Where do the sick go? Health insurance and employment in small and large firms.

Kapur, Kanika ; Escarce, Jose J. ; Marquis, M. Susan 等

1. Introduction

The difficulties that small firms face in obtaining and maintaining health insurance for their employees have been widely documented (Brown, Hamilton, and Medoff 1990; McLaughlin 1992; Fronstin and Helman 2000). Only 45% of firms with fewer than 50 employees offer health insurance, compared to 97% of firms with 50 or more employees (Agency for Healthcare Research and Quality 2002). This low proportion has been attributed, in part, to the high administrative cost of health insurance for small firms, the low demand for insurance among workers in these firms, and the unwillingness of insurers to take on small firm risks (McLaughlin 1992; Monheit and Vistnes 1999; Fronstin and Helman 2000).

In recent decades, small firms that provide health insurance to their employees were in a precarious position. Their premiums were calculated yearly, based on the expected value of their health care utilization. Hence, a single high-cost employee could lead to a substantial surcharge on the premiums for the firm (Zellers, McLaughlin, and Frick 1992). In a survey of small employers that did not offer health insurance, 75% claimed that an important reason for not offering insurance was high premium variability (Morrisey, Jensen, and Morlock 1994). Concerns about these problems fueled the passage of numerous state small group health insurance reforms in the 1990s that implemented premium rating reforms and restrictions on pre-existing condition exclusions. While a few states have implemented premium rating reforms that have severely restricted small group insurers' ability to use health status to set premiums, in most states, these reforms have been moderate.

Assuming that firms are unable to perfectly tailor individual wages to individual health insurance costs, unexpectedly high premiums may impose a large burden on small firms. Paying high premiums, possibly financed by borrowing at high interest rates, may increase the risk of bankruptcy. If small firms choose not to pay high premiums, and instead drop insurance coverage, they renege on the implicit compensation contract with workers. Employers may opt to raise employee contributions to cover higher costs, but large increases may lead to healthier employees dropping coverage. Faced with this predicament, small firms may choose to prevent expensive premium variability by maintaining a workforce that has a low-expected utilization of health care services. Thus, the link between employment and health insurance in small firms may result in a welfare loss if it prevents individuals with high-expected health costs from being employed in small firm jobs in which they may have high match-specific productivity.

Employers may obtain information about employees' medical conditions in several ways. Before the passage of the 1990 Americans with Disabilities Act (ADA), half of all employers conducted pre-employment medical examinations (U.S. Congress 1988). Most small group employers required the completion of a family health questionnaire for insurance coverage (Zellers, McLaughlin, and Frick 1992; Cutler 1994). While the ADA now restricts employer inquiries on employee health, it does not apply to firms with under 15 employees. In addition, employer compliance with the ADA may be hindered because its stipulations about pre-employment health inquiries are vague. Medical inquires are allowed if they pertain to the applicant's ability to perform the job. In addition, medical information is explicitly allowed in the use of medical underwriting for insurance (Epstein 1996). The media continues to report cases in which employers easily obtained employee medical records (Rubin 1998) or in which employees have been laid off because of high health costs (O'Connor 1996) or in which employee premiums have been adjusted to reflect the employee's claims experience (Kolata 1992).

The Health Insurance Portability and Accountability Act of 1996 (HIPAA) includes a nondiscrimination provision that bars a group health plan or issuer from discriminating in eligibility or contributions on the basis of a health status-related factor. However, HIPAA allows medical underwriting and allows insurers to rate groups of employees based on health status as long as the premium rate for all employees is blended. This stipulation prevents employers from requiring higher cost employees to contribute a higher premium share, but it does not shield employers from bearing the costs for a sick worker.

Economists have typically believed that health insurance is an attribute of "good jobs" and that workers do not choose jobs based on whether or not the job provides health insurance. In fact, this precept is behind the notion that employment is a mechanism for minimizing adverse selection in the market for health insurance (see, for example, Gruber and Levitt 2000). However, a number of recent studies have suggested that worker demand for health insurance may play an important role in employment decisions. Workers with high-expected family costs may prefer jobs that offer health insurance, and, conversely, workers with low preferences for health insurance may sort into jobs that lack health insurance (Monheit and Vistnes 1999, 2006; Bundorf and Pauly 2004; Royalty and Abraham 2006).

In this paper, we use the Medical Expenditure Panel Survey (MEPS) from 1996 to 2001 to examine the magnitude of employment distortions for workers with high-expected health costs. Since health insurance and employment are linked, health insurance may be an important determinant of employment outcomes. High-expected health costs may reduce the probability that workers are employed in firms in which they have the highest match-specific productivity. We estimate the magnitude of distortions in hiring, employment, and separations. Furthermore, we examine the effect of state small group health insurance reforms that restrict insurers' ability to deny coverage and restrict premium variability on employment distortions in small firms relative to large firms. Estimating the magnitude of employment distortions in insured small firms and understanding the effect of small group regulation on these distortions is essential in deciding optimal public policy toward the small group health insurance market.

2. Literature Review

The first literature that is relevant to this paper relates to small firms and health insurance. Cutler (1994) finds evidence that small firms are subject to a higher degree of premium variability than are large firms. Moreover, small firms with young workers, high turnover, or low wages tend to have the highest premium variability. The possibility of employment screening as a result of the incentives created by the small group health insurance market has been previously noted in the literature (Aaron and Bosworth 1994; Madrian 1994). Monheit and Vistnes (1994) find that the risk selection practices of insurers segment the small-group market so that only persons who are favorable health risks obtain employment-related insurance. They find that the employees and dependents with coverage from small firm policies are in better health than those with non-group policies (when firm coverage was not available) or those who had no coverage. While these results may indicate the presence of employment distortions due to health insurance, it is also possible that we may see these results if individuals in good jobs that offer health insurance are in better health than those who are not offered health insurance. Olson (1993) finds that individuals who say that they are in bad health are far less likely to have health insurance in industries that have a high proportion of small firms than in industries that have a high proportion of large firms. Using the 1987 National Medical Expenditure Survey (NMES) data, Kapur (2004) finds evidence of employment distortions in small firms that is consistent with underwriting rules in the small group health insurance market. Extension of this analysis to the 1996 MEPS is limited by the relatively small sample sizes of insured workers with adverse health conditions. Although not focused on small firms, Buchmueller (1995) finds that men in worse health are less likely to be insured.

Another relevant literature examines the impact of health insurance costs on wages and employment. There is evidence to indicate that rising health insurance costs have led firms to increase hours worked by employees rather than employing more workers (Cutler and Madrian 1998). Other work shows that for certain groups, the wages and the probability of being hired are sensitive to health insurance costs (Gruber 1995; Scott, Berger, and Garen 1995; Sheiner 1995). However, several recent papers fail to find robust evidence of the expected relationship between wages and health insurance (Jensen and Morrisey 2001; Levy and Feldman 2001; Simon 2001).

Using the 1987 NMES data and the 2000 MEPS data, Monheit and Vistnes (1999, 2006) provide evidence that worker preferences play a role in employer-provided health insurance, showing that workers with low preferences for health insurance sort into firms that do not offer health insurance. Royalty and Abraham (2006) demonstrate that workers with access to spouse health insurance sort into jobs that do not offer health insurance, again indicating that worker demand for health insurance may play an important role in job choice. Bundorf and Pauly (2004) also find evidence that individuals who have high-expected health costs are more likely to obtain health insurance in the group market and in the individual health insurance market.

Research on the impact of state small group health insurance reform generally has shown a small effect or no effect on small firms' propensity to offer health insurance or on employees' insurance coverage (Sloan and Conover 1998; Hall 1999; Jensen and Morrisey 1999; Zuckerman and Rajan 1999; Buchmueller and DiNardo 2002; Marquis and Long 2002; Monheit and Schone 2004). However, a few studies do find modest effects of the reforms on insurance (Uccello 1996; Buchmueller and Jensen 1997; Hing and Jensen 1999; Simon 2005). In addition, some work has demonstrated that stronger reforms increased insurance coverage for high-risk workers relative to low-risk workers (Monheit and Schone 2004; Davidoff, Blumberg, and Nichols 2005). Most of these studies exploit cross-sectional and time-series variation in the implementation of state reforms to identify the effect of the reform on insurance coverage and do not focus on analyzing employment and employment flows in small and large firms as a result of the reforms. Two existing studies examine the labor market effects of small group health insurance reform and find small or no effects; however, neither of these studies included access to detailed family health data (Kaestner and Simon 2002; Kapur 2003).

