(2003) Amy Wolaver, Timothy McBride, Barbara Wolfe - Mandating Insurance Offers For Low-Wage Workers An Evaluation of Labor Market Effects

Mandating Insurance Offers for Low-Wage Workers: An Evaluation of Labor Market Effects
Amy Wolaver Bucknell University Timothy McBride University of MissouriSt. Louis Barbara Wolfe University of WisconsinMadison
Abstract
Employing a simultaneous model of part-time status, health insurance offers, and wages, we examine the impacts on employment and health insurance coverage of nondiscrimination rules in the tax code governing employer-sponsored health insurance. Using 1988 and 1993 Employee Benets Supplements to the Current Population Surveys and variations in health insurance premiums and minimum wages, we nd that health insurance coverage among low-wage primary earners is increased by at most 31 percent by the policy, at a cost of an estimated 0.8 5.4-percentage-point decrease in full-time employment for low-wage workers.
Since the failure of the Clinton health care reform initiative, comprehensive health insurance coverage in the United States has fallen off the political map, to be replaced by piecemeal measures to extend or improve the generosity of coverage for various subgroups of the population. Despite these efforts and a booming economy in the 1990s, a signicant proportion of the population, 14 percent in 2000 (Mills 2001), remained without health insurance coverage for the entire year. Lack of insurance is a problem because the uninsured disproportionately go without needed medical care or delay care. In 2002, 28 percent of uninsured adults did not receive needed medical care, compared to only 6 percent of insured persons. The uninsured are also more likely than the insured to postpone care (41 percent compared to 20 percent) and more likely to skip lling a prescription (26 percent compared to 11 percent) (Kaiser Family Foundation 2003).
Journal of Health Politics, Policy and Law, Vol. 28, No. 5, October 2003. Copyright 2003 by Duke University Press.
884
Journal of Health Politics, Policy and Law
With the exception of the boom economic years, 1999 and 2000, the number of uninsured in the United States had been steadily growing and, in particular, low-skilled workers lost ground relative to high-skilled workers. Between 1979 and 1993, employer-provided health insurance among workers with the lowest wages (the bottom fth) fell thirteen percentage points, whereas among those in the highest wage group (the top fth) coverage fell only three percentage points. This trend continued into the mid-1990s: between 1987 and 1996, employer-provided coverage of low-wage workers (those earning less than $7 per hour) fell from 54 to 42 percent, whereas coverage of high-wage workers (those earning more than $15 per hour) increased from 87 to 90 percent (OBrien and Feder 1999). Early anecdotal reports of low-income families improperly losing Medicaid coverage in the wake of the welfare reforms and joining the ranks of the uninsured (Pear 1999) have since been conrmed with data from ve states (Ellwood 1999), making it clear that the problem has grown worse. The majority of the uninsured are in families of the working poor. Among the uninsured in 1999, 71 percent were in families with at least one full-time worker, and 65 percent were in families with incomes less than 200 percent of the federal poverty level (Kaiser Family Foundation 2001). Lack of coverage among children is mostly found in families with working parents. In 2000, 7.3 million of the 9.2 million parents of uninsured children worked, 5.4 million of them full time (Hoffman and Pohl 2002). In general, these are families that earn too much to qualify for Medicaid but do not have access to employer-sponsored health insurance or cannot afford the premiums. In this article, we argue that the nondiscrimination rules governing the treatment of employer-sponsored health insurance have had the unintended side effect of harming low-wage workers, leading to underemployment and little, if any, increase in the probability of coverage. Of particular concern is the nondiscrimination rule included in the tax code in 1978 to extend self-insured health insurance coverage to low-wage workers. Since passage of the nondiscrimination laws in 1978, employers with self-insured plans have been obliged to offer fringe benets to essentially all full-time workers (not just highly compensated employees) in order to receive the full benets of the favorable tax treatment. Thus pensions, health insurance, and a variety of other benets if offered must be offered across the board to all full-time employees. The intended impact of this policy was to increase the access of low-wage workers to these benets, including employer-provided health insurance, with its tax and riskpooling advantages over the individual health insurance market. We argue
Wolaver et al.
Mandating Insurance Offers for Low-Wage Workers
885
here that this policy, meant to benet low-wage workers, may have decreased full-time employment for the group that the law was intended to help. The details of the rules and a copy of the text are found in Appendix A. In the next section we describe the provisions of the tax code governing employer insurance and use economic theory to generate hypotheses about how this policy might lead to negative employment effects for lowwage workers. Then we use data from the Current Population Survey (CPS) and other sources to test these hypotheses, using a model of parttime employment status, health insurance offers, and wages. We comment upon the implications of these ndings for tax policy and on initiatives to encourage increased health insurance coverage.
How Can Current Tax Policy Disproportionately Harm Low-Wage Workers?
Our current health insurance system is heavily weighted toward the employer provision of health insurance for two reasons. First, the group health insurance market has several advantages over the individual market, including economies of scale in administrative costs and the sharing of risk.1 The second reason is the tax exemption for fringe benets that has been in place since World War II. Workers pay for the insurance in pretax dollars, increasing their demand for compensation in this form rather than in the form of wages. Nondiscrimination rules amended the tax code in Section 105(h) of Title 26 in 1978. If the nondiscrimination rules are violated at rms that selfinsure, the differential value of the employers premium contribution to employees that are highly compensated relative to nonhighly compensated employees is added to taxable income. In Section 89 of the 1986 Tax Reform Act (Pub. Law 99514) an abortive attempt was made to strengthen the nondiscrimination rules, applying them to all plans and setting new, complex standards for compliance. The new rules were intended to go into effect on 1 January 1989, but before the end of 1989 Congress retroactively repealed the Section 89 rules in the face of protests from small businesses and others, so that they were never enforced (Pub. Law 101140). Congress nevertheless favored retaining some nondiscrimination rules
1. Economies of scale in insurance purchase exist because of xed administrative costs that must be covered regardless of group size. Large groups also avoid an adverse selection premium.
886
because, rst, tax expenditures for employer-provided health insurance should not disproportionately favor highly compensated workers and, second, because the rules were felt to promote(s) the broad health care coverage of rank-and-le employees (Rostenkowski 1989). Expanding health insurance coverage is a laudable goal, but another effect is possible. The nondiscrimination clause indirectly raises the effective minimum wage for rms that offer a self-insured health insurance policy. Economic theory regarding mandated benets predicts that, to the extent that workers value health insurance, they will be willing to accept lower wages in exchange for the benet. If health insurance does not affect productivity, rms should care only about the total compensation paid to workers, not the mix of wages and health insurance.2 If wages can be lowered enough to cover the full cost of the benet, then rms demand for workers will not change as a result (Summers 1989; Gruber and Krueger 1991; Gruber 1999).3 However, as Figure 1 shows, minimum wage laws interfere with this adjustment by preventing wages from falling below the legal oor. The left graph shows a hypothetical labor market for highly skilled workers. In response to the increased costs of labor from providing health insurance, rms decrease their demand, shifting the demand curve from D0 to DHI by the amount of the per-worker cost of insurance. A change in labor supply is also likely, making the total effect on employment ambiguous. If workers value receiving the health insurance, they will increase their supply of labor; in the case shown, from S0 to SHI, workers value the insurance dollar for dollar and employment levels do not change. Wages adjust downward, from W*0 to W*HI, exactly covering the cost of the insurance to the employer. The second panel shows the same effects for a group of lowskilled workers, but in this case, the minimum wage law prevents wages from adjusting fully, and the quantity of full-time, low-skilled workers supplied, ESHI, exceeds the quantity demanded, EDHI. Clearly, as minimum wages are raised or as health insurance premiums rise (causing the demand curve to shift even further), this constraint becomes more binding and larger numbers of full-time workers face the risk that employers will reduce
2. The form of compensation will affect payroll taxes, about which, at rst blush, rms may be concerned. However, Brittain 1971 shows that payroll taxes are passed onto the worker in the form of reduced wages, therefore negating their effect on employment. They may also result in reduced employment for low-wage workers, but since they are a variable cost and not a xed cost per worker, we do not expect to see employment effects for payroll taxes as large as those for health insurance. 3. It could also be argued that health insurance could increase rms demand for labor, if healthier workers are more productive workers, decreasing the likelihood of adverse employment effects.
Wolaver et al.
887
Wages
S0
SHI
S0 Wages SHI
W0 W0 WHI WHI Minimum wage
E*0 = E*HI High-skilled, Full-time Workers
EDHI <ESHI Low-skilled, Full-time Workers
Figure 1 Labor Market Changes with Employer Provision of Health Insurance
demand for their services. Health insurance is a benet even more likely to cause employment effects than pensions, because health insurance is a xed cost per worker whereas pension contributions are likely to be a variable cost tied to wages and earnings. As a result of the xed cost, employerprovided health insurance (EHI) is a higher fraction of total compensation for low-wage workers than for high-wage workers, meaning that the health insurance provision is more likely to hit the minimum wage constraint than the pension provision. Even if these workers would be willing to take jobs with wages that were below the legal minimum but that also offered health insurance, federal and state laws would prohibit this arrangement. The policy may therefore raise the labor costs of certain workers above the market value of their labor. The nondiscrimination clause acts, then, as a partial employer mandate. Employers are not compelled to offer any health insurance to any workers, but offers cannot be extended selectively to only the highly skilled workers. If rms employ some workers whose productivity is less than the hourly cost of insurance plus the minimum wage and other workers whose productivity is above this level, they face a problem in complying with the law. Firms may want to cover very productive workers to attract higherquality workers than their competitors or to reduce turnover costs, but if they offer health insurance to those workers, they must also cover their full-time, low-skilled employees. Firms would be forced to pay these lowskilled workers more than they are worth to the rm.
888
Firms have several ways to avoid this situation. First, they could choose to not sponsor health insurance entirely or to lower the generosity of coverage for all employees. In this case, the nondiscrimination clause may actually decrease total health insurance coverage. Second, rms could decide not to self-insure; they would therefore not be subject to the clause. This option is costly for rms, since they would then become subject to state benet mandates and state premium taxes. Third, rms could cheat: they could continue to offer health insurance to only their highly skilled employees but run the risk of discovery and the loss of the favorable tax treatment. Fourth, rms could alter waiting periods and contribution rates to minimize the coverage of their low-wage workers. Fifth, rms could reduce the number of low-wage workers but increase the hours of the remaining workers to spread the costs of the health insurance over more hours. Finally, since the rule applies solely to the coverage of full-time employees, rms may hire low-skilled workers under alternative work arrangements, legally freeing them from offering these workers the same level of health insurance benets that they provide to highly compensated employees while retaining the favorable tax treatment. These arrangements might include part-time or part-year work, hiring temporary workers, or contracting/outsourcing tasks formerly done by direct employees of the rm. Although any response that results in a loss of coverage is of concern to us, the employment effect is the primary area of concern. Lowskilled workers would not only remain largely uninsured but some would also lose full-time employment. Not all nonhighly compensated employees under the legal standard are likely to fall into the category of at risk for employment effects. Workers who fall into this category have low productivity relative to the minimum wage and to their health insurance premium costs, as clearly shown in Figure 1, and are a subset of the nonhighly compensated employees as dened in the tax code. Other forces and policies undoubtedly also affect the provision of health insurance by rms. For instance, there are market incentives to offer broad coverage within the rm, such as shifting health risk from older, highly compensated employees to younger employees with lower compensation. This effect would tend to lower the probability of nding employment effects as a consequence of the nondiscrimination rules. In this analysis we assume that changes in other policies and market forces over time and across states are similar for our two dened groups of workers, but we test this assumption through a variety of robustness checks. We identify policy effects by using changes in the minimum wage
Wolaver et al.
889
