مرتب سازی کارگران،تفاوتهای جبران خسارت و بیمه سلامت: شواهدی از کارگران آواره

This article introduces an empirical strategy to the compensating differentials literature that (i) allows both individual observed and unobserved characteristics to be rewarded differently in firms based on health insurance provision, and (ii) selection to jobs that provide benefits to operate on both sides of the labor market. Estimates of this model are used to directly test empirical assumptions that are made with popular econometric strategies in the health economics literature. Our estimates reject the assumptions underlying numerous cross sectional and longitudinal estimators. We find that the provision of health insurance has influenced wage inequality. Finally, our results suggest there have been substantial changes in how displaced workers sort to firms that offer health insurance benefits over the past two decades. We discuss the implications of our findings for the compensating differentials literature.

Without a national health insurance system most Americans receive health benefits from their employers. As recent years have been characterized by rapid inflation in health care costs and health insurance premiums,1 there are increasing reports that employers have either reduced or even stopped offering coverage, increasing the ranks of the uninsured.2 In aggregate, the percentage of workers with employment-based health insurance has dropped from 70.0% in 1987 to 59.8% in 2004. Economists generally argue that profit maximizing employers would respond to the costs associated with the provision of health insurance by reducing wages, thus maintaining the total reward paid to the employees. Whether employers truly adjust wages for the provision of health insurance is a long-standing question in health economics. Evidence of a wage/fringe benefit trade-off has been difficult to establish empirically, in part because jobs that offer health insurance may differ substantially from those where this benefit is not provided.3 Empirical researchers have traditionally attempted to answer this question using wage regressions. Researchers arrive at conclusions by examining the sign and significance of the estimated coefficient on the health insurance receipt variable. Studies generally differ in the source of variation used to identify health insurance receipt in these regressions. The early literature on this topic typically did not explicitly contain a discussion of the source of variation in the health insurance variable and researchers implicitly assume that employees have implicitly made these choices when accepting a job offer. Using cross-sectional data researchers typically recover wrong-signed estimates of the compensating wage differential particularly if they ignore the inherent endogeneity of health insurance receipt. Longitudinal data that explicitly account for individual specific permanent unobserved heterogeneity have also been used, yet results from these studies also do not support the theoretical prediction of a significant negative compensating differential.4 To identify the trade-off, many of these studies must overcome the endogeneity of job switching. In particular, if job lock is an important factor in the labor market, then the sample used in the analysis for identification will disproportionately contain those individuals who do not feel locked into their job, possibly attributable to a lower valuation of health insurance. To overcome this empirical challenge, Simon (2001) uses panel data on displaced workers who switch jobs for exogenous reasons, but also fails to find evidence of compensating wage differentials, which she attributes to difficulty in empirically dealing with the heterogeneity of job-skill matching. In contrast, a recent literature attempts to use a credible source of variation in the costs of health insurance, which is arguably exogenous to workers employment decisions, to identify whether these additional costs are borne by workers. These studies generally find evidence consistent with the theory. For example, Gruber (1994), exploiting changes in state legislation that made coverage for childbirth a mandatory part of all health insurance policies, finds that the wages of workers expected to benefit from this policy were shifted down by 59–90%. Sheiner (1999) uses variation in health care costs across cities and finds that older workers in cities with higher health costs earn lower wages. While these studies are unable to disentangle whether the shifting of health insurance costs is on the group at the workplace or on the individual, Pauly and Herring (1999) find a flatter wage-tenure profile among job-insured workers presenting evidence consistent with an individual wage adjustment for health insurance. In his review of the compensating differentials literature, Pauly (2001) sustains the existing studies do not provide compelling evidence, either in favor, or against the existence of this trade-off. The interest in achieving a more solid answer to whether this trade-off truly exists is unquestionable, given the numerous implications for public policy. To shed light on this issue as well as provide an explanation for the diversity of the findings in this literature, we introduce a new panel data estimator to the health economics literature and re-examine the experience of displaced workers who change jobs for arguably exogenous reasons over an 18-year period. The estimation strategy originally developed in Lemieux (1998) allows observed and unobserved characteristics to be rewarded differently in firms that provide and do not provide health insurance, and it generates estimates robust to both employer and employee selection on unobservables. Estimates from our model are used to first, evaluate the existence and robustness of any potential wage and health insurance trade-off and second, to decompose the wage gap between firms that offer and do not offer health insurance. Our approach additionally allows us to test implicit assumptions underlying conventional empirical approaches used in the health economics literature. Examining the wage decomposition provides an opportunity