3. Theoretical Motivation

Employment Distortions

There are several reasons for which the health insurance market may distort employment decisions in small firms. If providing health insurance for workers with high-expected health costs is more expensive for small firms, they may screen out high-cost workers. Furthermore, small firms may be in a better position to screen out high-cost employees compared to larger firms. On the other hand, workers with high-expected health costs are expected to prefer jobs that offer health insurance. The existence of employment distortions is predicated on the assumption that firms are unable to perfectly tailor individual compensation to health insurance costs. Minimum wage laws, HIPAA non-discrimination provisions, and notions of fairness and equity are likely to prevent adjustment of wages at the individual level. We consider several theoretical reasons why small firms may choose to screen out workers with high-expected health costs. We also consider the rationale behind sorting based on worker demand.

First, the health insurance market may lead to employment distortions in small firms by creating premium variability. If small firms hire randomly from an available pool of workers, they will face a more variable health insurance premium bill. Therefore, small firms are likely to be more sensitive to the presence of a high-cost worker, because they are less able to diversify health insurance premium costs internally than are larger firms. (1) Assuming that firms are unable to fully pass on the premium bill to workers, at least in the short run, unexpectedly high premiums may be expensive for firms, since they may have to borrow to finance the high premiums. Employers may opt to raise employee contributions to cover higher costs, but large increases may lead to healthier employees dropping coverage. To avoid the cost of high health insurance premiums, small firms may choose to screen out workers who are likely to cause their group health insurance premiums to be high.

Second, it is possible that the administrative cost of charging each employee his or her health insurance cost may be large. As a result, firms may elect to charge each employee the average health insurance cost of the employee pool. If the search costs of finding a new job are higher than the extra share of health insurance costs, healthy workers will not quit. This pooling strategy implies that large firms can spread the high health insurance costs of one worker over their entire employee pool. However, small firms are unlikely to be able to successfully employ this strategy, since health insurance costs are spread over a much smaller number of workers. As a result, small firms may be compelled to charge each employee his or her own health insurance costs or to screen out high-cost employees.

Third, small firms are likely to have higher health insurance costs than large firms. Specifically, administrative costs decrease dramatically with firm size. Administrative costs are about 40% of claims paid in very small firms (under five employees), while they are only 5.5% of claims paid in the largest firms with 10,000 or more employees (Helms, Gauthier, and Campion 1992). A worker with high health costs may be more expensive to insure in a small firm because of the fact that the administrative costs of processing claims is relatively larger for small firms than it is for large firms. (2)

Fourth, small firms may have a lower cost of screening out sick employees. It is likely that employment decisions are centralized in the hands of a single entrepreneur in a small firm. In contrast, employment decisions are more likely to be decentralized in large firms. The single entrepreneur in a small firm may find it easier to screen out an individual with high-expected health costs. (3) Moreover, most small firms are required to provide data on the family health status of potential employees to their health insurance companies (U.S. Congress 1988). This implies that the information required for employment screening is readily available to small firms, and that, coupled with centralized decision making, results in individuals with high-expected health costs being more likely to be screened out of small firms than large firms.

Worker demand for health insurance may also affect employment outcomes. Workers with high-expected health costs value jobs that provide health insurance more than other workers. Bhattacharya and Vogt (2006) have developed a model for worker sorting due to health insurance demand. In this model, workers with poor health prefer jobs that offer health insurance; however, job switching costs may prevent some from moving to insured jobs. In addition to simply preferring insured jobs, workers with high-expected health costs may prefer insured large firm jobs to small firm jobs. The greater stability of large firm jobs and the increased plan choice in large firms may play a role in worker sorting. In addition, even though health insurance policy quality is similar across small and large firms along many important dimensions, small firm policies tend to have higher deductibles, and this feature may discourage workers with high-expected expenses from seeking insured small firm jobs. (4) Health insurance policy quality may also be viewed as a tool for small firms to discourage workers with high-expected health costs from seeking employment and may therefore act as an implicit employer screening mechanism.

State Small Group Health Insurance Reform

During the 1990s, most states implemented small group health insurance reform. These reforms tended to include the following components: guaranteed issue/renewal laws that mandated that insurance companies issue/renew some or all health insurance policies for small firms; portability and pre-existing condition limitation reforms that limited the time that insurers could exclude coverage for certain medical conditions; and premium rating reforms that restricted the factors that could be used to set health insurance premiums and restricted premium variability. Detailed descriptions of these reforms are in found in Kapur (2003). In 1996, the federal HIPAA mandated guaranteed issue of health insurance for small firms starting in 1997. For many states, the existing state small group reform was equivalent to HIPAA's provisions; however, for others, HIPAA's mandates changed the guarantee issue requirements.

Theoretically, the effect of state small group health insurance reforms and HIPAA on employment outcomes for workers with high-expected health costs is ambiguous. Firm screening and worker demand may exert opposing effects on employment outcomes. Preexisting condition limitations may increase insured small firms' screening of individuals with adverse family health conditions, since these health conditions must be covered by health insurance. On the other hand, since pre-existing condition limitation reforms lead to more extensive small firm health insurance coverage, these reforms may increase worker demand for small firm health insurance. However, ultimately, the worker response depends on workers' valuation of the additional health benefit relative to the cost in terms of reduced wages and/or higher health insurance premiums. Rating reforms that regulate premium variability and use of health factors in setting premiums can be expected to reduce the higher premiums associated with adverse health. Therefore, these reforms should increase insured small firm employment of workers with high-expected health costs. However, if low-risk workers and/or firms leave the market in response to the higher premiums, average market premiums may increase and discourage some high-risk workers from seeking insured small firm employment. Guaranteed issue reforms are important in their interaction with rating reforms, since they mandate that insurers must continue to offer health insurance to small firms even at the newly regulated premiums. Without guaranteed issue reform, rating reforms are toothless, since insurers can cherry-pick low-cost firms. Reforms were passed in packages in all states--the theoretical effect of these reform packages on the employment distortions for individuals with adverse health is ambiguous.

4. Data

We use the Household Component (HC) of the 1996-2001 MEPS conducted by the Agency for Healthcare Research and Quality (AHRQ). The MEPS has an overlapping panel design, in which the sample selected in any given year is followed for two calendar years (Cohen 2000). Each family in the HC participated in five rounds of data collection over a two-year period. During each round, information on all family members' employment (including size of employer) and health insurance coverage (including whether employer-provided health insurance was offered) was collected. In addition, a detailed set of variables on health, demographics, and health utilization was collected.

We also use a state small group reform database constructed using the information in Marquis and Long (2002), Kaput (2003), and Simon (2005). Our primary reform measure is a three-category variable: no/weak reform, moderate reform, or full reform. States with guaranteed issue reform and rating reform that restricted premium variability using rate bands were classified as having moderate reform. States with guaranteed issue and rating reform that included community rating or modified community rating were coded as having full reform. All other states were classified as having no reform or weak reform. We experimented with a host of other specifications for the reforms. We coded individual components of the reforms, such as guaranteed issue reform, rating, and portability/pre-existing condition exclusion reforms (Marquis and Long 2002; Kapur 2003). We also used a measure of premium variability allowed by the rating reforms. We developed several alternative measures of the package of reforms based on the extent of allowable premium variability, guaranteed issue, and renewal reforms. After the implementation of HIPAA, all states were coded as having guaranteed issue reforms.

The public-use MEPS data do not include an identifier for state of residence; therefore, we conducted the state reform analysis at the AHRQ Data Center in Rockville, Maryland, where we had access to a database that merged our reform variables to the MEPS-HC analysis files.

5. Econometric Framework: Employment Distortions

We estimate the magnitude of three types of employment distortions: (i) Hiring Distortions: Are small firms that offer health insurance less likely to hire workers with sick families?; (ii) Employment Stock Distortions: Are small firms that offer health insurance less likely to employ workers with sick families?; and (iii) Separation Distortions: Are small firms more likely to lay off insured workers with sick families?