and health insurance premiums over time and across states. If the nondiscrimination rules have the impacts we anticipate, we expect to see the following differences between high- and low-wage workers:
I
Low-wage workers should be less likely to work at rms that offer any health insurance coverage; At rms where coverage is offered to some employees, low-wage workers should be more likely to be ineligible for those offers; Low-wage workers should be more likely to be part time and ineligible for the health insurance offers at rms where coverage is offered to some employees; and The difference between low- and high-wage workers in the rates of joint part-time status and ineligibility for rm offers should increase with rm size, because rates of self-insurance increase with rm size.
We can estimate the magnitudes of the potential benets and costs of this policy by comparing the rates of health insurance offers, coverage, and part-time employment between low- and high-wage workers. If there are no differences between high- and low-wage workers, then the law, in combination with the minimum wage, is not binding and employment effects should be of little concern to policy makers. Previous theoretical and empirical work on the effects of employer mandates similar to this policy has found few or no employment effects. That work, however, concentrated on unemployment as the outcome, rather than on part-time or part-year work or other alternative work arrangements (Summers 1989; Gruber and Krueger 1991). Furthermore, the employment of low-wage workers has not been examined separately from that of highwage workers; aggregate analysis may hide the detrimental effects of the policy on low-wage workers. It is theoretically possible, depending on the relative value workers place on the health insurance, for benets to increase the employment of some workers at the same time as the employment of others is decreased, resulting in little or no measured aggregate effect on employment. The important question in evaluating any policy is whether the benets outweigh the costs. The cost of this policy is fewer full-time, permanent jobs for low-wage workers; the benet is increased health insurance coverage for other workers. A brief review of economy-wide trends is consistent with an assumption that costs outweigh benets. Over the past two decades, alternative work arrangements, from temporary services and contracted work to part-time work, have increased in frequency. At the same time, health insurance coverage for the working poor has eroded, opening
890
the door for other policy interventions. Studies such as Farber and Levy 2000 found that lack of eligibility was responsible for the decline in coverage for part-time workers (old and new) during their period of analysis. Garrett, Nichols, and Greenman 2001 found that nearly 90 percent of workers in rms that offer coverage are eligible for coverage but that part-time workers and workers with short tenure (less than one year) are the least likely to be eligible. These statistics are complemented by survey data that nd that 21.3 percent of employers cite saving on wage and/or benets as one reason for hiring part-time workers (Houseman 2001). And Blumberg and Nichols 2001: 39 40 states that According to the 1997 Contingent Worker Supplement to the CPS, 3.7 million (about 18 percent) of 20.3 million uninsured workers were in offering rms but were not eligible for that coverage. An additional 6.4 million workers with insurance coverage from another source were also ineligible for their own employers coverage. Of all those ineligible for their ESI [employer-sponsored insurance] offers, 53 percent reported the reason as being that they dont work enough hours per week or weeks per year. About 8 percent said that contract or temporary workers are not allowed in the plan. Twenty-seven percent said that they had not worked for the employer long enough to qualify, and 1 percent cited a pre-existing condition. About 11 percent cited other reasons. . . . While the lack of eligibility for workers with an offering employer represents a minority of those workers who are uninsured, it is a status which appears to be growing in importance over time. Between 1988 and 1997, eligibility conditional on an employer sponsored insurance offer fell from 94.3 percent to 91.3 percent. But eligibility for peripheral workers (those on the job less than a year and those in part-time jobs) fell more dramatically, from 79.8 percent to 69.6 percent, over the same period. The largest percentage point decline (23) was for part-time workers in jobs for less than a year their eligibility fell from 58.6 percent to 35.5 percent. Part-time workers in longer standing jobs saw their eligibility rates fall by 10 percentage points, to 67 percent in 1997, and eligibility for full-time workers in new jobs fell to 80 percent.
Methods
Our analysis estimates the probability of part-time work, health insurance offers, and wages simultaneously, since these aspects of a job will be determined together and affect each other. We create a categorical vari-
Wolaver et al.
891
able that equals 0 if the worker is full time with offer of health insurance, 1 if full time without offer, 2 if part time with offer, and 3 if part time without offer. Although it is logical that full-time employment with an offer of insurance is the most desirable category and part time without an offer the least, it is not as clear whether part-time employment with an offer dominates full time without. To allow for exibility, we model this categorical variable using multinomial logit specication. (See Appendix B for a detailed description of the econometric methods and for selected regression results.) These equations are estimated simultaneously, for the full sample and for the low- and high-wage samples separately, using generalized method of moments. They are run in two stages, rst with the probability that a rm offers health insurance to any worker and second with the probability that the worker is individually offered insurance. Because of the simultaneous nature of the estimations, covariates have both direct and indirect effects on the outcomes. The simultaneity in combination with the nonlinearity of the multinomial logic makes direct interpretation of the coefcient estimates impossible. Therefore, in addition to presenting coefcient estimates in Appendix Table B1, we construct predicted probabilities and change the values of one variable of interest at a time to infer the total direct and indirect marginal effects. In order to control for the possibility that the worker works part time voluntarily and to help identify the model, we include household characteristics, such as the presence of a child younger than age six, the female dummy variable interacted with a child younger than age six, and household size, in the insurance/part-time status equations. We also include the insurance status of the spouse and an imputed premium to capture the expected cost to the rm (see below) in these equations. In the wage equation only, we include region indicators to control for labor market conditions; we do not expect preferences for part-time work or health insurance to differ by region. We also use two denitions of part-time work in the analyses: all part-time workers and those who indicate that they are part time for economic reasons (our involuntary part-time measure). The estimates using the latter denition shift some part-time workers to full-time employees though they work part-time hours.
Data
The data are pooled from the 1988 and 1993 Employee Benets Supplements to the Current Population Surveys, which ask questions about cur-
892
rent employment and benets. We restrict our analysis to men and single women because we are most concerned with involuntary part-time employment, and married women are traditionally more likely to desire fewer work hours and exible work arrangements. This restriction strengthens our conclusions regarding the employment effects of the policy. The primary advantage of these data is the information on whether or not a rm offers coverage, even if the individual was not eligible for the offer or turned down coverage. We concentrate on these years because the growth of health insurance premiums was substantial and the minimum wage increased over this time period.4 The Current Population Survey does not contain information about actual health insurance premiums for the respondents. Instead, we use an imputation based on regression results, using the 1987 National Medical Expenditure Survey (NMES) and worker and job characteristics to predict the cost of health insurance for each worker. These characteristics include job hazards merged from the Dictionary of Occupational Titles, occupation, industry, rm size, marital status, household characteristics, education, age, job tenure, urbanrural status, and region. We estimate the model using a maximum-likelihood exponential regression on the logged premium and construct the predicted log premium using the coefcient estimates and the characteristics of the CPS respondents. By using the exponential regression instead of a linear regression on the premium levels, we avoid the retransformation problem. Appendix Table B1 contains the premium regression results from the 1987 NMES. We inate the premium using the medical care price index for individuals in the 1988 and 1993 sample and then use the CPI-U to convert the values into real 1988 dollars. Our method of identifying the effects of the law is to divide workers into two groups, those who are at risk for employment effects and those who are not, strictly on the basis of the theoretical model. We do this by using the real minimum wage (in 1988 dollars) in the year and state and the real hourly cost of health insurance for each worker. A low-wage worker is dened as a worker whose wages are less than the hourly cost of health insurance plus the minimum wage. By this denition, 11.7 percent of primary earners are considered to be low-wage workers. It is possible, however, because the minimum wage and health insurance premiums vary, that some workers classied as high wage might have lower
4. From 1995 to 1999 health insurance premiums grew at a lower rate than cash compensation. This has now reversed again as premium growth is outstripping the growth of cash compensation (see below for more detail on this).
Wolaver et al.
893
hourly wages than some workers classied as low wage. Approximately 9.6 percent of high-wage workers have lower hourly wages than the maximum wage in the low-wage sample. Because high- and low-wage workers vary in other characteristics, Table 1 provides descriptive statistics for selected variables used in the analysis for the full sample and for the high- and low-wage samples separately. We focus on employment effects, using results from multiple regression analysis controlling for many factors that affect the probability of health insurance offers, wages, and part-time work.
Results and Analysis: Did the Nondiscrimination Clause Increase the Coverage of Low-Wage Workers?
The nondiscrimination clause is limited in its potential for expanding health insurance coverage because it applies only to businesses that offer self-insured health insurance plans. These rms, however, employ a substantial portion of the workforce. A conservative estimate is that 37 percent of primary earners are offered a self-insured plan. Larger rms are more likely to self-insure; estimates of the percentage of insured employees in self-insured plans vary from 11.1 to 16.5 percent in rms with fewer than one hundred employees to 57.8 to 62.6 percent in rms with more than ve hundred employees in 1987 and 1993 (Acs et al. 1996). On the basis of these data and the distribution of high- and low-wage workers by rm size in the CPS data, we estimate that about 27 percent of low-wage workers work at rms that offer health insurance and self-insure, compared to 39 percent of high-wage workers.5 The difference in probabilities between low- and high-wage workers is a product of the lower probability that low-wage workers work at large rms as well as the lower probability that they work at rms that offer any insurance.6 The nondiscrimination clause does not provide any means to increase the take-up of insurance offers. Some workers indeed, a growing percentage turn down insurance that is offered to them by their employer,
5. Authors calculations, based on the distribution of rm size by wage group, rm offers by wage group, and the distribution of employees in self-insured plans by rm size from Acs et al. 1996. Among workers who are offered health insurance, however, low- and high-wage workers are closer in the probability that the plan is self-insured -40 and 42 percent for low- and highwage workers, respectively. 6. Throughout this article, some empirical results will be cited that cannot be found in the accompanying tables or graphs. These ndings are based on research completed by the authors and further explanations or details are available upon request.
894
at least in part because their costs increased owing to higher employee premium shares and higher premiums. The average proportion of the premium paid by the employer has declined: for single coverage, between 1988 and 1996, the proportion paid by small rms (those with fewer than two hundred employees) fell from 88 to 67 percent, whereas in large rms it fell from 87 to 78 percent (Gabel, Ginsburg, and Hunt 1997). Some workers who turn down coverage are insured by other means, perhaps through a spouse, but many remain uninsured. As Table 1 shows, low-wage workers are less likely to have potential coverage through a spouse and are also less likely to have any coverage. Our analysis shows that 20 percent of primary earners (men and single women) do not have employer-provided insurance themselves or access through a spouse; 11 percent of these workers therefore have no insurance coverage at all. Further examination of these workers (not in Table 1) shows that 25 percent of the workers without their own insurance were not offered coverage through their employer; the rest turned down an offer. Only a small group of ineligible workers is employed at rms that offer insurance benets to some, but not all, of their employees. Figure 2 illustrates this point; 93 percent (81.3/87.1 percent) of workers at rms that offer insurance to any of their employees are eligible for the offered health insurance. The differences across wage groups and between part-time and fulltime status are clear. Recall our hypothesis that employers would like to avoid offering health insurance coverage to low-wage workers because the minimum wage prevents the worker from paying for the coverage via lower wages. Figure 2 illustrates the percentage of workers at rms that offer insurance to any employees and the percentage of workers who have their own offer of health insurance, according to whether they work part or full time and their wage status. We expect that high-wage workers are more likely to work at rms that offer health insurance and more likely to be eligible for that offer. Because the nondiscrimination clause excludes part-time workers, we expect part-time workers to have fewer offers than full-time workers. Over 90 percent of the employers of high-wage, full-time workers offer health insurance, and almost all workers are individually eligible for that offer (Figure 2). Low-wage workers, however, are less likely to work at rms that offer any coverage, are less likely to be individually eligible for that coverage, and are less likely to accept those offers. Only 83 percent (50.8/61.5 percent) of low-wage, full-time workers are eligible for the offered rm coverage. There are several possible reasons for the differences in the eligibility of low-wage, full-time workers relative to high-
