Is there job lock? Evidence from the pre-HIPAA era.

• Author: Berger, Mark C.
• Position: Do workers stay in jobs due to health insurance considerations

1. Introduction

   The fundamental economic issue with job lock is the quality of matches between workers and jobs. Provision of health insurance through the place of employment complicates the labor market decisions of both workers and firms. If there are costs of changing jobs associated with changing health insurance plans, or in the extreme case, if a worker is unable to change jobs because a chronic health condition is covered by their current employer's health insurance but would not be covered by a new employer's plan, then the overall quality of matches between workers and employers in the labor market is reduced. The economy functions less efficiently, and workers are on average less productive and receive correspondingly lower pay because they are employed in less-than-optimal jobs.

   Empirical analyses of job lock have emphasized the possible reduction in mobility that occurs when some portion of compensation comes in the form of health insurance. Having to change health insurance policies when changing jobs adds to the transactions costs of a job transition, and thus will reduce the number of job changes. These costs are potentially enormous if a family member has a preexisting medical condition that would be excluded from coverage under the new employer's health insurance. A number of economists have attempted to identify and quantify the reduction in mobility caused by employer provision of health insurance, including Madrian (1994), Holtz-Eakin (1994), Monheit and Cooper (1994), Gruber and Madrian (1994), Buchmueller and Valletta (1996), Penrod (1995), Kapur (1998), Anderson (1998), and Gilleskie and Lutz (1999). (1)

   Early research by Madrian (1994) and Cooper and Monheit (1993) found that health insurance reduces mobility. They focused on transaction costs aspects of job changes, which are exacerbated if a job change also involves a change in health insurance plans. Later work that incorporates health status and employer-provided health insurance, especially Holtz-Eakin (1994), Penrod (1995), and Kapur (1998), turn up little if any evidence of job lock. Others find job lock effects for some labor force groups or equation specifications (for example, Gilleskie and Lutz 1999 for unmarried men; Buchmueller and Valletta 1996 for women; and Penrod 1995 for some equations for married men).

   The perceived magnitude of the job lock problem was a major motivation for the 1996 Health Insurance Portability and Accountability Act (HIPAA). (2) In their 1998 Health Confidence Survey, the Employee Benefits Research Institute (EBRI 1998) found that 27% of individuals reported that they or an immediate family member had experienced some form of job lock, up from 13% in 1991. Although HIPAA specifically addressed the job lock issue by largely eliminating preexisting conditions exclusions and by providing access to individual market coverage regardless of health status, after the passage of HIPAA and immediately after its implementation workers still perceived job lock to be a significant problem. (3) But the critical question is whether the actual behavior of workers is consistent with their survey responses.

   Toward that end, this paper analyzes the two primary labor market effects of job lock, reduced mobility and lower wages. Similar to Kapur (1998), we choose indicators of chronic, serious illness and functional impairments for family members that are highly predictive of health insurance demand. If anyone is job locked, it should be employees who have a family member with a serious health problem. For them, all else equal, we should see longer employment durations. At the same time, if they are in less-than-optimal jobs, they should be observed to have lower wages than other similar individuals. (4)

   We use the 1987 and 1990 Survey of Income and Program Participation (SIPP) panels to empirically examine the effects of job lock on mobility and wages during the period prior to the implementation of HIPAA. Using data prior to the passage and implementation of HIPAA gives the best chance for job lock effects to appear empirically if they are important. The SIPP enables us to follow workers over a longer period of time than most previous studies. Using the SIPP, we examine the mobility effects of job lock by estimating employment duration models that appropriately account for job tenure, which numerous studies have found to be an important determinant of the likelihood of job mobility. Unlike previous studies, we also use the SIPP to estimate wage equation models that test for the existence of job lock effects. (5) An important advantage of the SIPP in analyzing job lock is that it contains detailed information about labor market behavior over an extended period of time, in addition to information about the health status of each member of the household. (6) We exploit these strengths of the SIPP in our estimation of duration and wage models. (7) This combination of data, models, and measures provides a more complete picture of the phenomenon of job lock than has been available to date.