To estimate the magnitude of the employment distortion, employees in small firms with health insurance should be compared to employees who have similar observable job and demographic characteristics but who work in larger firms. The means presented in Table 1 show the characteristics of employees in four groups--those employed in small firms that offer health insurance, large firms that offer health insurance, small firms that do not offer health insurance, and large firms that do not offer health insurance. For most demographic and job characteristics, such as age, sex, marital status, education, and wage, employees in small firms with health insurance are more similar to employees in large firms with health insurance than to employees without health insurance. Therefore, the preferred comparison group for insured small firm workers is insured large firm workers. However, there are several statistically significant differences in job and demographic attributes between insured small firm workers and insured large firm workers. The model estimation controls for these differences.

Hiring Distortions

To test whether hiring distortions exist, we estimate a multinomial logit model with a fourlevel dependent variable, Yi, with different values for each of the following employment outcomes: small firm worker and offered health insurance, large firm worker and offered health insurance, small firm worker and not offered health insurance, and large firm worker and not offered health insurance: (5)

Pr ([Y.sub.i] =j) = f([[alpha].sub.j] + [[beta].sub.j] * [HC.sub.i] + [Z.sub.i] '[[gamma].sub.j]).

HC denotes the expected health costs. The sample consists of all workers who had tenure of less than one year at the interview date. (6) We construct three measures of expected health costs. These include a count of medical conditions in the family, an indicator for the presence of any family medical conditions, and an index of expected health costs (log transformed) based on family medical conditions. We focused on medical conditions that were chronic, persistent measures of health, since temporary shocks to health are likely to have a much smaller effect on employment outcomes. (7) The matrix Z consists of control variables. Demographic controls include schooling, sex, age, age squared, marital status, race, family size, and spouse work status. We also included job controls--industry indicators, occupation indicators, wage and union status, and region, Metropolitan Statistical Area, and year indicators. Since job controls are possibly endogenous to the employment outcome, we also ran these models excluding the job controls and found very similar results. (8)

We estimated the model on three samples--the full sample, single workers, and married workers. Models for married workers included an additional set of control variables that captured the characteristics of the worker's spouse. These included spouse education and spouse employment characteristics (wage, union, industry, and occupation), if employed. Married workers may have a muted worker sorting response, since their sorting behavior is affected by their own characteristics and by spouse characteristics. Furthermore, the health effects for married workers hinge on both the worker's own health and the health of his/her dependents, whereas for single workers, the health effects depend solely on the worker's own health. Therefore, separate models for single and married workers may be informative. For married workers, we also estimated an additional model that disaggregated family health into worker and dependent health. Poor worker health may result in a stronger employment distortion than poor dependent health, since the former is likely to be more readily observable.

We constructed the index of expected health costs by estimating separate models for adult and child medical expenditures on the subsample of individuals with private health insurance using a one-part generalized least-squares model with a gamma distribution and a square-root link. (9) This model provided the best fit for health costs (Manning and Mullahy 2001). The health cost models included a full set of medical condition indicators and controls for demographic variables. Coefficients on medical conditions from these models provided the weights used to construct the predicted cost index. (10) A concern with this model is that health insurance could be endogenously determined. Given that we have no suitable instrumental variables in our data set, we are unable to adjust for selection formally. However, we use multiple measures of health, and only one of them is subject to an endogeneity concern. Comparing results across the various health measures provides a natural robustness check.

The multinomial logit model treats both small firm employment and health insurance as endogenous variables that are jointly incorporated into the four-level dependent variable. A Hausman test shows that assumption of the Independence of Irrelevant Alternatives cannot be rejected in this application. We correct the standard errors for clustering within family.

We use the estimates from the multinomial logit model to compare the effect of HC on insured small firm new hires to its effect on insured large firm new hires by reporting relative risk ratios (RRR) for insured small firm employment, with insured large firm employment as the base category. As discussed earlier, this strategy enables us to compare groups that are comparable in demographic and job characteristics. The RRR based on 13, the coefficient of interest, is the effect of the expected health cost of a worker on the probability that he or she is employed in an insured small firm relative to an insured large firm. We would expect this RRR to be less than 1 and significant if individuals with high-expected health costs are less likely to be employed at small firms with health insurance.

Even if employees in small firms with health insurance are found to have lower expected health costs than those in large firms with health insurance, this result could be attributed to all small firms being less likely to employ sick workers. For example, if small firms have only a single worker experienced in a certain task, absenteeism due to sickness could be a bigger problem for a small firm, and thus these firms may prefer to employ healthier workers, regardless of health insurance. Alternatively, a worker demand story may indicate that workers with high-expected health costs prefer insured large firm jobs to insured small firm jobs because they are more stable and have greater health plan choice. Therefore, we compare the effect of health on the probability of being in a small firm with health insurance with the effect of health on the probability of being in a small firm without health insurance using estimates from the multinomial logit model.

We also use the estimates from the multinomial logit model to demonstrate the effect of health on the probability of being employed in a large firm that offers health insurance compared to the probability of being hired in a large firm that does not offer health insurance. These results can be compared to those that contrast small firms with health insurance to small firms without health insurance to show that effects of health on being hired into a small firm with health insurance are not generalizable to large firms.

Table 2 presents the results from the multinomial logit models. The models vary in their definition of firm size (less than 25 employees vs. less than 50 employees), in their family health measures (number of family conditions, whether there are any family conditions, and predicted medical expenditures), and in their estimation sample (full sample, single workers, and married workers). The first two columns of Table 2 contain the results for insured small firm workers relative to the base category of insured large firm workers. The third and fourth columns contain results for insured small firm workers relative to uninsured small firm workers, and the last two columns contain results for insured large firm workers relative to uninsured large firm workers. Significance tests for RRRs test the difference of the RRR from 1.11

For the full sample, the results in columns 1 and 2 of Table 2 show that workers with adverse family health are significantly less likely to be employed in insured small firms relative to insured large firms. This result is statistically significant for five of the six full sample models in columns 1 and 2. For the indicator for "any conditions" in the model in which small firms are defined as those that employ less than 25 people, the relative risk of being an insured small firm worker relative to an insured large firm worker is 86% as a result of having a family medical condition. The RRR in columns 3 and 4 show that workers with sick families are less likely to be new hires in small firms that offer health insurance relative to small firms that do not offer health insurance. This comparison shows that hiring workers with healthy families is not simply a characteristic of all small firms. For the full sample, columns 5 and 6 show that there is no effect of family health on whether workers are new hires in large firms that offer health insurance. However, for the married sample, there is some evidence that families with poor dependent health are more likely to be insured new hires in large firms compared to uninsured new hires in large firms. In summary, the results from Table 2 show that workers with adverse family health are less likely to be hired into small firms that offer health insurance compared to any other type of firm. The full results from the multinomial logit models reported in the paper are available on request. (12)

We cannot reject the hypothesis that the effect of health is the same for the single and married samples, and in several cases, the results become more imprecise after disaggregating the sample. (13) However, the results appear stronger for single individuals who are not likely to have access to alternative sources of employment-based coverage. For small firms containing under 50 employees, disaggregating the health measures into worker and family components shows that worker health appears to have a statistically significant effect, whereas dependent health tends to have a statistically insignificant effect, consistent with the notion that worker health is more readily observable to employers. However, in most cases, we are unable to reject the possibility that worker health has a statistically different effect from dependent health.

Stock Distortions

We estimate multinomial logit models to test employment distortions in the stock of employees, following the same structure as the models for new hires. (14) For the full sample, the results in Table 3 show that workers with unhealthy families are less likely to be employed in small firms that offer health insurance compared to large firms that offer health insurance. For the full sample, this result is statistically significant for five of the six models presented in columns 1 and 2 (Table 3). For the indicator for "any conditions" in the model in which small firms are defined as those that employ less than 25 people, the relative risk of being an insured small firm worker relative to an insured large firm worker is 91% as a result of a having a family medical condition.

Even though the point estimates for the full sample indicate that workers with sick families are less likely to be employed in small firms that offer health insurance compared to small firms that do not, these estimates are not statistically significant. For the full sample, the results that compare workers in large firms that offer health insurance to workers in large firms that do not offer health insurance show no consistent pattern on the employment of workers with sick families. We expect that the largest effect of health would occur at the time of hiring; therefore, it is not surprising that the effects for all employees are slightly weaker than those for new hires.