Table 1 Full Sample .117 (.32) .87 (.34) .83 (.38) .020 (.061) .019 (.14) .85 (.36) (.11) (.16) (.39) (.42) (.31) (603) (6.05) (.19) (.30) (11.4) (.45) (.50) (.49) (1.87) 2274 11.82 .023 .074 38.4 .26 .48 .37 2.48 .012 (.11) .027 (.16) .81 (.39) .76 (.43) .11 (.32) 2250 (604) 10.93 (6.10) .039 (.19) .10 (.30) 37.5 (11.4) .30 (.46) .46 (.50) .39 (.49) 2.49 (1.88) .010 .013 .87 .82 .071 (.10) (.11) (.34) (.38) (.26) (608) (5.84) (.15) (.26) (11.1) (.44) (.50) (.48) (1.84) .020 (.14) .020 (.14) .85 (.36) .014 (.12) .009 (.09) .89 (.31) .117 (.32) .87 (.34) .82 (.38) .91 (.29) .88 (.33) .56 (.50) .42 (.49) .067 (.25) .099 (.30) .49 (.50) .032 (.18) .13 (.34) .39 (.49) .31 (.46) .40 (.49) 2065 (523) 3.90 (.75) .17 (.37) .32 (.46) 33.1 (12.5) .50 (.50) .27 (.44) .58 (.49) 2.59 (2.07) Weighted Sample High-Wage Sample Low-Wage Sample
Variables Means and Standard Deviations (in Parentheses)
Variable
Low-wage worker Firm offers insurance Worker is eligible for offer Involuntarily part-time Firm offers insurance Firm does not offer insurance Full-time,a rm offers insurance Involuntarily part-time Individual offer of insurance No individual offer of insurance Full-time,a individual offer Covered by own employer Not covered by any source Annual premium ($) Real wage ($) Involuntarily part-time Part-time Age (years) Female Firm 1,000+ employees Firm < 100 employees Household size
.013 .027 .81 .76 .11 2249 10.90 .040 .10 37.8 .29 .45 .39 2.50
Table 1 Full Sample (.38) (.15) (.49) (.46) (.34) (.37) (.42) .18 (.39) .024 (.15) .57 (.49) .31 (.46) .14 (.35) .17 (.37) .23 (.42) 24,039 .18 .019 .56 .34 .13 .13 .25 21,234 (.39) (.14) (.50) (.47) (.41) (.34) (.43) Weighted Sample High-Wage Sample
Variables Means and Standard Deviations (in Parentheses) (Continued) Low-Wage Sample .15 (.36) .066 (.25) .65 (.48) .10 (.30) .20 (.40) .40 (.49) .068 (.25) 2,805
Variable
Presence of child < 6 Female Presence of child < 6 High school education College education Nonwhite Less than 1 year job tenure Union membership N
.18 .025 .57 .31 .13 .16 .23 24,039
Sources: 1988 and 1993 Employee Benets Supplements, Current Population Surveys, Primary Earner Sample (U.S. Census Bureau and U.S. Bureau of Labor Statistics 1988, 1993). aFull-time + voluntary part-time workers.
Wolaver et al.
897
100 92.2
90 80 70 61.5 63 76.1
89.9
87.1 81.3
60 Percent
50.8
50 40 30 -
44.4
23.6
20 10 0Low-wage, Part-time Low-wage, Full-time High-wage, Part-time High-wage, Full-time Total
Firm Offers Health Insurance
Individual Is Offered Health Insurance
Figure 2 Health Insurance Offers, by Wage and Part-time Status Note: Males and single females only. Sources: U.S. Census Bureau and U.S. Bureau of Labor Statistics 1988, 1993.
wage, full-time workers. Low-wage workers are more likely to work at smaller rms, which are less likely to self-insure and therefore not subject to the nondiscrimination law. Moreover, part-time status is not the only exemption in the clause; rms were also exempt from including workers under the age of twenty-ve, part-year workers, and workers with fewer than three years of job tenure. Each of these characteristics is also associated with lower wages. If we assume that no low-wage, full-time workers would have been offered coverage by an employer in the absence of the law, an upperbound value for the benet of the law can be estimated by the fraction of these workers who are offered and accept employer-provided health insurance coverage. From Table 1, 42 percent of low-wage workers are eligible for an offer, but only 31 percent of low-wage workers are covered by their own employer. Assuming that none of these workers would have
898
been offered insurance or would have been covered without the nondiscrimination clause a strong assumption that maximizes the estimated benets of the policy our results suggest that the benets of the clause are an increase of thirty-one percentage points in coverage among lowwage workers. These estimates are clearly overly optimistic; since highly compensated employees tend to be older, rms may have previously extended health insurance offers to the younger workers to increase the risk pool, subsidizing the costs of the older, less healthy employees.
Did the Nondiscrimination Clause Reduce Full-Time Employment for Low-Wage Workers?
Recall that economic theory predicts that low-wage workers, for whom the minimum wage limits the wage adjustment that can be made to cover their health insurance costs, should have higher rates than high-wage workers of part-time work with ineligibility for coverage, who should not be subject to any employment effects because their employers can fully lower wages to adjust for the health insurance premium. If workers are not willing to accept lower wages (that is, they do not value the health insurance as much as the benet costs the employer), then the high-wage and the low-wage groups should both manifest increased rates of parttime work and ineligibility for coverage; these preferences would be taken into account in the estimates presented. The impact of the clause would be overestimated, however, if low-wage workers are less willing than high-wage workers to give up wages, and the observable characteristics used in the regression analysis do not adequately control for this fact. As discussed above, part-time or part-year hiring is among the available options if rms want to provide insurance to only part of their workforce. Some employees prefer part-time work arrangements because they allow exible scheduling, but many workers would prefer full-time employment; these are the workers that are potentially harmed by the nondiscrimination rules. We have taken into account some of these differences by controlling for factors that are associated with voluntary part-time work, such as the presence of a small child in the household and household size. In our sample, the number of workers desiring exible scheduling should be small, because we selected only primary earners (men and single women). In one set of regressions, we also limit our denition of part time to only those who indicated that they were part time due to economic reasons. Our last alternative to take into account individual pref-
Wolaver et al.
899
erences regarding part-time work is to assume that the same fraction of high- and low-wage workers desires part-time jobs. An estimate of the impact of the law on part-time employment is, therefore, the difference between the fraction of low-wage workers and the fraction of part-time, high-wage workers who are ineligible for health insurance coverage. Based on the raw regression results (sample in Appendix Table B2), in Table 2A we predict employment effects of the following magnitude: 11 percent of at-risk low-wage workers (those at rms that offer insurance) on average will be part time and ineligible for an offer of health insurance, compared to only 1 percent of high-wage workers, producing an estimate of the effect of the nondiscrimination clause of 10 percent. Although we are examining only heads of households, some of these low-wage workers could still be part time by choice. Thus, we also compare, in Table 2B, rates of coverage for high- and low-wage workers by limiting the denition of what is a part-time job. The differences are smaller, but still striking; 5.8 percent of at-risk low-wage workers compared to 0.4 percent of high-wage workers are both involuntarily working part time and ineligible for coverage, resulting in an estimate of 5.4 percent increased part-time employment for at-risk workers. These differences are statistically signicant.
Differences by Firm Size
As discussed above, at any rm size, low-wage workers should be more likely than high-wage workers to be part time and ineligible for coverage. Our rst robustness test uses the variations in self-insurance status by rm size to examine whether the law has an effect. Tables 2A and 2B show our regression-based predicted probabilities that a worker is both part time and ineligible for a health insurance offer according to rm size and wage status. (Standard errors are estimated by bootstrapping.) For high-wage workers, the risk of part-time employment should be invariant across rm size; low-wage workers should be increasingly part time and ineligible for coverage as rm size increases. We nd that very few high-wage workers are both part time and ineligible for health insurance, regardless of employer size, whether we use all part-time work or involuntary part-time work to dene the dependent variable. Low-wage workers, in contrast, are more likely to be employed part time (Table 2A) or to be involuntarily employed part time (Table 2B) and ineligible for insurance in rms of all sizes. As rm size grows, the rate for involuntary part-time workers grows from 4.3 to 7.2 percent, and for
Table 2A Predicted Fraction of Primary Earners Working Part-time and Ineligible for Health Insurance Coverage (Standard Error in Parentheses): Workers at Firms That Offer Insurance to at Least Some of Their Employees Difference (b a) .09* (.02) .07* (.03) .11* (.02) .10* (.01) .11 .09 = .02 (0.2) .027 .024 .053 .029 Low-Wage Workers, High-Wage Coefcient (c) Difference (b c) .076 .027 .091 .083 .091 .076 = .015
Firm Size ( N employees) .10 (.02) .08 (.02) .12 (.02) .11 (.01)
High-Wage Workers Low-Wage Workers (a) (b)
100 or fewer 100 999 1000 or more
.011 (.003) .013 (.003) .013 (.002)
Total
.012 (.003)
Difference in difference
Largest to smallest rms
Note: Category a and b predictions based on sample-specic, weighted worker characteristics and regression coefcients; category c predictions used lowwage worker weighted characteristics and high-wage worker regression coefcients. Primary earners are all males and single females, between the ages of 18 and 64. Low-wage workers are dened as those with wages less than the minimum wage plus an imputed hourly cost of health insurance. Part-time is dened as all part-time workers. Involuntary part-time is dened as part-time for economic reasons. Sources: Authors calculations from regression results using 1988 and 1993 Employee Benets Supplements, Current Population Surveys (U.S. Census Bureau and U.S. Bureau of Labor Statistics 1988, 1993). Bootstrapped standard errors. *Statistically signicant at the .05 level.
Table 2B Predicted Fraction of Primary Earners Working Part-time and Ineligible for Health Insurance Coverage (Standard Error in Parentheses): Workers Involuntarily Part-time and Ineligible for Coverage, by Firm Difference (b a) .039* (.013) .068* (.034) .067* (.020) .054* (.011) .067 .039 =.028 (.02) .009 .008 .009 .011 Low-Wage Workers, High-Wage Coefcient (c) Difference (b c) .035 .063 .061 .049 .061 .035 = .026
Firm Size ( N employees) .043 (.013) .072 (.035) .072 (.020) .058 (.012)
High-Wage Workers Low-Wage Workers (a) (b)
100 or fewer 100 999 1,000 or more
.0036 (.0018) .0044 (.0018) .0052 (.0016)
Total
.004 (.0014)
Difference in difference: Largest to smallest rms
Note: Category a and b predictions based on sample-specic, weighted worker characteristics, and regression coefcients; category c predictions used lowwage worker weighted characteristics and high-wage worker regression coefcients. Primary earners are all males and single females, between the ages of 18 and 64. Low-wage workers are dened as those with wages less than the minimum wage plus an imputed hourly cost of health insurance. Part-time is dened as all part-time workers. Involuntary part-time is dened as part-time for economic reasons. Source: Authors calculations from regression results using 1988 and 1993 Employee Benets Supplements, Current Population Surveys (U.S. Census Bureau and U.S. Bureau of Labor Statistics 1988, 1993). Bootstrapped standard errors. *Statistically signicant at the .05 level.
902
those who are both part time it rises from 9 to 11 percent. It is striking that the rates of part-time work vary so little (by at most one-tenth of one percent) for high-wage workers as rm size varies; this nding is consistent with our theoretical story that higher-income workers should not generally be subject to employment effect because employers can adjust wages to pay for the employees health insurance.
Robustness Tests of Worker Preference Differences
Several other tests of the hypothesis are possible to account for other differences between high- and low-wage workers. First, we predict the fraction of low-wage workers who would be expected to be part time on the basis of their observable characteristics and the coefcient results from the high-wage regressions (column c in Tables 2A and 2B). Intuitively, these estimates take into account differences in preferences between highand low-wage workers captured by differences in their unobservable characteristics. In other words, the difference between column c and column a in Tables 2A and 2B is an alternative estimate of the fraction of lowwage workers who are involuntarily part time due to the policy. Under these assumptions, the predicted effect of the law falls to between 4.9 and 8.3 percent of workers (depending on which denition of part-time work is used as an outcome). We can also identify a subgroup of the high-wage workers that is closer to low-wage workers than is the full sample of high-wage workers. Some workers who in 1988 were dened as high wage would be reclassied as low-wage workers under the conditions of 1993, when real health insurance premiums and the minimum wage were higher; we term this subgroup pseudolow wage 1993 (column a in Table 3). The pseudolow wage 1993 workers are those whose wages in 1988 could be adjusted downward to pay for health insurance, but who would be subject to employment effects in 1993. This is a subgroup of high-wage workers who should be more similar to low-wage workers. Similarly, some workers dened as low wage in 1993 would be reclassied as high-wage workers under the conditions of 1988, when real health insurance premiums and the minimum wage were lower; we term this subgroup pseudohigh wage 1988 (column b in Table 3). This is a subgroup of low-wage workers who should have greater similarities with high-wage workers than with other lowwage workers. Finally, some workers would be dened as low wage in either year (column c in Table 3).
