans. Woodbury
and Huang's simulation results suggest that the full taxation
of health benefits may result in up to a 15% decline in the amount
of health insurance. They calculated these estimates for the
1986 US tax codes, and marginal tax rates have increased since
then. Forcing consumers to understand fully the costs of health
care may not be good politics, but, in my view, it is good economics.
Allen, Steven G., Robert L. Clark and Ann A. McDermed. "Pensions,
Bonding, and Lifetime Employment" Journal of Human Resources
(28) Summer 1993 463-81.
Daniel, Kermit. "Does Marriage Make Men More Productive?"
mimeo., University of Pennsylvania, 1994.
Diamond, Peter. "Organizing the Health Insurance Market"
Econometrica (50) November 1992 1233-55.
Dorsey, Stuart. "The Economic Function of Private Pensions:
An Empirical Analysis" Journal of Labor Economics
(5) s171-89.
Dye, Ronald A. and Rick Antle. "Self-Selection via Fringe
Benefits" Journal of Labor Economics (2) July 1984
388-411.
Farber, Henry. "The Analysis of Interfirm Mobility"
Journal of Labor Economics (12) October 1994 554-93..
Friedman, Milton. "The Folly of Buying Health Care at the
Company Store" The Wall Street Journal, February
3, 1993, p. A20.
Garen, John E. "Empirical Studies of the Job Matching Hypothesis"
Research in Labor Economics (9) 1988 187-224.
Gruber, Jonathan. "The Incidence of Mandated Maternity
Benefits" American Economic Review (84) June 1994
622-41.
Gruber, Jonathan and Alan B. Krueger. "The Incidence of
Mandated Employer-Provided Insurance: Lessons from Workman's
Compensation Insurance" in David Bradford, ed., Tax Policy
and the Economy Cambridge, MA: M.I.T. Press, 1992.
Gruber, Jonathan and Brigitte Madrian. "Limited Insurance
Portability and Job Mobility: The Effects of Public Policy on
Job-Lock" Industrial and Labor Relations Review (48)
October 1994 86-102.
Holtz-Eakin, Douglas. "Health Insurance Provision and Labor
Market Efficiency in the United States and Germany" in Rebecca
Blank's Protection Versus Economic Flexibility: Is There a
Tradeoff? Chicago: University of Chicago Press, 1994.
Hutchens, Robert. "Delayed Payment Contracts and the Firm's
Propensity to Hire Older Workers" Journal of Labor Economics
(4) October 1986 439-57.
________. "A Test of Lazear's Theory of Delayed Payment
Contracts" Journal of Labor Economics (5) October
1987 s153-70.
Ippolito Richard A. "The Labor Contract and True Economic
Pension Liabilities" American Economic Review (75)
December 1985 1031-43.
Long, James and Frank Scott "The Income Tax and Nonwage
Compensation" Review of Economics and Statistics (64)
May 1982 211-9.
Luzadis, Rebecca A. and Olivia S. Mitchell. "Explaining
Pension Dynamics" Journal of Human Resources (26)
Fall 1991 679-703.
Madrian, Brigitte. "Employment-Based Health Insurance
and Job Mobility: Is There Evidence of Job Lock?" Quarterly
Journal of Economics (109) February, 1994 27-54.
Monheit, Alan C. and Philip Cooper. "Health Insurance and
Job Mobility: Theory and Evidence" Industrial and Labor
Relations Review (48) October 1994 68-85.
Mortensen, Dale T. "Wages, Separation and Job Tenure:
On-the-Job Specific Training or Matching?" Journal of
Labor Economics (6) October 1988 445-72.
Olson, Craig. "Parttime Work, Health Insurance Coverage
and the Wages of Married Women" 1994.
Scott, Frank A., Mark C. Berger, and Dan A. Black. "Effects
of the Tax Treatment of Fringe Benefits on Labor Market Segmentation"
Industrial and Labor Relations Review (42) January 1989
216-229.
Scott, Frank, Mark Berger, and John Garen. "Do Health Insurance
Costs and Nondiscrimination Policies Reduce the Job Opportunities
of Older Workers?" Industrial and Labor Relations Review
forthcoming.
Woodbury, Stephen. "Substitution Between Wage and Nonwage
Benefits" American Economic Review (73) March, 1983
166-82.