As with the new hire analysis, the results in columns 1 and 2 of Table 3 appear stronger for single individuals than for married individuals, although in most cases the effects are not statistically different. In columns 3 and 4, sicker single workers appear to be significantly less likely to be employed in insured small firms compared to uninsured small firms (for the under50 employees definition). In several models that contrast insured small firms with uninsured small firms and insured large firms with uninsured large firms, we find that poor dependent health increases the likelihood of being in an insured firm, consistent with the notion that families with poor health have a higher demand for insured jobs.

Separation Distortion

Ex ante, we do not expect separation distortions to be large, since separations are costly in terms of lost on-the-job training and unemployment insurance taxes. Firms should prefer to screen at the time of the hiring decision, and workers should prefer to make decisions on their preferred jobs at the time of hiring. However, given that expected health costs at the time of hiring could be lower than the expected costs at a later date as a result of imperfect information about future health costs, separation distortions could be present.

To test the importance of layoff and quit distortions, we determine if workers with high family health costs employed in small firms with health insurance are more likely to be laid off or more likely to quit than workers in small firms without health insurance and workers in large firms. Unlike the hiring distortion estimation, here the desired insurance variable is whether or not a worker holds employer-provided health insurance. It is not an indicator of whether a worker was offered health insurance, since only individuals who hold health insurance contribute to employers' health insurance costs.

We estimate a multinomial logit model in which Y, the job transition, can take any of three values denoted by j--stay, lay off, or quit. (15) The sample consists of individuals employed at any time during the sample period.

Pr(Y = j) = f([[alpha].sub.j] + [[beta].sub.1j] * HC + [[Beta].sub.2j] * Small Firm + [[beta].sub.3j] * HI + [[gamma].sub.1j] * HC * Small Firm

+ [[gamma].sub.2j] * HC * HI + [[gamma].sub.3j] * HI * Small Firm + [[delta].sub.j] * HC * HI * Small Firm + Z'[[eta].sub.j]),

where Z is the vector of controls, HC is a vector of expected health costs, and HI denotes employer-provided health insurance. If small firms screen high-cost workers, the coefficient on the interaction of expected health cost, small firm and health insurance, [delta]j, would be positive when j denotes a layoff, assuming that wages are relatively inflexible. Alternatively, if wages are flexible, we expect that [[beta].sub.3j] would depend on worker valuation of health insurance relative to wages when j denotes a quit. (16)

Since insured small firm workers are most similar to insured large firm workers, we determine the sensitivity of our results to this specification by reestimating this model on only insured workers and by estimating the effect of interest with an interaction term between small firm employment and adverse health. We also reestimate the model only on small firm workers, and we examine the interaction between adverse health and health insurance to determine if sickness has a differential effect on insured small firm workers compared to uninsured small firm workers.

Table 4 presents the results from the multinomial logit models for separations. We present results for the full sample, single workers, and married workers. Our results for the sample of all workers (insured and uninsured), in the first set of columns, show that workers with families with adverse health who are employed in small firms with health insurance are less likely to quit their jobs and are also less likely to be laid off. The three rows following the full sample RRR show predictions for the marginal effects at the 25th, 50th, and 75th percentiles of the distribution. (17) The marginal effects show small and negative effects on separations. For the insured subsample, these results are only statistically significant when small firms are defined as employing less than 50 workers. However, results from the small firm sample show no statistically different effects of sickness for insured small firm workers compared to uninsured small firm workers.

The results are somewhat mixed across the various samples. The results from the small firm sample that show no effect of health on separations appear to be more plausible than the results that show that insured small firm workers are less likely to separate than workers in insured large firms. If small firms were screening out workers with high health costs, we would expect these workers to be more likely to be laid off. If, on the other hand, the separation distortions were due to worker demand for health insurance, we would not expect workers in small firms with health insurance to be less likely to separate from their jobs than workers in large firms with health insurance. Some models show that poor worker health is associated with the expected increase in separation in insured small firms. However, given the mixed pattern of results and their small magnitude in marginal effects, our findings regarding separation distortions remain ambiguous.

Supply Side versus Demand Side

An intriguing question is whether employment distortions are driven by firms screening out sick workers (a demand side effect) or by sick workers choosing not to work in small firms (a supply side effect). In a market-clearing model with fully adjusting wages, it is impossible to distinguish between a demand and a supply side effect. However, if we assume that wages are somewhat rigid, then the distinction between demand and supply side effects becomes meaningful. Determining whether employment distortions are driven by demand or supply side considerations is important for designing economic policy. For instance, strengthening anti-discrimination legislation would reduce distortions if firm screening were a main factor, whereas improving small firm health insurance quality would be more effective if worker sorting was responsible for employment distortions.

Our results have not provided any definitive evidence of either a firm-side or worker-side story; most likely both play some role. We have found that workers with adverse family health are less likely to be hired into small firms that offer health insurance compared to any other type of firm. If worker demand were responsible for our results, we would expect to see workers with high-expected costs flock to jobs that offer health insurance in large and small firms. However, we find that workers with adverse health are no more likely to be new hires in large firm jobs that offer health insurance than in large firm jobs with no health insurance. They are also less likely to be in small firm insured jobs than in small firm uninsured jobs. This evidence indicates that firm screening plays a role in employment distortions. On the other hand, we do find some evidence for a worker demand story from the results on dependent health--we find that workers with dependents in poor health are more likely to be employed in insured small firms compared to uninsured small firms, and they are more likely to be employed in insured large firms compared to uninsured large firms. However, the results from the separation models did not shed any light on the role of firm-side versus worker-side effects. Most likely, both firm and worker effects play a role in determining the distribution of employment between small and large firms.

We also empirically check if wages for insured and sick small firm employees are lower than wages for insured and sick large firm employees. If we assume that sick workers employed in insured small firms are no more productive than sick workers employed in insured large firms, then similar wages by firm size would again indicate that small firms would have an incentive not to employ workers with high-expected health costs. We find no difference in the wages between insured workers with high-expected health costs in large and small firms.

Effect of State Reform on Employment Distortions

We reestimated the multinomial logit models for new hires, the stock of workers, and for separations, after including measures of state small group health insurance reform and interactions of the reforms with the family health measures. We find no evidence that reforms changed employment patterns. The results of the interaction between reform and health are statistically insignificant for all models. Results from the multinomial logit models described in this section are available on request.

We also found small and statistically insignificant effects for state reforms on each type of employment distortion using alternative measures of state reforms, such as individual indicators for portability, type of rating reform, guaranteed issue reforms; for alternative definitions of the package of reforms; and for measures of the allowable premium variability.

6. Conclusion

This paper determines if the link between employment and health insurance leads to distortions in small firm employment. The estimation results show that workers with high-expected health costs are less likely to be new hires in small firms that offer health insurance than any other type of firm. These workers are also less likely to be employed in small firms that offer health insurance. The magnitude of these findings indicates that the employment effects are relatively small--workers with high-expected health costs are 86% as likely to be new hires in insured small firms and 91% as likely to be employed in insured small firms relative to insured large firms.

Our results indicate that both firm screening and worker sorting play a role in the observed employment outcomes. Small firms may be using explicit and implicit mechanisms to screen out workers with high-expected health costs. Workers with high-expected health costs are likely to prefer jobs with high-quality health insurance policies. But our results can be only partially explained by worker sorting. We find that workers with high-expected health costs are less likely to be new hires in small firms that offer health insurance than in small firms that do not offer health insurance, contrary to the predictions of a worker sorting framework. However, we also find some evidence that workers with sick dependents are more likely to be employed in insured small firm jobs than in uninsured small firm jobs, and they are more likely to be employed in insured large firm jobs than uninsured large firm jobs. Most likely, a combination of firm screening and worker sorting factors are responsible for the employment outcomes that we observe.

A pertinent question in evaluating our results is whether firms can legally screen workers with high-expected health costs from employment. The ADA prohibits employment screening for workers with qualifying disabilities; however, the ADA does not apply to firms with fewer than 15 employees, leaving these firms open to employment screening. Furthermore, we observe the strongest evidence for firm screening during the hiring process, when screening may be hard to detect or prove.