Table 3 1993 Low Wage, Would Be 1988 High Wage (b) Difference (b a) All Other Low-Wage Workers (c) Difference (c a)
Comparisons of a Subset of High-Wage Workers to Low-Wage Workers
1988 High Wage, Would Be 1993 Low Wage (a)
5.1
30.2 (45.9) 14.1 (34.9) 11.6 (32.0) 5.9 (23.5) 10.8 7.4 1128
8.4
2.0
3.6
Workers at rms that offer health insurance Part-time 6.2 18.36 (24.2) (38.8) Involuntary part-time 3.5 7.9 (18.3) (27.0) Part-time and ineligible 3.2 8.3 (17.6) (27.7) Involuntary part-time and ineligible 2.3 4.3 (14.8) (20.3) Predictions from regression results (only workers at rms that offer insurance) Part-time and ineligible 1.4 10.5 Involuntary part-time 0.5 4.1 and ineligible Number of observations 314 446 (those with rm offer) 9.1 3.6
9.4 6.9
Notes: Category a workers would change category from high-wage to low-wage because of real state or federal minimum wage increase or health insurance premium ination between 1988 and 1993. Category b workers would change category from low-wage to high-wage because of lower real state or federal minimum wage or lower health insurance premiums in the earlier year. Category c workers are all low-wage workers except for those in category b. The predictions from the regressions are based on the actual worker characteristics, including the sample year they are drawn from, but workers who would switch wage categories are given the coefcients from the counterfactual wage regression. Part-time is dened as all part-time workers. Involuntary part-time is dened as part-time for economic reasons. Sources: Authors calculations from 1988 and 1993 Employee Benets Supplements, Current Population Surveys (U.S. Census Bureau and U.S. Bureau of Labor Statistics 1988, 1993).
904
Table 3 shows the comparisons between these three groups based on both the raw data and the predictions based on the regression results. These results are consistent with the hypothesis that the nondiscrimination clause increased part-time employment for low-wage workers. The rates of all part-time work for the subgroup of high-wage workers who are closest to low-wage workers are 3.2 percent, compared to 8.3 percent for the subgroup of low-wage workers who are closest to high-wage workers and 11.6 percent of low-wage workers would not change status, regardless of the sample year. The comparisons using involuntary part-time work as the dependent variable follow the same pattern: 2.3 percent compared to 4.3 percent and 5.9 percent. When the predictions from the regression results are used, the effects become starker: only 1.4 and 0.5 percent of the subgroup of high-wage workers are predicted to be part time or involuntarily part time and ineligible respectively, compared to 10.5 and 4.1 percent for the low-wage subgroup and 10.8 and 7.4 percent for the always low-wage subset, respectively. These gures imply that the policy led to an estimated increase in parttime employment of between 2.0 and 8.4 percent, somewhat lower in magnitude than other estimates, but still a signicant proportion of low-wage workers at risk of employment effects. Using the predicted percentages from the regression results to measure the differences, the effects are larger, ranging from 3.6 to 9.4 percent of low-wage workers affected by the legislation, depending on the denition of part-time work used.
Robustness Tests Accounting for Other Economic Changes: Year Effects
It is possible that the methods used here overestimate the effects of the law on employment. These effects could, in part, be attributable to changes in the economy or other policies between 1988 and 1993. We account for changes in the minimum wage and health insurance premiums, but rates of part-time work were certainly increasing over this period, perhaps for many other reasons. Mishel, Bernstein, and Schmitt 1997 calculates that rates of total part-time work increased from 18.1 to 18.8 percent between 1989 and 1993, roughly the time period of the sample; rates of involuntary part-time work increased from 4.3 to 5.5 percent. Our predictions are based on the regressions, which control for year but are run over the combined 1988 and 1993 samples; they may thus be capturing other time differences in addition to minimum wage changes and ination in health insurance premiums. As a test, we compare how the fractions of low- and
Wolaver et al.
905
Table 2C Predicted Fraction of Primary Earners Working Part-time and Ineligible for Health Insurance Coverage: Involuntarily Part-time and Ineligible for Coverage, by Year Sample Year High-Wage Workers Low-Wage Workers (a) (b) .017 .012 .0061 .0044 .078 .106 .066 .058 Difference (b a) .061 .094 .060 .054
All part-time 1988 1993 Involuntary part-time 1988 1993
Note: Category a and b predictions based on sample-specic, weighted worker characteristics and regression coefcients. Primary earners are all males and single females between the ages of 18 and 64. Low-wage workers are dened as those with wages less than the minimum wage plus an imputed hourly cost of health insurance. Part-time is dened as all part-time workers. Involuntary part-time is dened as part-time for economic reasons. Source: Authors calculations from regression results using 1988 and 1993 Employee Benets Supplements, Current Population Surveys (U.S. Census Bureau and U.S. Bureau of Labor Statistics 1988, 1993).
high-wage workers who are predicted in Table 2C to be both part-time and ineligible for coverage change between 1988 and 1993, using the coefcients on the year dummies, for both denitions of part-time work. Using the involuntary part-time work denition, the impact of the law actually dampens over time, so our results for this denition do not appear to be due to other economic factors. The opposite is true for all part-time work, however; we estimate that those results may be inated by as much as 1.8 percentage points.7
Difference-in-Difference Estimates by Firm Size
As in most ex post studies of policy changes, the data for a perfect test of the effects of the nondiscrimination clause are not available. The data used here are less than ideal in two respects: (1) The time period does not pre- and postdate the inception of the nondiscrimination rules; instead we rely on variation in minimum wage policy and premium costs over time and place to identify the role of the nondiscrimination policy. (2) The lack
7. Percentage of the 3.3 (9.4 minus 6.1) percentage point increase in the difference between low- and high-wage worker rates of part-time work and ineligibility.
906
of exact information on self-insured status hampers our identication. Firm size is the best correlate of self-insured status available and offers some help in identication.8 If other policies or market changes explained the difference in involuntary part-time work between low- and high-wage workers under our denition, they would not only have to operate differently for both low- and high-wage workers but would also have different effects by rm size, coincident with minimum wage changes and growth in health insurance premiums cross-sectionally and over time. An estimate of the policys effects that could potentially account for the above possibility is the difference-in-difference method. This method takes, rst, the difference in the rates of part-time work and insurance ineligibility by wage status and, then, the differences in those rates between the smallest and largest rms. Under this method, we estimate that the law might have increased the rate of part-time work among potentially eligible low-wage workers by 1.5 to 2.8 percent depending on the denition of part-time work used (see the difference-in-difference in the fraction of workers who are part time and ineligible for coverage in Tables 2A and 2B). Another, simpler comparison is the difference between rates of parttime work and ineligibility among low-wage workers at the largest and smallest rms. Using this as a measure, the impact of the law ranges from 2 to 2.9 percentage points. The difference-in-difference estimates are best viewed as underestimates or lower bounds of the true effect of the legislation. In fact, these estimates cannot be cleanly interpreted because rates of self-insurance are not perfectly correlated with rm size. If no small rms and all large rms (or close to it) are self-insured, then the difference-in-difference results would be the best estimate of the impact of the clause. However, since 11.1 (in 1988) or 16.5 (in 1993) percent of workers in small rms should be affected by the legislation, compared to 60.9 (in 1988) or 60.6 (in 1993) percent of workers in large rms, the identication of the effects using this method is weakened. The results indicate that the nondiscrimination clause may affect as few as 1.5 and as many as 10 percent of low-wage primary earners at rms that offer health insurance.
8. Using the 1993 Robert Wood Johnson Foundation Employer Health Insurance survey, the authors investigated other possibilities, including presence of union employees, for-prot versus nonprot status, age distribution of the rm, and industry. The variation in self-insurance rates with these variables is far less than that by rm size, and for some of these variables, for instance, industry, rates of part-time work vary widely for reasons exogenous to self-insurance status. Disentangling these outside rates from the policy effects would require more information than is currently available.
Wolaver et al.
907
Premium Imputation Effects
An implicit assumption made in the analysis is that the incidence of the health insurance benets is based on an average premium. For self-insured rms, the premium imputation method will systematically overestimate the health insurance costs for healthy and underestimate the costs for unhealthy workers. This misclassication is likely to bias the results away from nding any effect of the clause for the following reasons. In the rst place, sicker workers are more likely to work part time and to have higher health care costs and higher premiums. Since we do not observe health status, we will tend to misclassify some sicker low-wage workers as high wage, because the premium we impute to them will be too low. Conversely, healthy workers are more likely to be misclassied as low-wage workers, because the imputed premium will overstate their true premium costs. In combination, these two measurement errors will bias our differenced rates of part-time work and ineligibility for health insurance to zero. Our results therefore underestimate the effects of the nondiscrimination clause. Another effect of imputation may be to overestimate premium costs for young workers and underestimate premium costs for older workers; this, combined with the higher probability nationally that younger workers are part time, may cause our results to be an overestimate of the effects of the law.9 Restricting our sample to primary earners only limits this bias; as a check, part-time workers in our sample are, on average, about a year younger than full-time workers.10 Henry David Thoreau pointed out in an 11 November 1850 entry to his journal that some circumstantial evidence is very strong, as when you nd a trout in the milk. In every real, predicted, and hypothetical comparison of low- and high-wage workers, we nd some evidence that the legislation has effects. Even though the evidence is circumstantial, we believe that those effects are present. Although we cannot offer a precise point estimate of the impact of the law, we have presented some bounds for the effects. Our lower-bound effect is very likely underestimated, whereas our upper-bound estimates may be either over- or underestimated, depending upon which uncontrolled effects dominate. The strict
9. As of 2001, the proportion of employed persons working part time was 26.3 for those aged twenty to twenty-four, 11.1 for those aged twenty-ve to fty-four, and 24.2 for workers aged fty-ve and above (U.S. Bureau of Labor Statistics 2000). 10. The authors wish to thank an anonymous referee for the difference-in-difference interpretation of the results and for calling attention to the implicit assumption of an average, rather than individual, incidence of the health insurance benet in the estimates.
908
division of low- and high-wage workers according to economic theory fully justies our base estimates for the impact of the law, but given the absence of pre- and postlegislation data, we have performed every other feasible comparison to test the robustness of our results. We present these estimates not as denitive measures of the impact of the law, but simply to establish the presence of effects, with a range of possible measurements. These are small economy-wide impacts, but as the following section suggests, they may produce large distributional effects. Ironically, the cost of the law is placed on the very workers that the law was intended to aid by increasing their access to health insurance coverage.
The Effect on Inequality
Economists have been grappling with an explanation for the increase in earnings and income inequality that began in 1979, after several decades of earnings convergence. These events cannot be attributed to any single cause, but the growing cost of fringe benets and the nondiscrimination rules may contribute. Figure 3 shows the growth and composition of the employment cost index. The ination in health insurance slowed in the mid1990s, but in general the costs of nonwage compensation have outpaced the growth in wages. Recently, however, the United States has experienced faster growth in nonwage compensation costs, as it did in earlier periods. As health insurance costs grow, rms will pay more attention to their offer of insurance coverage and to recovering the costs of their rms contributions.
Earnings Inequality Effects
If low-wage workers are now increasingly likely to work fewer hours because of the nondiscrimination policy, their labor earnings are further depressed. These low-wage workers compose roughly the bottom 12 percent of the wage distribution of primary earners. According to the estimates presented here, between 0.6 and 5.6 percent of all low earners (based on the percentage of low-wage workers at rms that offer any insurance) experience reduced hours of work because of the nondiscrimination clause. Depending on how many fewer hours they work, the annual earnings loss could range between $975 and $3,900, with an average loss of $2,925.11 This loss represents 12.5 50 percent of the average
11. The average hourly wage of these workers in 1988 dollars was $3.90. If the worker receives ve fewer hours per week for fty weeks a year, the income loss is $975. If the worker
Wolaver et al.
909
9 -8 -7 -6 --
Percent Percent
5 -4 -3 -2 -1 -0 -82 83 84 85 86 87 88 89 90 91 92 93
Benets
Wages and Salaries
94
95
96
97
98
99
2000
Year Quarter Year and Quarter

