The effect of health insurance on married female labor supply.

I. Introduction

Health insurance in the United States largely is employment based: nearly 90 percent of Americans with private insurance are covered through their employer or the employer-provided plan of a family member. The national debate over health insurance reform that occurred in the early 1990s therefore generated a spate of new research on the labor market effects of health insurance provision. Because health insurance is the largest nonwage component of total compensation, accounting for 34 percent of expenditures on voluntary employee benefits and 7 percent of total compensation (U.S. BLS 1994a), these effects potentially are large. Recent research has examined the extent to which employer-provided health insurance reduces job mobility ("job-lock;" see Madrian 1994, Cooper and Monheit 1993, Holtz-Eakin 1994, Penrod 1993, Buchmueller and Valletta 1996) and affects retirement decisions (Karoly and Rogowski 1993, Gustman and Steinmeier 1994, Madrian 1994, Gruber and Madrian 1995). Other papers report that related public policies such as "continuation of coverage" laws (Gruber and Madrian 1994, 1996) and mandated maternity benefits (Gruber 1994) have significant labor market effects. Overall, recent research suggests large labor market effects of our current system of employment-based health insurance and correspondingly large effects of proposed federal reform.

A neglected potential effect of employer-provided health insurance is on hours worked and labor supply. One simple reason for such effects is that health insurance constitutes a lump-sum employment benefit that may induce a negative income effect on desired hours of labor supply. However, due to the unique nature of the good and quasi-fixed costs associated with employment-based health insurance, the effect of health insurance on desired labor supply may be more complex. For a number of reasons that we will discuss, employers seldom offer health insurance to part-time workers. Thus, some employees who would otherwise work part time may be working full time in order to acquire health coverage for themselves and their families. This behavior should be particularly pronounced for women, due to the tendency for men to work full time regardless of their wives' hours.

The joint distribution of employer-provided health insurance and hours worked, however, provides limited information on the labor supply effects of such employer-set hours constraints. In a standard labor supply regression, the effect of an hours threshold associated with the provision of health insurance is indistinguishable from pure preference heterogeneity. In the extreme, all workers with preferences for long hours will have health insurance on the job, while all those with preferences for short hours will lack health insurance on the job. The resulting joint distribution of hours and health insurance is not due to the threshold in a meaningful way, and in this pure heterogeneity case changing or eliminating the threshold will not change hours worked. Thus, we can not exploit variation in own insurance coverage as a means of identifying the effect of health insurance availability on desired hours of labor supply.

Instead, we use a family labor supply framework and incorporate data on spouse health insurance. Women who are not covered by health insurance through their husbands may work longer hours to acquire health insurance. This will be reflected in lower hours worked for women who are covered by husbands' insurance. We investigate the extent of this behavior using data on married couples from the April 1993 Current Population Survey (CPS) Benefits Supplement (U.S. Department of Commerce, Bureau of the Census 1994), a large, nationally representative labor market survey with extensive information on health insurance coverage and offers, labor supply, and related demographic and economic variables. Two other papers have investigated similar hypotheses (Olson 1995; Wellington and Cobb-Clark 1996). As described in detail in Section II, we offer several departures from this work, particularly along methodological lines. Our econometric models include reduced form Tobit and probit labor supply equations and a multinomial logit model that provides a stronger test of our hypothesis. These models generally indicate a large negative effect of husbands' insurance on married female labor supply, particularly in families with children. The multinomial logit results imply that this effect arises from women with covered husbands shifting from full-time jobs with insurance to non-employment or part-time jobs without coverage.

The results of our analysis have direct policy implications. Various federal health insurance reform proposals have contained provisions that would reduce or eliminate the hours threshold associated with the provision of health insurance by employers.(1) Such changes will reduce labor supply to the extent that workers have increased their hours of work to acquire health coverage. Reduced labor supply may in turn have implications for alternative means of financing health reform.

II. Health Insurance and Family Labor Supply

Although the link between health benefits and labor supply has not been extensively analyzed, several recently published papers examine the labor supply effects of Medicaid benefits (Winkler 1991, Moffitt and Wolfe 1992, Wolfe and Hill 1995, Yelowitz 1995). These studies exploit variation in the value of Medicaid benefits across states and individuals to identify labor supply effects.

Similar analysis of employer-provided health insurance effects on labor supply is difficult due to limited observable variation in such benefits across individuals and geographic areas. Furthermore, such analysis is made complex by employers' tendency to restrict health insurance benefits to full-time workers. According to tabulations by the U.S. Bureau of Labor Statistics (BLS), 28 percent of part-time workers in firms with 100 or more employees receive health insurance benefits, compared to 83 percent of full-time workers in such firms (U.S. BLS 1993). In firms with fewer than 100 employees the gap is even greater: 5 percent of part-time workers and 71 percent of full-time workers receive health benefits (U.S. BLS 1994b).

These coverage rate differentials can not be explained by differences in the characteristics of part-time and full-time workers (Blank 1990). They relate instead to the unique characteristics of health insurance benefits. First, health insurance benefits represent "quasi-fixed" costs that vary with the number of workers rather than the number of hours (Oi 1962). This makes health insurance more expensive to provide to part-time workers (Montgomery and Cosgrove 1993). A negative relationship between individual health status and labor supply provides a second possible rationale for limiting health benefits to full-time workers: doing so screens out some high risk employees (Buchmueller 1995). Indeed, in the small group market insurers often require that insurance be limited to full-time employees and that a minimum fraction of eligible employees take coverage, in order to limit adverse selection.

A third potential explanation for the gap in benefits between full-time and part-time workers is that it reflects employer efforts to tailor compensation packages to average worker preferences. Married women, teenagers, and senior citizens account for a disproportionate share of part-time workers (Nardone 1995). Because these workers are often covered by the health insurance of a spouse or parent or Medicare, they will prefer a compensation package that is weighted more heavily toward wages. A related explanation is that IRS nondiscrimination rules limit the ability of firms to offer different benefit packages to different groups of full-time employees but allow the differential treatment of full-time and part-time employees. Thus, excluding part-time workers from fringe benefit plans is one way that employers can provide benefits to high-wage employees but not to low-wage employees without violating IRS non-discrimination rules (Scott, Berger, and Black 1989).

The resulting employer-set hours threshold for insurance coverage creates a non-convex budget constraint. Figure 1 depicts the labor supply effects of such employer-set hours thresholds. We assume initially that an employee faces a standard consumption (C) and leisure (L) tradeoff, subject to an hourly wage rate and total hours available (T). The resulting budget constraint is represented by the solid line segments in Figure 1. The right vertical portion of the budget constraint is generated by nonwage income. We assume that the worker can earn a real wage that is constant across hours worked.(2) In the absence of health insurance provision, the worker maximizes her utility at point 10 and works T - [1.sub.0] hours. Availability of health insurance on jobs that require a minimum of T - [1.sub.ht] hours worked generates the second vertical portion at [1.sub.ht]; the height of this second vertical portion represents the consumption value of health insurance coverage.(3) Given the availability of health insurance on her job, the worker shown will maximize utility at [l.sub.ht], assuming no alternative source of coverage. We call the horizontal distance between [1.sub.0] and [1.sub.ht] the "threshold effect" because it is generated by the employer-set hours threshold associated with the provision of health insurance.

Now consider the effect of availability of an alternative source of health insurance, such as coverage under a …