State small group health insurance reforms coupled with federal HIPAA regulations appear to have had little effect on the pattern of employment distortions for workers with high-expected health costs. Employment distortions appear to have persisted in small firms, despite the wide implementation of these insurance reforms. Premium rating regulations, an important component of the state small group reforms, were quite weak in most states and could explain the muted effects we find of reforms on small firm employment patterns.

Received December 2005; accepted November 2006.

References

Aaron, Henry, and Barry Bosworth. 1994. Economic issues in reform of health care financing. Brookings Papers on Economic Activity Microeconomics 1994:249-9.

Agency for Healthcare Research and Quality (AHRQ), Center for Cost and Financing Studies. 2002. 2002 Medical expenditure panel survey--Insurance component. Rockville, MD: AHRQ.

Ai, Chunrong, and Edward Norton. 2003. Interaction terms in logit and probit models. Economics Letters 80:123-9.

Bhattacharya, Jayanta, and William Vogt. 2006. Employment and adverse selection in health insurance. NBER Working Paper No. 12430.

Brown, Charles, James Hamilton, and James Medoff. 1990. Employers large and small. Cambridge, MA: Harvard University Press.

Buchmueller, Thomas. 1995. Health risk and access to employer-provided health insurance. Inquiry 32:75-86.

Buchmueller, Thomas, and John DiNardo. 2002. Did community rating induce an adverse selection death spiral? Evidence from New York, Pennsylvania, and Connecticut. American Economic Review 92:280-94.

Buchmueller, Thomas, and Gail Jensen. 1997. Small group reform in a competitive managed care market: The case of California, 1993 to 1995. Inquiry 34:249-63.

Bundorf M. Kate, and Mark Pauly. 2004. The uninsured: Risk, income and affordability of coverage. ERIU Working Paper, University of Michigan.

Cohen, Steven. 2000. Sample design of the 1997 medical expenditure panel survey household component. MEPS Methodology Report 11, Rockville, MD.

Cutler, David. 1994. Market failure in small group health insurance. NBER Working Paper No. 4879.

Cutler, David, and Brigitte Madrian. 1998. Labor market responses to rising health insurance costs: Evidence on hours worked. RAND Journal of Economics 29:509-30.

Davidoff, Amy, Linda Blumberg, and Len Nichols. 2005. State health insurance market reforms and access to insurance for high-risk employees. Journal of Health Economics 24:725-50.

Epstein, Richard. 1996. Antidiscrimination in health care. Oakland, CA: The Independent Institute.

Fronstin, Paul, and Ruth Helman. 2000. Small employers and health benefits: Findings from the 2000 small employer health benefits survey. EBRI Issue Brief No. 226, Washington, DC.

Gilleskie, Donna, and Byron Lutz. 2002. The impact of employer-provided health insurance on transitions. Journal of Human Resources 37:129-62.

Gruber, Jonathan. 1995. The incidence of mandated maternity benefits. American Economic Review 84:622-41.

Gruber, Jonathan, and Larry Levitt. 2000. Tax subsidies for health insurance: Costs and benefits. Health Affairs 19:72-85.

Hall, Mark. 1999. Health insurance market reform study. Wake Forest University, Winston-Salem, NC. Available http:// www.phs.wfubmc.edu/insure/.

Helms, David, Anne Gauthier, and Daniel Campion. 1992. Mending the flaws in the small-group market. Health Affairs 11:7-27.

Hing, Esther, and Gail Jensen. 1999. Health Insurance Portability and Accountability Act of 1996: Lessons from the states. Medical Care 37:692-705.

Jensen, Gail, and Michael Morrisey. 1999. Small group reform and insurance provision by small firms, 1989-1995. Inquiry 36:176-87.

Jensen, Gail, and Michael Morrisey. 2001. Endogenous fringe benefits, compensating wage differentials and older workers. International Journal of Health Care Finance and Economics 1:203-26.

Kaestner, Robert, and Kosali I. Simon. 2002. Labor market consequences of state health insurance regulation. Industrial and Labor Relations Review 56:136-56.

Kaiser Family Foundation-Health Research and Educational Trust (KFF-HRET). 2004. Employer health benefits: 2004 annual survey. Menlo Park, CA: KFF.

Kapur, Kanika. 1998. The impact of health on job mobility: A measure of job lock. Industrial and Labor Relations Review 51:282-97.

Kapur, Kanika. 2003. Labor market implications of state small group health insurance reform. Public Finance Review 31:571-600.

Kapur, Kanika. 2004. The impact of the health insurance market on small firm employment. The Journal of Risk and Insurance 71:63-90.

Kolata, Gina. 1992. New insurance practice: Dividing sick from well. New York Times. 4 March, p. A1.

Levy, Helen, and Roger Feldman. 2001. Does the incidence of group health insurance fall on individual workers? International Journal of Health Care Finance and Economics 1:227-48.

Madrian, Brigitte. 1994. Employment-based health insurance and job mobility: Is there evidence of job-lock? Quarterly Journal of Economics 109:27-54.

Manning, William, and John Mullahy. 2001. Estimating log models: To transform or not to transform? Journal of Health Economics 20:461-94.

Marquis, M. Susan, and Stephen Long. 2002. Effects of "second generation" small group health insurance market reforms, 1993 to 1997. Inquiry 38:365-80.

McLaughlin, Catherine. 1992. The dilemma of affordability: Private health insurance for small business. In American health policy: Critical issues for reform, edited by Robert Helms. Washington, DC: American Enterprise Institute, pp. 133-51.

Monheit, Alan, and Barbara Schone. 2004. How has small group market reform affected employee health insurance health insurance coverage? Journal of Public Economics 88:237-54.

Monheit, Alan, and Jessica Vistnes. 1994. Implicit pooling of workers from large and small firms. Health Affairs 13:301-14.

Monheit, Alan, and Jessica Vistnes. 1999. Health insurance availability at the workplace. Journal of Human Resources 34:770-85.

Monheit, Alan, and Jessica Vistnes. 2006. Health insurance enrollment decisions: Preferences for coverage, working sorting, and insurance take up. NBER Working Paper No. 12429.

Morrisey, Michael, Gail Jensen, and Robert Morlock. 1994. Small employers and the health insurance market. Health Affairs 13:149-61.

O'Connor, Matt. 1996. Firm fired worker for having HIV, suit claims. Chicago Tribune. 11 July, p. 1, Metro Chicago section.

Olson, Craig. 1993. Health insurance and adverse selection in the labor market. Unpublished paper, Industrial Relations Research Institute, University of Wisconsin-Madison.

Royalty, Anne Beeson, and Jean Abraham. 2006. Health insurance and labor market outcomes: Joint decision making within households. Journal of Public Economics 90:1561-77.

Rubin, Alissa. 1998. Records no longer for doctors' eyes only. Los Angeles Times. 1 September, p. AI.

Scott, Frank, Mark Berger, and John Garen. 1995. Do health insurance and pension costs reduce the job opportunities of older workers? Industrial and Labor Relations Review 48:775-90.

Sheiner, Louise. 1995. Health care costs, wages and aging: Assessing the impact of community rating. Unpublished paper, Federal Reserve Board of Governors.

Simon, Kosali. 2001. Displaced workers and employer-provided health insurance: Evidence of a wage/fringe benefit tradeoff? International Journal of Health Care Finance and Economics 1:249-72.

Simon, Kosali. 2005. Adverse selection in health insurance markets? Evidence from state small group health insurance reforms. Journal of Public Economics 89:1865-77.

Sloan, Frank, and Christopher Conover. 1998. Effects of state reforms on health insurance coverage of adults. Inquiry 35:280-93.

Uccello, Cori. 1996. Firms' health insurance decisions: The relative effects of firm characteristics and state insurance regulations. Washington, DC: The Urban Institute.

U.S. Congress, Office of Technological Assessment. 1988. Medical testing and health insurance. Washington, DC: U.S. Congress.

Zellers, Wendy, Catherine McLaughlin, and Kevin Frick. 1992. Small business health insurance: Only the healthy need apply. Health Affairs 11:174-80.

Zuckerman, Stephen, and Shruti Rajan. 1999. An alternative approach to measuring the effects of insurance market reforms. Inquiry 36:44-56.