Figure 3 Employment Cost Index for Civilian Workers, 19822001, Changes in Wages and Salaries and Benet Costs, Twelve-Month Percentage Change Source: U.S. Bureau of Labor Statistics 2001.
annual earnings of these workers at full-time, full-year hours, and for them the cost is substantial, even though the employment effect is a small percentage of the total workforce. The mandate regarding health insurance coverage can be compared to another mandate, the minimum wage, which is also intended to help lowskilled workers. In both cases, one can ask whether the mandate achieved its goal. Critics of the minimum wage have argued that it is not successful because (1) many minimum-wage workers are not low income but from families with incomes well above the poverty line and (2) the minimum wage reduces employment because it requires a rate of pay that may be in excess of a workers contribution to the rm. The latter point is clearly directly relevant to the investigation in this article. The evidence is mixed, although the current view is that the minimum wage has either none or a small negative effect on employment, including hours worked (Card and Krueger 2000, Neumark and Wascher 2000). Since the nondisloses twenty hours a week, the income loss is $3,900. The average part-timer in the sample works twenty-ve hours a week, representing a loss of fteen hours per week translated into real annual earnings of $2,925; for comparison, an average worker at forty hours a week for fty weeks per year earns $7,800.
910
crimination clause raises the effective minimum wage for full-time workers at self-insured rms, our estimates of its small negative employment effects are consistent with those of Neumark and Wascher 2000 for the minimum wage.
Additional Equity Effects
A second distributional problem lies in the tax exemption for the value of the health insurance premiums. Workers with higher wages are likely to fall into higher tax brackets. The tax treatment represents a sizable tax break for those workers who least need the help. Low-wage workers are less likely to receive any tax benets, since they are less likely to be offered and to accept coverage from an employer. The taxes of those low-wage workers who are covered are reduced, but not by as much as a high-wage earners taxes on an identical policy. An additional equity concern is that the nondiscrimination mandate does not cost the same across rms or workers in terms of geography, industry, or occupation because health insurance costs exhibit considerable variation along these dimensions. These differences are compounded by state variations in the minimum wage and create horizontal inequities in the incidence of the law. Workers in some states, industries, and occupations will be more likely to be part time due to the increased health insurance or minimum wage costs. Workers in poor health will also be more likely to be subject to these employment effects. All of these costs should be compared to the gains to those workers who remain employed and receive the benets from the increased health insurance coverage. In the next section we attempt to provide some insight into the benets in increased coverage of this legislation.
Discussion and Conclusion
Our tests of the effects of the nondiscrimination policy on employment in this article have in every case found them to be negative, even though rms have several other possible responses, such as noncompliance or shifting away from self-insurance. Comparisons of low-wage workers, dened as those whose estimated total productivity is lower than the minimum wage plus the costs of providing health insurance, to workers who are able to pay for health insurance through reduced wages always show higher rates of part-time employment coupled with ineligibility for health insurance. Employment effects are greater for low-wage workers in large
Wolaver et al.
911
rms in which self-insurance is more common and the policy is more likely to have an impact. These effects remain when many other characteristics of the workers and rms that may affect coverage and part-time work decisions are taken into account. Finally, comparisons of subsets of high- and low-wage workers, who are more similar to each other in their characteristics than the broader groups of workers, also show the hypothesized effects. This article illustrates the unforeseen effects which sometimes plague policy interventions. Certainly, the reduction of full-time employment opportunities for low-wage workers the group the policy was expected to help was an unintended consequence. Although the fraction of the primary-earner workforce that is part time as a consequence of the law is low, probably less than half a percentage point, the set of workers at risk consists of those who will suffer the most from underemployment. The maximum estimated benet of the law to low-wage workers is an increase in employer-provided health insurance of thirty-one percentage points. Casting the provision in the best possible light, this result implies that for every 9.4 low-wage workers who gain insurance coverage, one low-wage worker, on average, loses full-time employment. Translating these effects into dollar terms, for every $19,411 in benets under the law there is an expected cost in lost earnings of $2,965.12 Using this average, the calculated cost of the law is 15 percent of the benets among affected workers. This calculation does not include any other employment effects not examined in this article: loss of all employment, shifts to part-year work, and any costs of increased temporary work.
Possible Impact of the Nondiscrimination Clause in the PostWelfare Reform Era
Welfare-to-work initiatives so prominent in welfare reform policies have, we believe, increased the population of workers vulnerable to the employment effects of the nondiscrimination clause. How has this policy affected the probability of private health insurance coverage for former welfare recipients now entering the workforce?
12. The average earnings loss in dropping from full-time to part-time hours for low-wage workers in this sample is $2,925; we assume that the average premium of $2,065 for low-wage workers is the dollar value of the benet of the law. The range of estimates for all part-time work is 0.8 to 5.6 percent of all low-wage workers affected, with a mean effect of 3.3 percent. Therefore, if the benet of the law is an increase in coverage of thirty-one percentage points, for every 31/3.3, or 9.4, low-wage workers, we expect benets of 9.4 $2,065 and a cost of $2,925.
912
We found that at most 31 percent of low-wage workers in our sample would have gained coverage as a result of the nondiscrimination clause. However, recent welfare leavers may not be comparable to the sample we have studied here for several reasons. First, they may have less job experience and more dependent children. Second, health care costs have increased in real terms, increasing the proportion of workers who would be in our low-wage group, and this may affect the estimates. The addition of welfare leavers into the low-wage labor market will also represent an increase in labor supply, further depressing market wages for this group as a whole, drawing more workers into the pool of those at risk for employment effects and possibly changing the responses of rms. What do recent data on private coverage in poor households tell us about these questions? On average, across the United States about 37 percent of those with incomes below 200 percent of federal poverty levels are covered by employer-sponsored insurance, whereas nearly 42 percent are covered by private insurance, Medicare, or Civilian Health and Medical Program for Uniformed Services (CHAMPUS) (Holahan 2002). But private coverage rates vary among states, from less than 30 percent in New York and California to more than 45.2 percent in Wisconsin. Insurance gaps also vary, from about 50 percent to more than 67 percent of households. Underlying this variation, Holahan 2002 suggests, are state-level factors such as industrial composition, extent of union membership, size of rms, and the human capital of workers. Shen and Zuckerman 2003 found that states with low ESI rates have populations with fewer skills and less human capital. Because of these workforce characteristics, these states get fewer of the high human capital, high paying jobs that also have ESI and as a result have high uninsurance rates.13 It is plausible that welfare leavers will be disproportionately concentrated in the low ESI states and therefore have more difculties nding private health insurance coverage and perhaps will be more subject to employment effects. Little is yet known about the long-term employment prospects of welfare leavers compared to other low-wage workers. Loprest 2002 nds that early welfare leavers (those who left prior to 1997) are more likely than working low-income mothers (those with incomes at or below 200 percent of the federal poverty line) to have full-time employment (69.4 compared to 64.2 percent) but less likely to have employer-sponsored health insurance (23.2 compared to 35.8 percent). This comparison is less than ideal because Loprests sample includes married women, which would
13. It appears to us that states with high ESI tend to be states with very low unemployment.
Wolaver et al.
913
tend to increase the rates of voluntary part-time work, and because early welfare leavers are more likely to be successful than women who exit the rolls later. Moreover, Loprests data refer to the period prior to the recession that began in 2001 and prior to the recent upswing in health insurance premium ination, both of which will impact coverage and employment rates of low-wage workers. Our reading of this literature is that public insurance is the primary source of health care coverage of former welfare recipients and especially their children. However, many former welfare recipients who remain off cash assistance and are employed at least part year are covered by neither public insurance nor their employer; Garrett and Hudman 2002 estimates that only about 20 percent of early welfare leavers had employer-based coverage in 1999. In addition, for many of these individuals, the nondiscrimination mandate may reduce employment opportunities. Our concern is not primarily that there is an efciency cost to the nondiscrimination clause, as shown by an increased probability of (involuntary) part-time work coupled with ineligibility for health insurance among workers who are not productive enough to pay for their coverage through lowered wages. Rather, we are concerned with equity and the best policy to increase health insurance coverage. If the efciency losses were concentrated among high-wage workers, for instance, the price of the nondiscrimination clause would be more palatable. If we are correct in our estimates, the consequences of the policy we explore are being borne by vulnerable members of the population, including recent welfare leavers. We believe the evidence is convincing enough that alternative policies to increase health insurance coverage among low-wage workers should be carefully examined. Several other policy instruments could increase the coverage of low-wage workers without the concomitant loss of full-time employment opportunities. These include Medicaid and/or state Child Health Insurance Program (SCHIP) expansions, tax credits, and universal coverage, although each has its own efciency and equity concerns. Universal health care coverage would eliminate the problems of adverse selection and provide more equity in coverage, but it is unlikely to become politically viable in the immediate future. Incremental reforms such as expansions of the public insurance system or tax credits are more viable policy options that should not have the negative employment effects associated with the nondiscrimination clause. Expansions in Medicaid and SCHIP (some states have included families in coverage as well as children) have an established record of increasing coverage among low-income populations. These programs avoid the
914
adverse selection problems of the individual health insurance market (Pauly and Nichols 2002). Further advantages are modest copayments and experience in provision of care among the low-income population. In some states, however, access to care is problematic because providers receive low reimbursements (Berman et al. 2002). Moreover, crowd-out may reduce employer-provided coverage, possibly increasing both the rate of uninsurance and government budgets by providing public insurance to some households that would have had private coverage in the absence of the policy (Cutler and Gruber 1996). If one of the goals of government intervention is to increase equity as well as health insurance coverage, then the only form of crowd-out that should be of concern is an increase in the uninsured rate. Tax credits are perhaps the most politically attractive option in the current legislative environment. Credits might increase health care coverage without the negative employment outcomes of the nondiscrimination clause if the credits are refundable and generous enough to make a difference.14 A study by Families USA (2001) nds that the tax credits proposed by the Bush administration, in which low-income uninsured individuals would receive tax credits of up to $1,000 and low-income uninsured families would receive tax credits of up to $2,000, would have little impact on the coverage of the low-income population. According to the report, in many states, $1,000 plans were not available and those that were generally provided incomplete coverage, had high deductibles and required high coinsurance or copayments. A Kaiser Family Foundation study (2001) found that individuals with even relatively minor health problems had difculty in obtaining affordable coverage in the individual insurance market; any policy should take into account this factor if it is to be successful at increasing coverage. Tax credits targeted to low-income families would make the tax treatment of health insurance more equitable. But since premiums decrease with the size of the group, low-income workers without access to employer-provided plans would still be at a great disadvantage relative to workers with generous fringe-benet plans. The prospect of crowd-out is not eliminated with the introduction of tax credits for individual purchase of health insurance. Nevertheless, compared to the nondiscrimination clause, such credits should not have negative employment consequences and, if made more generous, could lead to an increase in coverage of this low-wage population.
14. In most plans, the proposed credits are roughly half the current annual premium for family coverage (Feder et al. 2001), which is still well below that provided on average in employerprovided plans.
Wolaver et al.
915
Now that health insurance premium ination is once again exceeding wage growth, as in the period from 1988 to 1994, the employment effects of the nondiscrimination clause may grow. In combination with a weak economy, the effects of the nondiscrimination clause on employment and insurance coverage of low-wage workers are likely to intensify. First, it is clear that low-wage workers are less likely to be employed by rms that offer health insurance. Even for those workers who have potential access to employer-sponsored coverage, the nondiscrimination clause can only have an impact if the plan is self-insured. The policy does nothing to make those plans more affordable for the low-wage worker and creates an incentive for rms to limit hours for workers it wishes to exclude from coverage. Moreover, this article provides some insight into current trends in the rates of the uninsured in the context of the recent welfare reforms and the recent economic downturn. It is not surprising that only a small percentage of former recipients of Medicaid, whether exiting the public safety net because they lost eligibility or not, has been able to nd affordable employer-based insurance coverage.
916
Appendix A: Details of the Nondiscrimination Clause