2. Effect of Job Lock on Mobility and Wages

   To determine the existence and magnitude of job lock effects on mobility and wages, we must first specify empirical models that allow these effects to be identified in the data. For this purpose, we estimate duration models in which workers are "at risk" of ending their current jobs, as well as fairly standard Mincer-type wage equations. We utilize a difference in differences approach that is now common in the literature, using the interaction of health insurance and health conditions measures or proxies to identify the potential job lock effects (e.g., Madrian 1994; Kapur 1998).

   Employment Durations

   Our identification of job lock effects on employment durations can be illustrated using the following logistic transformation of the hazard rate of leaving employment:

   (1) ln[h/(1 - h)] = x[[beta].sub.0] + health[[beta].sub.1] + hi[[beta].sub.2] + hi X health[[beta].sub.3],

   where h is the hazard rate, x is a vector of control variables, health is a health status indicator for workers' families, hi is an indicator that the worker has employer-provided health insurance, hi*health is an interaction variable between the health indicator for workers' families and the health insurance indicator, and the [[beta].sub.i]s are the corresponding parameters to be estimated. (8)

   Our measures of health are dummy variables indicating that the worker has a family member with health problems, hi is a dummy variable indicating that the worker has a family health insurance plan provided by an employer, and hi*health is an interaction variable indicating that a family member with health problems is covered by the insurance plan. The coefficient [[beta].sub.1] estimates the impact of having a family member with health problems for those without health insurance. The coefficient [[beta].sub.2] corresponds to the impact of employer-provided health insurance on workers with healthy family members. The difference in differences estimate of the job lock effect is the coefficient [[beta].sub.3]. It estimates the additional effect of family health problems for those workers with employer-provided family health insurance plans.

   Instead of direct measures of family health problems, several earlier studies have used proxies such as family size (Madrian 1994) and spouse health insurance coverage (Madrian 1994; Buchmueller and Valletta 1996) to test for the possibility of job lock, reasoning that larger families have higher expected medical expenditures and those with access to spouse health insurance coverage are less likely to be job locked. We also estimate versions of Equation 1 that use these proxies instead of direct family health measures for comparison purposes.

   Although we agree that those with access to spouse health insurance may have greater mobility for many reasons, including worker heterogeneity, a stronger test of job lock in the presence of spouse-provided health insurance is obtained by including both the direct measures of health and the measures of spouse health insurance. The intuition is that a person with a family member with a chronic health condition and employer-provided health insurance would be less locked if he or she were also eligible for employer-provided health insurance through the spouse. (9) This expanded model can be written

   (2) ln[h/(1 - h)] = X[[beta].sub.0] + health[[beta].sub.1] + hi[[beta].sub.2] + hi x health[[beta].sub.3] + sphi[[beta].sub.4] + sphi x hi[[beta].sub.5] + sphi x health[[beta].sub.6] + sphi x hi x health[[beta].sub.7],

   where sphi refers to employer-provided health insurance through the spouse. In Equation 1, variation in spouse employer-provided health insurance does not come into play, and the test of job lock is [[beta].sub.3] < 0. In fact, for those without eligibility for spouse-provided health insurance, this remains the test for job lock in Equation 2. We would expect, however, that the extent of job lock would be less for those with access to spouse employer-provided health insurance, suggesting that [[beta].sub.7] > 0. Thus, Equation 2 provides an even richer examination of the job lock hypothesis than is available in Equation 1.

   Wages

   If job lock interferes with workers' ability to accept alternative job offers, we would expect workers who are locked to have lower wages than workers who are not locked. Further, to the extent that employers know that a worker is job-locked, they may adjust wage offers to take advantage of the worker's immobility. To analyze this type of evidence, we next turn to estimating Mincer-type earnings equations of the form

   (3) ln y = x[[gamma].sub.0] + health[[gamma].sub.1] + hi[[gamma].sub.2] + health x hi[[gamma].sub.3] + [epsilon],

   where In y is the natural log of weekly earnings, [epsilon] is a random disturbance, and the other variables are defined as in Equation 1. If employers offer lower wages to job-locked workers or...