(1) Small firms should have a greater incentive to insure against inter-temporal premium changes; however, the market for inter-temporal insurance is inadequate (Cutler 1994). Since the 1990s, small group health insurance reforms may have improved the availability of inter-temporal insurance.

(2) The lower administrative costs in large firms may be due to the fact that large firms tend to have a benefits manager to coordinate health claims and complete paperwork and because there are economies of scale in coordination. The benefits office in large firms acts as an intermediary between employees and insurers, reducing administrative burden for large firm insurers. Large firms are also less likely to drop insurance, resulting in lower transition costs for insurance companies.

(3) It is also possible that small firms may have a relative advantage in wage adjustment, compared to large firms. However, there are several reasons that a small firm employer may find it as difficult as a large firm to adjust wages. First, both small and large firms are subject to minimum wage laws that impede their ability to adjust wages beyond a point. Given that health expenditures can be very high, this may be a binding constraint even for higher wage workers. Second, jobs are often advertised with wage ranges, making it problematic to adjust wages after a position is advertised. Third, different wages for similar jobs within a firm may result in employee dissatisfaction and, so, may be less preferred as an adjustment strategy.

(4) The Kaiser-HRET employer surveys on health insurance benefits have showed that there are no statistically significant differences among small firm plans and large firm plans in their offerings of prescription drugs, adult physicals, outpatient mental services, inpatient mental services, annual OB/GYN visits, oral contraceptives, and well-baby care. Only the propensity to offer prenatal care and chiropractic care differed significantly. Small firm policies were more likely to have no policy limit (60% in small firms and 45% in large firms) and more likely to have a limit on out-of pocket spending than were large firm policies (87% in small firms and 77% in large firms), and they were more likely to have higher deductibles ($559 in large firms and $280 in small firms for single coverage) (Kaiser Family Foundation Health Research and Educational Trust 2004).

(5) The health insurance offer question asks if the person was offered health insurance through the employer. The employment size question asks the number of employees at the person's establishment of employment. Since health insurance decisions are likely to be made on the basis of firm size rather than establishment size, we also used a question that asked if the firm had multiple locations to restrict the sample to single location firms in which the categorization of small versus large employer based on firm size is identical to the categorization based on establishment size. The results were similar to those including all establishments.

(6) Our models of hiring distortions would not capture the form of employment distortion in which an individual is potentially unemployed as a result of higher-expected health costs.

(7) We used the following adult medical conditions: angina, myocardial infarction, cancer, diabetes, arthritis, stroke, chronic obstructive pulmonary disease, asthma, obesity, liver disease, depression, HIV, renal disease, congestive heart failure, hypertension, hyperlipidemia, anxiety, psychological disorder, irritable bowel disease, epilepsy, thyroid disease, ulcers, and migraine. We used the following child conditions: asthma, diabetes, sinusitis, upper respiratory infections, seizures, cerebral palsy, mental retardation, cancer, appendicitis, hemanemia, congenital heart disease, renal disease, urinary tract infection, depression, attention deficit hyperactivity disorder, otitis media, and ache.

(8) One concern is that health may be important because of the physical demands of the job. As a check, we included interactions of the occupation and industry indicators with worker health and found that the results on family health were robust.

(9) We found virtually identical results when we used expected health plan payments as an alternative dependent variable in the models.

(10) To reduce the possible endogeneity of medical conditions with respect to policy quality and job change, we re-specify the health measures to include only those conditions that are discovered before the survey year. We find no qualitative change in the results.

(11) The relative risk ratio for insured small firm employment relative to insured large firm employment for the number of conditions measures the effect of a 1-unit increase in conditions on Prob(Insured Small Firm Worker)/Prob(Insured Large Firm Worker).

(12) In general, the effect of age on employment distortions is quite different from the effect of health, even though age and health are correlated. Age is correlated with seniority, experience, and human capital accumulation. Therefore, even though age is correlated with health, we cannot make clear predictions about how insured small firms will value older workers compared to younger workers.

(13) We found some evidence that workers with spouses who were employed and those who had spouses belonging to unions were more likely to be new hires in insured small firm jobs compared to insured large firm jobs. However, most of the spousal variables were statistically insignificant. Furthermore, controlling for spousal characteristics does not change the effect of health on employment among married individuals.

(14) The sample used in estimating models for the stock of all workers includes workers who have tenure of less than one year (the new hire sample).

(15) Using a Hausman test, we determined that this model did not violate the Independence of Irrelevant Alternatives assumption of the multinomial logit model.

(16) The separation analysis bears some similarity to the job lock literature. Like the job lock literature, our models focus on workers with a high demand for health care in insured firms; however, unlike the job lock literature, we contrast small and large firms. If individuals in insured small firms have worse policies and are less likely to be subject to job lock than individuals in insured large firms, we may expect the results in the separation models to be affected by job lock. However, evidence from the Kaiser-HRET survey indicates that policies in small and large firms are quite similar. Furthermore, the literature on job lock is quite mixed, and several studies suggest that job lock may be quite small (Kapur 1998; Gilleskie and Lutz 2002).

(17) The RRR and coefficients of interaction terms in non-linear models can be misleading; therefore, we have calculated predictions of the marginal effects and their distribution (Ai and Norton 2003). We have also estimated linear probability models instead of multinomial logit models and found very similar results.

Kanika Kapur, * Jose J. Escarce, ([dagger]) M. Susan Marquis, ([double dagger]) and Kosali I. Simon ([dagger])

* School of Economics, University College Dublin, Belfield, Dublin 4, Ireland; E-mail kanika.kapur@ucd.ie; corresponding author.

([dagger]) Division of General Internal Medicine and Health Services Research, Department of Medicine, David Geffen School of Medicine at UCLA, Los Angeles, CA 90095, USA; E-mail escarce@rand.org.

([double dagger]) RAND, 1200 South Hayes Street, Arlington, VA 22202, USA; E-mail susanm@rand.org.

([section]) Department of Policy Analysis and Management, Cornell University, Ithaca, NY 14853, USA; E-mail kis6@cornell.edu.

This research was supported by a grant from the Economic Research Initiative on the Uninsured (ERIU) at the University of Michigan. We thank Paul Devereux and participants at the ERIU July 2005 conference and the Geary Institute seminar for helpful comments. We also thank Jessie Riposo for programming assistance.

Table 1. Means in the Medical Expenditure Panel Survey (1996-2001)

 Small Firms

 Don't
 Offer HI Offer HI

Demographics
 Age 38.93 34.57
 Female 0.49 0.53
 Married 0.48 0.38
 White 0.80 0.71
 Black 0.09 0.09
 Hispanic 0.09 0.16
 Other race 0.03 0.04
 Family size 2.83 3.09
 Less than high school 0.07 0.22
 High school 0.57 0.60
 College 0.19 0.09
 More than college 0.07 0.03
 Other degree 0.10 0.06

Health
 No. of medical conditions 1.96 2.04
 Any medical conditions 0.55 0.55
 Predicted medical expenses 1437.85 1559.82

Job characteristics
 Wage 14.42 8.90
 Union 0.09 0.01
 Dual worker family 0.66 0.68

Other characteristics
 Region: Northeast 0.17 0.16
 Region: Midwest 0.25 0.22
 Region: South 0.36 0.37
 Region: West 0.22 0.25
 MSA 0.80 0.78

Number of observations 9010 9018

 Large Firms

 Don't
 Offer HI Offer HI

Demographics
 Age 40.11 32.97
 Female 0.47 0.57
 Married 0.51 0.33
 White 0.76 0.66
 Black 0.12 0.15
 Hispanic 0.08 0.15
 Other race 0.04 0.05
 Family size 2.82 3.13
 Less than high school 0.07 0.19
 High school 0.50 0.58
 College 0.23 0.13
 More than college 0.11 0.04
 Other degree 0.08 0.06

Health
 No. of medical conditions 2.02 2.00
 Any medical conditions 0.57 0.55
 Predicted medical expenses 1460.56 1501.52

Job characteristics
 Wage 16.95 9.47
 Union 0.21 0.06
 Dual worker family 0.67 0.68

Other characteristics
 Region: Northeast 0.19 0.19
 Region: Midwest 0.25 0.23
 Region: South 0.35 0.33
 Region: West 0.21 0.25
 MSA 0.84 0.83

Number of observations 28,989 6446

HI = health insurance.