The clause only applies to employer-provided plans where the rm itself takes on the risk of workers medical expenses, rather than purchasing policies from insurance companies. These rms may contract with an insurer for administrative tasks such as claims processing. Self-insurance affords rms several advantages. These plans are exempt from state mandates and state premium taxes that add to the cost of the insurance, ceteris paribus (Acs et al. 1996). In 1991, approximately 40 percent of employees who had employer-provided insurance were in a self-insured plan (Sullivan et al. 1992). The history of the nondiscrimination clause exemption for non-selfinsured plans is somewhat convoluted. The Internal Revenue Code of 1986 extended the nondiscrimination rules to all employer-provided health insurance; the provisions were scheduled to go into effect after 31 December 1988, but public law 101-104 repealed this extension retroactively. The law stipulates that the value of any differences in coverage between highly compensated and other workers becomes taxable. In discussing the repeal of section 89 of the 1986 Internal Revenue Code, the House Congressional Record states that the justication for the revenues lost due to the tax expenditure for employer-provided health insurance is through increased rates of insurance. On the other hand, the Congress believed that the cost to the Federal Government of tax-favored employer-provided accident and health coverage is not justied if such coverage disproportionately benets highly compensated employees. In order to achieve this objective, nondiscrimination rules were enacted to permit the full exclusion from income of employer-provided health benets only if the benets are provided to required numbers of nonhighly compensated employees (Congressional Record, 101st Cong., 1st sess., vol. 135, pt. 155, 7 November 1989). There are some exceptions that allow employers to exclude some full-time employees; for example, workers during a probationary period and workers under the age of 25 may be exempted from offers. Plans negotiated by unions are also exempt from nondiscrimination rules. The following is the relevant passage from the 1978 legislation requiring that employers offer insurance to all employees (reprinted from U.S. Tax Code, Title 26, Subtitle A, Chapter 1, Subchapter B, Part III, Section 105: (h); accessed on-line at www.fourmilab.ch/ustax/www/t26-A-1-B-III-105.html).
Wolaver et al.
917
(h) Amount paid to highly compensated individuals under a discriminatory self-insured medical expense reimbursement plan (1) In general In the case of amounts paid to a highly compensated individual under a self-insured medical reimbursement plan which does not satisfy the requirements of paragraph (2) for a plan year, subsection (b) shall not apply to such amounts to the extent they constitute an excess reimbursement of such highly compensated individual. (2) Prohibition of discrimination A self-insured medical reimbursement plan satises the requirements of this paragraph only if (A) the plan does not discriminate in favor of highly compensated individuals as to eligibility to participate; and (B) the benets provided under the plan do not discriminate in favor of participants who are highly compensated individuals. (3) Nondiscriminatory eligibility classications (A) In general A self-insured medical reimbursement plan does not satisfy the requirements of subparagraph (A) of paragraph (2) unless such plan benets (i) 70 percent or more of all employees, or 80 percent or more of all the employees who are eligible to benet under the plan if 70 percent or more of all employees are eligible to benet under the plan; or (ii) such employees as qualify under a classication set up by the employer and found by the Secretary not to be discriminatory in favor of highly compensated individuals. (B) Exclusion of certain employees For purposes of subparagraph (A), there may be excluded from consideration (i) employees who have not completed 3 years of service; (ii) employees who have not attained age 25; (iii) part-time or seasonal employees; (iv) employees not included in the plan who are included in a unit of employees covered by an agreement
918
between employee representatives and one or more employers which the Secretary nds to be a collective bargaining agreement, if accident and health benets were the subject of good faith bargaining between such employee representatives and such employer or employers; and (v) employees who are nonresident aliens and who receive no earned income (within the meaning of section 911(d)(2)) from the employer which constitutes income from sources within the United States (within the meaning of section 861(a)(3)). (4) Nondiscriminatory benets A self-insured medical reimbursement plan does not meet the requirements of subparagraph (B) of paragraph (2) unless all benets provided for participants who are highly compensated individuals are provided for all other participants. (5) Highly compensated individual dened For purposes of this subsection, the term highly compensated individual means an individual who is (A) one of the 5 highest paid ofcers, (B) a shareholder who owns (with the application of section 318) more than 10 percent in value of the stock of the employer, or (C) among the highest paid 25 percent of all employees (other than employees described in paragraph (3)(B) who are not participants). (6) Self-insured medical reimbursement plan The term self-insured medical reimbursement plan means a plan of an employer to reimburse employees for expenses referred to in subsection (b) for which reimbursement is not provided under a policy of accident and health insurance.
Wolaver et al.
919
Appendix B: Econometric Methods