to understand how displaced workers sort into new positions in the labor market. The equalizing differences literature that underlies the wage health insurance trade-off implicitly assumes that workers should sort into jobs with different attributes based on their preferences for those attributes.5 In contrast, comparative advantage would suggest that workers sort into firms or sectors in which they would perform relatively better than other potential employees in the labor market. Understanding the sorting process potentially could yield insights into how the labor market functions and how inequality develops.6 For example, if the underlying distribution of skills among displaced workers were stable over time, any changes in the manner in which these workers sort to new jobs would lead to a potentially different allocation of these skills across firms over time.7 While it is well known that displaced workers experience large and persistent earnings loss,8 this group is becoming a topic of increased policy relevance. Since many Americans are a pink slip away from losing their health insurance coverage,9 numerous policies have been introduced in the last five years to promote the continuation of health insurance coverage for displaced workers. For example, the Health Coverage Tax Credit introduced federally in 2002 covers 65% of the premium amount paid by eligible displaced individuals for health insurance coverage, where eligibility is primarily based on prior industry. By reducing the costs of health insurance while unemployed, these polices may alter the search behavior for newly displaced workers by increasing the take-up of unemployment benefits and the length of unemployment spells. In addition, several individual states have introduced health plans for displaced workers and new policies are being continually debated at both the federal and state level. To provide some guidance for policymakers regarding the impacts of such programs, we use our data and estimates from the wage decomposition to ascertain whether there have been any changes in how displaced workers search for and sort into a new job following displacement (based on the provision of employer provided health insurance) over the last two decades. Our analysis yields three major findings for the health economic literature on the impacts of employer based health insurance provision. (1) We find that the diversity of results in the health economics literature on the existence of a compensating wage differential may be more a consequence of imposing too stringent assumptions on the empirical model rather than a failure of the underlying theory. Our empirical estimates reject assumptions that underlie single index control function, OLS, matching, fixed effects and first difference strategies. Specifically, we clearly reject the assumptions that permanent individual specific unobserved (to the analyst) heterogeneity is constant and that selection of jobs by displaced workers can be explained by variables observed to the analyst. Further, since selection of this new job is based in part on factors unobserved to the analyst, our results suggest that this does not operate exclusively on the workers’ side of the labor market but also affects hiring decisions by firms. Researchers in this area must account for both employee and employer decisions regarding employer provision of health insurance. (2) Our empirical results indicate substantial changes have occurred over the past two decades both in how displaced workers sort across firms when seeking re-employment and how firms select workers for employment. We observe that, if on one hand, the importance of selection bias in explaining the unadjusted wage gap has diminished by over 40%, on the other hand, the portion of this bias due only to unobserved (to the analyst) characteristics, such as ability or innate health status has more than doubled. Finally, we find that recently displaced workers are searching nearly three weeks longer for jobs that provide health benefits, suggesting that those who need health insurance shop for it. (3) We find that the provision of health insurance has substantially influenced wage inequality. Similar to the large established literature that has explored the determinants of wage structure in the U.S. labor market,10 we estimate that, in the past decade, firms that provide health insurance are offering increasingly larger returns to observed individual characteristics. In particular, we find that there is nearly a 30% increase in the returns to a college education in firms that provide health benefits. Residual wage dispersion has also increased over the past two decades as our results indicate that the return to unobservable skills, such as motivation, innate health status or cognitive ability has risen. While we do not find any evidence of a health insurance compensating wage differential, we find that the effect of health insurance on wage of workers in firms providing health insurance has increased by 50% between decades. In addition, we present evidence that the role of unobserved characteristics is increasingly affecting the dispersion of wages between sectors. Our results suggest that a larger fraction of the displaced work force is seeking jobs with health insurance benefits and, faced with a more heterogeneous pool of applicants, employers in the health insurance sector are increasingly rewarding observed skills. We hypothesize that the inability to identify the impact of health insurance on wage levels is likely a consequence of omitting firm characteristics that arise from aggregate worker sorting rather than the heterogeneity of individual job-skill matching. The rest of the paper is organized as follows. In the next section, we introduce the economic model and empirical method that we employ to estimate the parameters of the model. Section 3 describes the data used in our analysis. Estimates of our model and other empirical results are presented and discussed in Section 4. A concluding section discusses what our findings on worker sorting imply for the compensating differentials literature and suggests avenues for further research.