Table 2. Multinomial Logit Models of New Hires (1996-2001)

 Small Firm HI vs. Small Firm HI vs.
 Large Firm HI Small Firm no HI

 Odds Standard Odds Standard
 Ratio Error Ratio Error

Small firm: Less than 25 employees

No. of conditions
 Full sample 0.942 0.026 ** 0.946 0.026 **
 Single workers 0.936 0.038 * 0.936 0.036 *
 Married workers 0.937 0.035 * 0.946 0.037
 Worker health 0.892 0.070 0.829 0.068 **
 Dependent health 0.955 0.044 0.995 0.048

Any conditions
 Full sample 0.860 0.062 ** 0.861 0.065 **
 Single workers 0.818 0.080 ** 0.809 0.079 **
 Married workers 0.908 0.097 0.909 0.105
 Worker health 0.856 0.101 0.822 0.104
 Dependent health 0.905 0.095 0.911 0.102

Predicted expenditures
 Full sample 0.981 0.010 * 0.978 0.010 **
 Single workers 0.979 0.013 0.973 0.013 **
 Married workers 0.981 0.014 0.981 0.015
 Worker health 0.980 0.016 0.972 0.017 *
 Dependent health 0.981 0.014 0.984 0.015

Small firm: Less than 50 employees

No. of conditions
 Full sample 0.952 0.023 ** 0.957 0.022 *
 Single workers 0.938 0.033 * 0.947 0.030 *
 Married workers 0.959 0.032 0.960 0.032
 Worker health 0.881 0.063 * 0.828 0.059 **
 Dependent health 0.989 0.040 1.015 0.040

Any conditions
 Full sample 0.901 0.061 0.891 0.058 *
 Single workers 0.877 0.079 0.858 0.072 *
 Married workers 0.939 0.092 0.915 0.092
 Worker health 0.796 0.086 ** 0.783 0.087 **
 Dependent health 0.988 0.095 0.945 0.091

Predicted expenditures
 Full sample 0.985 0.009 * 0.982 0.009 **
 Single workers 0.984 0.012 0.978 0.011 **
 Married workers 0.986 0.013 0.983 0.013
 Worker health 0.970 0.014 ** 0.966 0.015 **
 Dependent health 0.996 0.013 0.993 0.013

No. of observations 12,059

 Large Firm HI vs.
 Large Firm no HI

 Odds Standard
 Ratio Error

Small firm: Less than 25 employees

No. of conditions
 Full sample 1.012 0.022
 Single workers 0.989 0.027
 Married workers 1.051 0.034
 Worker health 0.951 0.064
 Dependent health 1.094 0.044 **

Any conditions
 Full sample 1.073 0.067
 Single workers 1.008 0.080
 Married workers 1.173 0.113 **
 Worker health 0.994 0.107
 Dependent health 1.071 0.101

Predicted expenditures
 Full sample 1.006 0.008
 Single workers 0.996 0.010
 Married workers 1.023 0.013 *
 Worker health 0.996 0.015
 Dependent health 1.016 0.013

Small firm: Less than 50 employees

No. of conditions
 Full sample 1.024 0.025
 Single workers 0.993 0.031
 Married workers 1.084 0.042 **
 Worker health 1.003 0.076
 Dependent health 1.120 0.053 **

Any conditions
 Full sample 1.120 0.078
 Single workers 1.027 0.091
 Married workers 1.287 0.140 **
 Worker health 1.125 0.136
 Dependent health 1.114 0.119

Predicted expenditures
 Full sample 1.012 0.010
 Single workers 0.999 0.012
 Married workers 1.036 0.015 **
 Worker health 1.013 0.017
 Dependent health 1.019 0.015

No. of observations

Categories: Small firm health insurance (HI), Large firm H1, Small
firm no HI, Large firm no HI. Alternative definitions of health and
firm size. All models include worker education, sex, age, marital
status, race, family size, wage, union, industry, occupation, spouse
work status, MSA, region, and year dummies.

* Significance at the 10% level.

** Significance at the 5% level.

Table 3. Multinomial Logit Models of Stock (1996-2001)

 Small Firm HI vs. Small Firm HI vs.
 Large Firm HI Small Firm no HI

 Odds Standard Odds Standard
 Ratio Error Ratio Error

Small Firm: Less than 25

No. of conditions
 Full Sample 0.985 0.012 0.989 0.014
 Single Workers 0.985 0.018 0.970 0.020
 Married Workers 0.986 0.015 1.011 0.020
 Worker Health 0.983 0.028 0.997 0.038
 Dependent Health 0.993 0.023 1.056 0.034 *

Any conditions
 Full Sample 0.909 0.030 ** 0.966 0.039
 Single Workers 0.903 0.041 ** 0.940 0.051
 Married Workers 0.916 0.042 * 0.993 0.058
 Worker Health 0.956 0.046 1.025 0.065
 Dependent Health 0.943 0.043 1.069 0.066

Predicted expenditures
 Full Sample 0.988 0.004 ** 0.993 0.005
 Single Workers 0.988 0.006 * 0.989 0.007
 Married Workers 0.988 0.006 * 0.999 0.008
 Worker Health 0.995 0.007 1.003 0.009
 Dependent Health 0.987 0.006 ** 0.995 0.008

Small Firm: Less than 50

No. of conditions
 Full sample 0.983 0.010 * 0.986 0.013
 Single workers 0.979 0.015 0.965 0.017 **
 Married workers 0.987 0.013 1.009 0.017
 Worker health 0.963 0.025 0.979 0.034
 Dependent health 0.995 0.021 1.059 0.030 **

Any conditions
 Full sample 0.923 0.028 ** 0.963 0.035
 Single workers 0.904 0.037 ** 0.923 0.045 *
 Married workers 0.946 0.039 1.004 0.052
 Worker health 0.925 0.039 * 0.988 0.056
 Dependent health 0.971 0.039 1.104 0.060 *

Predicted expenditures
 Full sample 0.989 0.004 ** 0.992 0.005
 Single workers 0.986 0.005 ** 0.986 0.006 **
 Married workers 0.992 0.005 1.000 0.007
 Worker health 0.989 0.006 * 0.998 0.008
 Dependent health 0.996 0.005 1.000 0.007

No. of observations 52,437

 Large Firm HI vs.
 Large Firm no HI

 Odds Standard
 Ratio Error

Small Firm: Less than 25

No. of conditions
 Full Sample 0.983 0.015
 Single Workers 0.980 0.017
 Married Workers 1.012 0.018
 Worker Health 0.952 0.035
 Dependent Health 1.084 0.033 **

Any conditions
 Full Sample 1.027 0.040
 Single Workers 0.963 0.047
 Married Workers 1.092 0.063
 Worker Health 0.949 0.058
 Dependent Health 1.223 0.073 **

Predicted expenditures
 Full Sample 1.001 0.005
 Single Workers 0.992 0.007
 Married Workers 1.011 0.008
 Worker Health 0.992 0.008
 Dependent Health 1.014 0.008 *

Small Firm: Less than 50

No. of conditions
 Full sample 0.989 0.016
 Single workers 0.970 0.020
 Married workers 1.019 0.022
 Worker health 0.959 0.039
 Dependent health 1.088 0.038 **

Any conditions
 Full sample 1.075 0.047 *
 Single workers 1.014 0.056
 Married workers 1.140 0.075 **
 Worker health 0.972 0.068
 Dependent health 1.227 0.085 **

Predicted expenditures
 Full sample 1.007 0.006
 Single workers 0.999 0.007
 Married workers 1.016 0.009 *
 Worker health 0.995 0.010
 Dependent health 1.016 0.009 *

No. of observations

Categories: Small firm health insurance (HI), Large firm HI, Small
firm no HI, Large firm no HI. Alternative definitions of health and
firm size. All models include worker education, sex, age, marital
status, race, family size, wage, union, industry, occupation, spouse
work status, MSA, region, and year dummies.

* Significance at the 10% level.