Logged wages are modeled as a linear function of a vector of characteristics, Xw and the categories of the insurance/hours variable. The omitted category is Full-time with an offer of insurance. The probability that a worker falls into one of the four insurance categories is given in equation 1 (Greene 1993): [1] Pr(Category j) = exp(j + j,wages Log(wages) + j,kXj,k)
j=0to3 exp(j
+j,wagesLog(wages) + j,kXj,k)
+i,k
As described in the text, economic theory suggests that rms decide simultaneously on the size of the total compensation package, the mix of wages and benets in this package, and whether to hire workers part time or full time. Thus we model the wage equation 2 in the following form, [2]Log(wages) = 0 + pt,no offer Part time,no offer + pt,offer Part time, offer + pt,no offer Full time,offer + Aw'Xw +w where X is a vector of worker/job characteristics. We employ Generalized Methods of Moments (GMM). There are other alternatives to the GMM technique, such as full-information maximum likelihood, but we chose GMM because it is computationally easier to estimate and in cases where errors in specication are more worrisome, yields consistent estimates where a full-information, maximum likelihood model will not (Lechner and Breitung 1996). Appendix Table B1 shows the results from the regression used to impute the health insurance premiums for individuals in the sample. Appendix Table B2 shows the raw coefcient results from the model, which are used to construct the predicted probabilities presented in the text.
920
Appendix Table B1 Results
Premium Imputation, Exponential Regression Coefcient Standard Error
Constant 7.24 .126 * .722 .344 * Hazardsa -.417 .402 PUSa -.163 .118 Mediuma -.387 .216 ** Heavya .415 .051 * Presence of familyb State and local government .033 .059 Industry 1 -.132 .312 Industry 2 .082 .220 Industry 3 .007 .138 Industry 4 .165 .086 Industry 5 .078 .092 Industry 6 -.020 .091 Industry 7 -.038 .100 Industry 8 .065 .124 Industry 9 -.093 .190 Industry 10 .236 .230 Industry 11 -.055 .076 Union .195 .048 * Firm 1 25 employees -.0625 .125 Firm 26 100 employees -.103 .070 *** West Central Region -.003 .113 North Central Region .089 .097 Southwest SC Region -.126 .107 South East SC Region -.130 .112 South Atlantic Region -.133 .095 Pacic Region .078 .101 Mountain Region .033 .12 Mid-Atlantic Region .025 .098 N = 3,045 Log likelihood = -3514.37 Model 2(28) = 154.95, Pr > 2 = 0
Source: Authors calculations from 1987 National Medical Expenditure Survey (U.S. Department of Health and Human Services 1987). aDictionary of Occupational Titles data on average scores by occupation. bMore than one person in the household. *Statistically signicant at the .05 level. **Statistically signicant at the .10 level. ***Statistically signicant at the .15 level.
Appendix Table B2 Involuntary Part-time w/ Offer Log Wage Std. Error Std. Error High-Wage Worker Set * ** ** * Std. Error Full-time, No Offer
Selected Results from Simultaneous Regression Model Individual Offer
Involuntary Part-time, No Offer
Variable
Std. Error
* ** ** * **
* * * *
* **
**
Constant Age age2 Union y1992 Hsdeg Colldeg frm1k1t frm1000p Indn Indman Indtrade Indallot tenlt1yr ten1t4 tenure5t9 west midwest neast priskfam spouseown * * * * * 1.22 1.21 8.77 0.71 ** 11.66 0.74 4.02 0.25 * *
-10.25 -15.97 19.14 -1.95 -0.33 -0.31 -0.88 0.21 0.29 0.42 -0.52 1.25 0.33 2.07 0.76 3.08
4.65 15.19 17.15 0.71 0.58 0.67 1.29 0.39 0.27 0.70 0.66 0.76 0.58 0.76 0.52 1.31
-17.12 -35.42 40.46 0.69 0.32 -2.10 -4.55 0.29 -0.33 -0.89 1.03 1.38 0.92 2.03 1.43 0.77
8.57 18.75 21.16 0.69 0.45 1.09 2.40 0.51 0.32 0.75 0.90 0.90 0.79 0.93 0.64 0.48
-7.89 -21.32 25.39 -1.45 -0.15 -0.67 -1.24 0.12 0.01 -0.72 -0.38 0.50 0.26 2.62 0.69 1.92
2.28 7.02 7.75 0.30 0.23 0.32 0.64 0.16 0.12 0.20 0.23 0.30 0.22 0.38 0.22 0.29
1.64 1.92 -1.90 -0.02 -0.04 0.18 0.33 0.07 0.16 0.04 -0.07 -0.03 -0.11 0.10 -0.04 -0.03 0.13 0.04 0.07
0.51 2.63 2.81 0.23 0.07 0.14 0.13 0.06 0.11 0.20 0.06 0.19 0.06 0.59 0.09 0.06 0.09 0.18 0.17
15.47 1.20
9.30 0.48
** *
Appendix Table B2 Involuntary Part-time w/ Offer Log Wage Std. Error Std. Error ** ** -30.39 1.36 1.11 Objective*N = 5.58 Low-Wage Worker Set 3.19 3.02 -4.34 1.50 0.69 -0.09 -1.12 0.65 0.06 -0.15 -0.22 0.27 27.11 14.62 17.67 0.93 1.18 0.87 3.55 0.39 0.75 4.59 0.96 1.35 5.22 -11.98 12.35 -0.84 -0.35 -0.08 -0.76 -0.62 0.17 0.51 0.01 0.59 15.12 14.64 16.64 1.23 0.66 0.50 1.52 0.74 0.34 1.81 0.66 0.60 1.88 -2.65 2.85 -0.05 -0.02 0.03 -0.09 -0.14 0.02 0.01 0.07 0.09 1.41 5.16 5.72 0.28 0.17 0.10 0.20 0.27 0.09 0.25 0.23 0.24 59.54 13.51 6.48 4.25 0.20 0.52 1.17 2.28 -0.10 -0.35 0.01 1.43 0.07 0.20 0.38 Std. Error 7.81 0.25 -0.15 1.61 Full-time, No Offer
Selected Results from Simultaneous Regression Model Individual Offer (Continued)
Variable
Std. Error
log wage Hhsize kidlt6 fkidlt6 wptnioff wptioff wftnioff
1.04 -0.16 0.25 -0.73
3.14 0.17 0.44 0.71
N = 19,324
constant age age2 union y1992 hsdeg colldeg frm1k1t frm1000p indn indman indtrade
4.42 -20.77 24.09 -1.36 -0.21 -0.77 -1.25 0.51 0.64 -0.52 -3.16 -0.31
33.71 29.34 30.93 1.76 0.84 1.21 3.56 0.94 0.65 2.82 2.29 0.61
Appendix Table B2 Involuntary Part-time w/ Offer Log Wage Std. Error Std. Error Low-Wage Worker Set 0.20 0.82 -0.20 -0.02 2.65 3.47 1.69 0.72 0.67 2.35 0.48 1.97 1.02 2.98 1.38 3.61 0.12 0.49 0.05 0.14 0.07 0.04 0.01 0.20 0.96 0.07 0.33 0.13 0.10 0.10 Std. Error Full-time, No Offer
(Continued)
Variable
Std. Error
-0.81 1.54 0.25 1.88
0.68 3.94 1.93 4.66
indallot tenlt1yr ten1t4 tenure5t9 west midwest neast priskfam spins log wages hhsize kidlt6 fkidlt6 wptnioff wptioff wftnioff -10.69 -0.20 -3.96 0.19 0.99 -1.29 56.80 0.99 27.22 0.32 1.25 3.84 9.33 0.10 -6.84 -0.09 0.30 -0.39 25.48 0.68 16.08 0.21 0.78 1.29 -0.22 -0.07 -2.33 Objective*N = 2.29
8.72 1.08 -4.15 -0.07 0.25 -0.38
55.37 1.35 31.90 0.39 1.19 2.98
2.46 2.97 4.65
N = 1,574
Note: Other variables included but not shown are: region, year, gender, class of worker, race, urbanicity, occupation and marital status. Source: Authors calculations from 1988 and 1993 Employee Benets Supplements, Current Population Surveys (U.S. Census Bureau and U.S. Bureau of Labor Statistics 1988, 1993). *Statistically signicant at the .05 level. **Statistically signicant at the .10 level.
924
References
Acs, G., S. H. Long, S. Marquis, and P. Short. 1996. Self-Insured Employer Health Plans: Prevalence, Prole, Provisions, and Premiums. Health Affairs 15(2):266 278. Berman, S., J. Dolins, S. Tang, and B. Yudkowsky. 2002. Factors That Inuence the Willingness of Private Primary Care Pediatricians to Accept More Medicaid Patients. Pediatrics 110(2):239 248. Blumberg, L. J., and L. Nichols. 2001. Why Are So Many Americans Uninsured? A Conceptual Framework, Summary of the Evidence, and Delineation of the Gaps in Our Knowledge. Paper prepared for a conference on the uninsured, Ann Arbor, MI, July. Brittain, J. A. 1971. The Incidence of Social Security Payroll Taxes. American Economic Review 61(1):110 125. Card, D., and A. B. Krueger. 2000. Minimum Wages and Employment: A Case Study of the Fast-Food Industry in New Jersey and Pennsylvania: Reply. American Economic Review 90(5):1397 1420. Cutler, D., and J. Gruber. 1996. Does Public Insurance Crowd Out Private Insurance? Quarterly Journal of Economics 111(2):391 430. Ellwood, M. 1999. The Medicaid Eligibility Maze: Coverage Expands but Enrollment Problems Persist: Findings from a Five State Study. Washington, DC: Kaiser Commission on Medicaid and the Uninsured. Employee Benets Research Institute (EBRI). 1999. Contingent Workers and Workers in Alternative Work Arrangements. Executive Summary. EBRI Issue Brief No. 207. Washington, DC: EBRI. Available on-line at www.ebri.org/ibex/ib207.htm. Families USA. 2001. A 10-Foot Rope for a 40-Foot Hole: Tax Credits for the Uninsured. Congress Daily, 13 September. Farber, H. S., and H. Levy. 2000. Recent Trends in Employer-Sponsored Health Insurance Coverage: Are Bad Jobs Getting Worse? Journal of Health Economics 19:93 119. Feder, J., L. Levitt, E. OBrien, and D. Rowland. 2001. Covering the Low-Income Uninsured: Assessing the Alternatives. Washington, DC: Kaiser Commission on Medicaid and the Uninsured. Available on-line at www.kff.org/content/2002/2258/2258.pdf. Gabel, J., P. Ginsburg, and K. Hunt. 1997. Small Employers and Their Health Benets, 1988 1996: An Awkward Adolescence. Health Affairs 16(5):103 110. Garrett, B., and J. Hudman. 2002. Women Who Left Welfare: Health Care Coverage, Access and Use of Health Services. Washington, DC: Kaiser Commission on Medicaid and the Uninsured. Available on-line at www.kff.org/content/2002/4050/4050.pdf (accessed 25 July 2002). Garrett B., L. M. Nichols, and E. Greenman. 2001. Workers without Health Insurance: Who Are They and How Can Policy Reach Them? Urban Institute Report to the Kellogg Foundation. Greene, W. H. 1993. Econometric Analysis. New York: Macmillan. Gruber, J. 1999. Health Insurance and the Labor Market. In Handbook of Health Economics, vol. 5, ed. A. J. Culyer and J. P. Newhouse. Amsterdam: Elsevier Science.
Wolaver et al.
925
Gruber, J., and A. Krueger. 1991. The Incidence of Mandated Employer-Provided Insurance: Lessons from Workers Compensation Insurance. In Tax Policy and the Economy, vol. 5, ed. D. Bradford. Cambridge: NBER and MIT Press. Hoffman, C., and M. B. Pohl. 2002. Health Insurance Coverage in America: 2000 Data Update (February). Washington, DC: Kaiser Commission on Medicaid and the Uninsured. Available on-line at www.kff.org/content/2002/4007/4007.pdf. Holahan, J. 2002. Variations among States in Health Insurance Coverage and Medical Expenditures: How Much Is Too Much? Washington, DC: Urban Institute. Houseman, S. 2001. Why Employers Use Flexible Stafng Arrangements: Evidence from an Employer Survey. Industrial and Labor Relations Review 55(1):149 70. Kaiser Family Foundation. 2001. The Uninsured: A Primer. Uninsured Basics (March). Washington, D.C. Kaiser Commission on Medicaid and the Uninsured. Accessed via the Kaiser Family Foundation Web site at www.kff.or/content/2001/2228/2228/pdf. . 2003. The Uninsured and Their Access to Health Care. Fact sheet (January). Washington, DC. Kaiser Commission on Medicaid and the Uninsured. Accessed via the Kaiser Family Foundation Web site at www.kff.org/content/2003/142004/ 142004.pdf. Lechner, M., and J. Breitung. 1996. Some GMM Estimation Methods and Specication Tests for Nonlinear Models. In The Econometrics of Panel Data: A Handbook of the Theory with Applications: Advanced Studies in Theoretical and Applied Econometrics, vol. 33, ed. L. Matyas and P. Sevestre. Dordrecht: Kluwer. Loprest, P. 2002. Families Who Left Welfare: Who Are They and How Are They Doing? Urban Institute Discussion Paper 99 02. Washington, DC: Urban Institute. Mills, R. J. 2001. Health Insurance Coverage: 2000. In Current Population Reports, September, 60 215. Accessed via the Bureau of the Census Web site at www.census.gov/prod/2001pubs/p60 215.pdf. Mishel, L. R., J. Bernstein, and J. Schmitt. 1997. The State of Working America: 19961997. Armonk, NY: M. E. Sharpe. Neumark D., and W. Wascher. 2000. Minimum Wages and Employment: A Case Study of the Fast-Food Industry in New Jersey and Pennsylvania: Comment. American Economic Review 90:1362 1396. OBrien, E., and J. Feder. 1999. Employment-Based Health Insurance Coverage and Its Decline: The Growing Plight of Low-Wage Workers. Washington, DC: Kaiser Commission on Medicaid and the Uninsured. Pauly, M. V., and L. Nichols. 2002. The Nongroup Health Insurance Market: Short on Facts, Long on Opinions and Policy Disputes. Health Affairs Web Exclusive (23 October) at www.healthaffairs.org/WebExclusives/Nongroup_Web_Excl_102302. htm. Pear, R. 1999. Poor Workers Lose Medicaid Coverage Despite Eligibility. New York Times, 12 April, A1. Rostenkowski, D. 1989. Announces Introduction of, and a Public Hearing on, H.R. 1864, a Bill to Simplify the Nondiscrimination Rules Applicable to Employee Benet Plans under Section 89 of the Internal Revenue Code. Committee on Ways and Means Press Release, No. 814, 14 April. Shen, Y.-S., and S. Zuckerman. 2003. Why Is There State Variation in EmployerSponsored Insurance? Health Affairs 22(1):241 250.
926
Sullivan, C., M. Miller, R. Feldman, B. Dowd, and J. Gabel. 1992. Employersponsored Health Insurance in 1991, Witness to a Thousand Stories: A Look at Insurance Data. Health Affairs 11(4):172 191. Summers, L. 1989. Some Simple Economics of Mandated Benets. American Economic Review 79(May):177 183. U.S. Bureau of the Census and U.S. Bureau of Labor Statistics. 1988. Employee Benets Supplements, Current Population Survey. Washington, DC: U.S. Bureau of Labor Statistics. .1993. Employee Benets Supplements, Current Population Survey. Washington, DC: U.S. Bureau of Labor Statistics. U.S. Bureau of Labor Statistics. 2000. Health Insurance Coverage: 1999. Report P60 211. Washington, DC: U.S. Department of Commerce, Economic Statistics Administration. . 2001. Employment Cost Index. Washington, DC: U. S. Bureau of Labor Statistics. U.S. Department of Health and Human Services. 1987. National Medical Expenditures Survey: Household Survey File. Washington, DC: Agency for Health Care Policy and Research.
Copyright of Journal of Health Politics, Policy & Law is the property of Duke University Press and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.