Woodbury, Stephen and Wei-Jang Huang. The Tax Treatment of Fringe Benefits Kalamazoo, MI: W.E. Upjohn Institute for Employment Research, 1991.
| Covered by some form of health insurance | ||
| Employed at firm that offers health insurance | ||
| Eligible for employer-provided health insurance | ||
| Covered by employer-provided health insurance | ||
| Refused employer-provided health insurance | ||
| Covered by spouse's health insurance | ||
| Sample size (n) |
Panel A: Husband's provision of health insurance to spouse
by wife's self-employment status
Husband does not provide spouse with employer-provided insurance |
| ||
Husband provides spouse with employer-provided insurance | |||
| Total percentage | |||
| n |
Panel B: Wife's provision of health insurance to spouse by
husband's self-employment status
Wife does not provide spouse with employer-provided insurance | |||
Wife does provides spouse with employer-provided insurance | |||
| Total percentage | |||
| n |
Panel A: Husband's employer offers
| health benefits and spouse is eligible for family health benefits | ||
| health benefits and spouse provides family health benefits | ||
| family health benefits and spouse is eligible for family health benefits | ||
| family health benefits and spouse provides family health benefits | ||
| family health benefits and spouse is eligible for health benefits | ||
| family health benefits and spouse receives health benefits |
Panel B: Wife's employer offers
| health benefits and spouse is eligible for family health benefits | ||
| health benefits and spouse provides family health benefits | ||
| family health benefits and spouse is eligible for family health benefits | ||
| family health benefits and spouse provides family health benefits | ||
| family health benefits and spouse is eligible for health benefits | ||
| family health benefits and spouse receives health benefits |
Note: To be included in this sample, workers must be working full-time and eligible for employer-provided health benefits. Spouses may or may not be eligible for health benefits, but must be full-time worker.
Panel A: Husband's decision to accept or refuse employer-provided
health insurance
Husband accepts employer-provided insurance | |||
Husband refuses employer-provided insurance | |||
| Total percentage | |||
| n |
Panel B: Wife's decision to accept or refuse employer-provided
health insurance
| |||
Wife accepts employer-provided insurance |
| ||
Wife refuses employer-provided insurance | |||
| Total percentage | |||
| n |
Note: To be included in this sample, workers must
be working full-time and eligible for employer-provided health
benefits. Spouses may or may not be eligible for health benefits,
but must be full-time worker..
|
| |||
| Worker is Hispanic |
|
| ||
| Worker is black |
|
| ||
| Worker is Asian |
|
| ||
| Worker is Native American |
|
| ||
| Worker's age |
|
| ||
| Age squared /100 |
|
| ||
| Worker has less than one year of tenure |
|
| ||
| Worker's tenure |
|
| ||
| Tenure squared / 100 |
|
| ||
| Number of children |
|
| ||
| Worker did not begin high school |
|
| ||
| Worker did not complete high school |
|
| ||
| Worker attended college but has no degree |
|
| ||
| Worker has a vocational degree from junior college |
|
| ||
| Worker has an associate degree |
|
| ||
| Worker has a bachelors degree |
|
| ||
| Worker has a masters degree |
|
| ||
| Worker has a Ph.D. degree |
|
| ||
| Worker has a professional degree |
|
| ||
| Constant |
|
| ||
| ||||
| Likelihood function | ||||
| Number of observations |
a. Mean of the dependent variable for column (1) 0.6465 and for column (2) is 0.7727. Absolute values of z-statistics are given in parentheses.
| Worker is Hispanic |
|
| ||
| Worker is black |
|
| ||
| Worker is Asian |
|
| ||
| Worker is Native American |
|
| ||
| Worker's age |
|
| ||
| Age squared /100 |
|
| ||
| Worker did not begin high school |
|
| ||
| Worker did not complete high school |
|
| ||
| Worker attended college but has no degree |
|
| ||
| Worker has a vocational degree from junior college |
|
| ||
| Worker has an associate degree | (0.30) |
| ||
| Worker has a bachelors degree |
|
| ||
| Worker has a masters degree |
|
| ||
| Worker has a Ph.D. degree |
|
| ||
| Worker has a professional degree |
|
| ||
| Number of children |
|
| ||
| Spouse is employed |
|
| ||
| Spouse is self-employed |
|
| ||
| Worker is covered by spouse's plan |
|
| ||
| Worker is covered by other plan |
|
| ||
| Constant |
|
| ||
| Likelihood function | ||||
| Number of observations |
a. Mean of the dependent variable for column (1) 0.096 and for column (2) is 0.069. Absolute values of z-statistics are given in parentheses.
| Worker is Hispanic |
|
| ||
| Worker is black |
|
| ||
| Worker is Asian |
|
| ||
| Worker is Native American |
|
| ||
| Worker's age |
|
| ||
| Age squared /100 |
|
| ||
| Worker did not begin high school |
|
| ||
| Worker did not complete high school |
|
| ||
| Worker attended college but has no degree |
|
| ||
| Worker has a vocational degree from junior college |
|
| ||
| Worker has an associate degree |
|
| ||
| Worker has a bachelors degree |
|
| ||
| Worker has a masters degree |
|
| ||
| Worker has a Ph.D. degree |
|
| ||
| Worker has a professional degree |
|
| ||
| Number of children |
|
| ||
| Spouse is employed |
|
| ||
| Worker is covered by spouse's plan |
|
| ||
| Worker is covered by other plan |
|
| ||
| Constant |
|
| ||
| Likelihood function | ||||
| Number of observations |
a. The mean of the dependent variable for column
(1) is 0.097 and for column (2) is 0.074. Absolute values of
z-statistics are given in parentheses.