** Significance at the 5% level.

Table 4. Multinomial Logit Models of Separations (1996-2001).
Categories: Quit, Layoff, Stay. Alternative Definitions of Health and
Firm Size

 Full Sample

 Quit/Stay Layoff/Stay

 Odds Odds
 Ratio SE Ratio SE

Small firm: Less than 25 employees

No. of conditions * HI * small firm
 Full sample 0.938 0.030 ** 0.928 0.038 *
 Prediction: 25th pct -0.008 0.002 * -0.004 0.0013
 Prediction: 50th pct -0.006 0.004 * -0.003 0.002
 Prediction: 75th pct -0.004 0.005 -0.002 0.0029
 Single workers 0.9463 0.042 0.9317 0.0542
 Married workers 0.929 0.046 0.919 0.054
 Worker health 1.077 0.155 0.798 0.142
 Dependent health 0.841 0.104 0.901 0.125

Any conditions * HI * small firm
 Full sample 0.859 0.084 0.786 0.114 *
 Single workers 0.9063 0.163 0.8835 0.2337
 Married workers 0.8152 0.181 0.7391 0.2378
 Worker health 1.241 0.298 0.725 0.233
 Dependent health 0.721 0.172 0.818 0.257

Predicted expenditures * HI * small firm
 Full sample 0.975 0.015 * 0.948 0.021 **
 Single workers 0.9823 0.023 0.988 0.0348
 Married workers 0.963 0.029 0.953 0.041
 Worker health 1.027 0.033 0.943 0.041
 Dependent health 0.934 0.028 ** 1.013 0.042

Small firm: Less than 50 employees

No. of conditions * HI * small firm
 Full sample 0.927 0.029 ** 0.926 0.037 *
 Prediction: 25th pct -0.009 0.002 * -0.004 0.001
 Prediction: 50th pct -0.007 0.004 * -0.003 0.002
 Prediction: 75th pct -0.004 0.005 * -0.002 0.003
 Single workers 0.8915 0.038 ** 0.9388 0.0527
 Married workers 0.984 0.047 0.870 0.048 **
 Worker health 1.229 0.165 0.802 0.135
 Dependent health 0.919 0.107 0.812 0.112

Any conditions * HI * small firm
 Full sample 0.811 0.068 ** 0.830 0.102
 Single workers 0.7897 0.137 0.8832 0.2232
 Married workers 0.952 0.207 0.644 0.194
 Worker health 1.510 0.346 * 0.594 0.179 *
 Dependent health 0.912 0.206 0.750 0.227

Predicted expenditures * HI * small firm
 Full sample 0.961 0.014 ** 0.945 0.020 **
 Single workers 0.9597 0.022 * 0.9825 0.0333
 Married workers 0.994 0.029 0.934 0.038 *
 Worker health 1.056 0.033 * 0.922 0.038 **
 Dependent health 0.960 0.028 0.984 0.039

 Insured Sample

 Quit/Stay Layoff/Stay

 Odds Odds
 Ratio SE Ratio SE

Small firm: Less than 25 employees

No. of conditions * HI * small firm
 Full sample 0.963 0.025 0.956 0.032
 Prediction: 25th pct -0.004 0.001 -0.001 0.001
 Prediction: 50th pct -0.003 0.002 -0.001 0.001
 Prediction: 75th pct -0.002 0.003 -0.001 0.002
 Single workers 0.9874 0.04 0.975 0.048
 Married workers 0.931 0.037 0.929 0.043
 Worker health 0.941 0.104 0.911 0.134
 Dependent health 0.925 0.088 0.894 0.092

Any conditions * HI * small firm
 Full sample 0.956 0.085 0.818 0.115
 Single workers 1.1011 0.15 0.975 0.215
 Married workers 0.801 0.136 0.787 0.206
 Worker health 0.883 0.160 0.991 0.246
 Dependent health 0.862 0.161 0.826 0.200

Predicted expenditures * HI * small firm
 Full sample 0.991 0.012 0.972 0.018
 Single workers 1.0117 0.02 0.997 0.029
 Married workers 0.964 0.022 0.962 0.033
 Worker health 0.982 0.024 0.987 0.033
 Dependent health 0.956 0.022 * 0.983 0.032

Small firm: Less than 50 employees

No. of conditions * HI * small firm
 Full sample 0.947 0.019 ** 0.963 0.028
 Prediction: 25th pct -0.005 0.001 * -0.001 0.001
 Prediction: 50th pct -0.004 0.002 * -0.001 0.001
 Prediction: 75th pct -0.002 0.003 * 0.000 0.001
 Single workers 0.9505 0.031 0.968 0.043
 Married workers 0.951 0.034 0.934 0.037 *
 Worker health 0.953 0.094 0.987 0.127
 Dependent health 0.973 0.083 0.863 0.080

Any conditions * HI * small firm
 Full sample 0.863 0.069 * 0.834 0.105
 Single workers 1.0102 0.13 0.964 0.194
 Married workers 0.753 0.117 * 0.787 0.183
 Worker health 0.947 0.154 0.896 0.198
 Dependent health 0.923 0.151 0.899 0.193

Predicted expenditures * HI * small firm
 Full sample 0.978 0.011 ** 0.975 0.016
 Single workers 0.9978 0.02 0.995 0.027
 Married workers 0.963 0.020 * 0.962 0.030
 Worker health 0.990 0.022 0.980 0.029
 Dependent health 0.961 0.020 * 0.982 0.028

 Small Firm Sample

 Quit/Stay Layoff/Stay

 Odds Odds
 Ratio SE Ratio SE

Small firm: Less than 25 employees

No. of conditions * HI * small firm
 Full sample 1.001 0.026 1.009 0.034
 Prediction: 25th pct 0.000 0.002 -0.001 0.001
 Prediction: 50th pct 0.000 0.003 -0.001 0.002
 Prediction: 75th pct 0.000 0.005 -0.001 0.003
 Single workers 1.014 0.0353 1.013 0.0475
 Married workers 0.964 0.040 0.966 0.046
 Worker health 1.214 0.149 0.892 0.131
 Dependent health 0.873 0.087 1.048 0.113

Any conditions * HI * small firm
 Full sample 1.006 0.113 1.187 0.296
 Single workers 1.084 0.155 1.182 0.2536
 Married workers 0.833 0.151 0.943 0.246
 Worker health 1.412 0.275 * 0.838 0.215
 Dependent health 0.797 0.155 1.227 0.305

Predicted expenditures * HI * small firm
 Full sample 0.998 0.015 1.022 0.022
 Single workers 1.007 0.019 1.024 0.0293
 Married workers 0.975 0.024 0.989 0.034
 Worker health 1.048 0.028 * 0.974 0.034
 Dependent health 0.946 0.023 ** 1.025 0.034

Small firm: Less than 50 employees

No. of conditions * HI * small firm
 Full sample 0.998 0.023 1.017 0.028
 Prediction: 25th pct -0.001 0.002 -0.001 0.001
 Prediction: 50th pct 0.000 0.003 -0.001 0.002
 Prediction: 75th pct 0.000 0.004 0.000 0.002
 Single workers 0.992 0.0302 1.019 0.04
 Married workers 0.994 0.036 0.961 0.038
 Worker health 1.255 0.128 ** 0.935 0.112
 Dependent health 0.926 0.081 1.019 0.093

Any conditions * HI * small firm
 Full sample 0.984 0.093 1.212 0.163
 Single workers 1.011 0.1226 1.172 0.2074
 Married workers 0.872 0.132 0.961 0.207
 Worker health 1.473 0.241 ** 0.793 0.169
 Dependent health 0.905 0.146 1.245 0.259

Predicted expenditures * HI * small firm
 Full sample 0.997 0.013 1.024 0.018
 Single workers 0.997 0.016 1.02 0.0242
 Married workers 0.987 0.020 0.991 0.029
 Worker health 1.055 0.023 ** 0.972 0.028
 Dependent health 0.963 0.020 * 1.015 0.028

All models include the reported health measure, firm size, health
insurance (HI), and all two-way interactions between these three
variables. SE = standard error; pct = percentage.
Models also include worker education, sex, age, marital status, race,
family size, wage, tenure union, industry, occupation, spouse work
status, MSA, region, and year dummies. For the insured sample, the HI
control and interactions are redundant and not included in the model.
For the small firm sample, the small firm control and interactions are
redundant and not included in the model.

* Significance at the 10% level.

** Significance at the 5% level.