(2003) Amy Wolaver, Timothy McBride, Barbara Wolfe - Mandating Insurance Offers For Low-Wage Workers An Evaluation of Labor Market Effects

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

(2003) Amy Wolaver, Timothy McBride, Barbara Wolfe - Mandating Insurance Offers For Low-Wage Workers An Evaluation of Labor Market Effects

Uploaded by

Copyright:

Available Formats

Mandating Insurance Offers for Low-Wage Workers: An Evaluation of Labor Market Effects

Journal of Health Politics, Policy and Law

Mandating Insurance Offers for Low-Wage Workers

Journal of Health Politics, Policy and Law

Mandating Insurance Offers for Low-Wage Workers

W0 W0 WHI WHI Minimum wage

E*0 = E*HI High-skilled, Full-time Workers

EDHI <ESHI Low-skilled, Full-time Workers

Figure 1 Labor Market Changes with Employer Provision of Health Insurance

Journal of Health Politics, Policy and Law

Mandating Insurance Offers for Low-Wage Workers

Journal of Health Politics, Policy and Law

Mandating Insurance Offers for Low-Wage Workers

Journal of Health Politics, Policy and Law

Mandating Insurance Offers for Low-Wage Workers

Journal of Health Politics, Policy and Law

Variables Means and Standard Deviations (in Parentheses)

.18 .025 .57 .31 .13 .16 .23 24,039

Mandating Insurance Offers for Low-Wage Workers

20 10 0Low-wage, Part-time Low-wage, Full-time High-wage, Part-time High-wage, Full-time Total

Firm Offers Health Insurance

Individual Is Offered Health Insurance

Journal of Health Politics, Policy and Law

Mandating Insurance Offers for Low-Wage Workers

High-Wage Workers Low-Wage Workers (a) (b)

100 or fewer 100 999 1000 or more

.011 (.003) .013 (.003) .013 (.002)

Largest to smallest rms

High-Wage Workers Low-Wage Workers (a) (b)

100 or fewer 100 999 1,000 or more

.0036 (.0018) .0044 (.0018) .0052 (.0016)

Difference in difference: Largest to smallest rms

Journal of Health Politics, Policy and Law

Comparisons of a Subset of High-Wage Workers to Low-Wage Workers

1988 High Wage, Would Be 1993 Low Wage (a)

Journal of Health Politics, Policy and Law

Mandating Insurance Offers for Low-Wage Workers

All part-time 1988 1993 Involuntary part-time 1988 1993

Journal of Health Politics, Policy and Law

Mandating Insurance Offers for Low-Wage Workers

Premium Imputation Effects

Journal of Health Politics, Policy and Law

Mandating Insurance Offers for Low-Wage Workers

Wages and Salaries

Year Quarter Year and Quarter

Journal of Health Politics, Policy and Law

Mandating Insurance Offers for Low-Wage Workers

Journal of Health Politics, Policy and Law

Mandating Insurance Offers for Low-Wage Workers

Journal of Health Politics, Policy and Law

Mandating Insurance Offers for Low-Wage Workers

Journal of Health Politics, Policy and Law

Appendix A: Details of the Nondiscrimination Clause

Mandating Insurance Offers for Low-Wage Workers

Journal of Health Politics, Policy and Law

Mandating Insurance Offers for Low-Wage Workers

Appendix B: Econometric Methods

Journal of Health Politics, Policy and Law

Appendix Table B1 Results

Premium Imputation, Exponential Regression Coefcient Standard Error

Selected Results from Simultaneous Regression Model Individual Offer

Involuntary Part-time, No Offer

Selected Results from Simultaneous Regression Model Individual Offer (Continued)

E0 = EHI High-skilled, Full-time Workers