Surveys of workers consistently rank health insurance as far and away the most important among all benefits offered in the workplace (Salisbury, 2001). Intuitively, health insurance has the potential to influence numerous labor market decisions and many individuals may be reluctant to consider working for companies that do not provide health benefits.34 At the same time, employers are increasingly facing large increases in the costs of providing these benefits. Hence, understanding how the provision of these benefits affects the labor market has substantial policy and human resource implications. In this paper, we extended the health economics literature on compensating differentials by introducing an empirical approach that allows (1) both individual observed and unobserved (to the econometrician) characteristics to be rewarded differently in different sectors of the economy, and (2) selection to operate on both sides of the labor market. An important feature of this model is that the estimates enable us to test several assumptions that are made with existing empirical strategies. We find that the assumption that unobserved attributes are rewarded equally in both sectors which underlies fixed effects and traditional differencing strategies is rejected. In addition, we find evidence that there is substantial selection on unobserved characteristics, which rules out matching and OLS strategies. Finally, we find that selection operates on both sides of the labor market, a feature that cannot be accounted for by traditional control function estimators. While we do not find any evidence of a health insurance compensating wage differential, we observe that health insurance has increasingly influenced wage inequality. We present evidence that the provision of health insurance is increasingly affecting the dispersion of wages across sectors, which is consistent with the findings in the compensating differentials literature that have explored the incidence question from the perspective of employers. Estimates from our model are also used to decompose the wage gap between the sectors and we find there are substantial changes in the selection of workers to firms that provide health insurance benefits. Specifically, we observe that there has been increasing sorting based on comparative advantage. Finally, we find that recently displaced workers are searching nearly three weeks longer for jobs that provide health benefits and that these workers on average have lower unobserved productivity attributes. An important limitation of this study is that the impacts we estimate are applicable to displaced workers only. The composition of displaced workers not only differs from other workers in the labor market, but has also changed over time.35 Yet, it should be noted that our finding of an increasing role of health insurance on wages is consistent with the recent health economics literature that indicates that the recent rise in health insurance premiums affects a variety of labor market outcomes.36 Consequently, it is likely our findings have some limited external validity. We do not view our failure to find evidence of a compensating differential despite using a more general estimation approach as a rejection of the underlying theory. The simple compensating differentials theory remains inconsistent with two empirical features of the U.S. labor market. First, there is large variability in costs of health insurance across similar firms. Understanding the sources of this variability may suggest additional variables that may need to be accounted for in wage regressions. In this study, we find strong evidence that the sorting of workers to jobs is consistent with comparative advantage.37 While, sorting in the labor market is likely to lead to strong correlations between observed and unobserved attributes of the individual with the corresponding variables of their co-workers, our evidence also indicates that in recent years the distribution of these unobserved attributes is more heterogeneous in the health insurance sector. Thus, sorting to firms on the basis of preferences for health insurance is not perfect and there still may remain heterogeneity at the firm level that is not a function of these individual specific heterogeneities, suggesting that frictions exist in the labor market. Accounting for such heterogeneity in addition to individual heterogeneity as well as the identification of the impacts of endogenous group formation at firms has typically been ignored (in part due to data limitations) in the health insurance compensating differentials literature.38 Second, employers cannot set employee specific compensation packages. An increasing body of evidence suggests that there is likely substantial heterogeneity regarding preferences for health insurance benefits among workers within firms. For example, Gruber and Lettau (2004) find that within firms, the median worker and workers in the highest quantiles of salary exhibit a disproportionate amount of influence on decisions related to health insurance coverage. These workers may also be willing to bear a disproportionate amount of the costs through lower wages, and if these tastes can indeed be proxied by observables such as salary, it may be useful to examine whether the trade-off exists by examining those most likely affected. For example, with longitudinal matched employee and employer data it is possible to examine within firms how the estimated trade-off varies across the salary distribution and how the distribution of residual wage dispersion varies across firms that differ in both benefit provision and the cost of benefits.39 In conclusion, future research needs to consider general empirical models with richer data sources to determine whether the compensating differential truly exists as well as shed more light on how health insurance provision affects the labor market.