The University of Michigan
555 South Forest Street
Third Floor
Ann Arbor, MI 48104-2531

T 734-936-9842
F 734-998-6341



Workers whose health insurance comes from their own jobs are far less likely to change jobs than workers whose coverage comes elsewhere or who do not have insurance. Economists have given considerable attention to understanding what part of the difference represents a loss of efficiency due to workers being "locked" into their jobs for fear of losing some or all of their coverage if they change jobs. Matthew Dey and Christopher Flinn posit a structural job search model that offers an alternative explanation for the observed patterns of health insurance offerings and job separations. Specifically, in the context of the job search model, lower rates of turnover among those with insurance stem from the match value of such jobs and positive effects of insurance on health status and productivity. In applying their model to data, they find little evidence that the observed rate of job change is inefficient unless there is a wide disparity across employers in their cost of health insurance.

Descriptive statistics of search and health insurance

  • Half (50.7%) of unemployment spells among white males ages 25-54 end with a transition into a job that provides health insurance.
  • Among the jobs that prime age adult men take after a spell of unemployment, wages among jobs with health insurance are 41% higher than among jobs without health insurance.
  • Prime age adult men who go from one job without health insurance to another without health insurance obtain wages on their new jobs that are 6% larger than the wages of those who go from jobs without to jobs with health insurance.

Assumptions of the search model.

  • The Dey-Flinn model allows for firms to be heterogeneous in their cost of providing health insurance to their workers and workers to be heterogeneous in their demand for health insurance. Workers and employers bargain over the surplus created from the job-worker pairing they create; their relative bargaining power is only weakly identifiable, and so results are shown with varying assumptions about this parameter.
  • Workers (and firms) make optimal choices, i.e., they take the job whose wage-health insurance pair makes them better off. Inefficiency arises when individuals decline job matches that would be otherwise desirable because the firm has high costs of providing coverage relative to other firms. Inefficiency occurs both as 'job lock' (staying rather than moving) and 'job push' (leaving even though the current job is a better match).
  • The empirical implementation of the model implies that about 3 percent of would be job transitions inefficiently do not take place because of job lock/job push.
  • In the specification which allows for worker and firm heterogeneity, the parameters imply that a job without health insurance will last about a year and a half while a job with health insurance will last 12 years.

The Dey-Flinn model matches many of the observed features of the data. The model differs from the data in the share of low wage jobs that come with health insurance; more low wage workers have it than the model implies. This suggests that in the long run fewer low wage jobs will come with health insurance.

The paper suggests that distortions to the labor market through what is known as "job lock" and "job push" are not substantial. The paper implies that if people could move between jobs independent of health insurance considerations, the labor market would not be substantially more efficient. Insurance market reforms would make the labor market only marginally more efficient.

The paper is based on a complex structural model estimated using data from the Survey of Income and Program Participation (SIPP). Its policy conclusions are not based on well identified econometric evaluations of policies designed to influence job lock. For example, while the model developed in the paper could simulate how portability reforms or coverage mandates might affect job mobility, the paper does not evaluate such policies.

The paper relies heavily on the assumption that the health effects of insurance are sufficient to meaningfully reduce the rate of adverse health shocks that result in job separations. It is difficult to verify that the phenomenon captured by this parameter is truly a health-related productivity effect as opposed to a preference effect that is more in the spirit of the earlier job lock literature. That said, the key identifying assumption of the model, which is that individuals at jobs not covered by health insurance are more likely to exit the job spell into unemployment is not imposed in the estimation of the model, but is overwhelmingly substantiated by the raw transition data and model parameter estimates.

What "locks" workers is described differently elsewhere in the "job lock" literature than in the Dey-Flinn approach. "Lock" is usually described as something that arises from a wedge between how a worker values coverage from a current job compared to a prospective job. Pre-existing conditions have been cited as a source of such a wedge. A worker (or dependent) has coverage for a condition with the current employer but would be subject to a "pre-existing condition" exclusion if the worker changed jobs (a result that is less likely after the enactment of the Health Insurance Portability and Accountability Act of 1996.) The cost of covering an individual would presumably be the same in both jobs. By contrast, the Dey-Flinn model looks to differences in cost between firms for the same health insurance coverage as the source of "lock" and "push."

The model makes assumptions about the distribution of worker productivity and errors (both log normal), and the division of surplus between workers and employers. It is not possible to formally test these assumptions, though the predicted wage distributions are not seriously at odds with their sample counterparts.

The estimation is done with a sample restricted to adult white males with at least a high school education, and the generalizability is limited to that sample.

Survey of Income and Program Participation, 1996 panel. The sample is white males between ages 25 and 54 with at least a high school education, excluding any individual who reports attendance in school, self-employment, military service, or participation in AFDC, WIC, or Food Stamps. A total of 10,121 individuals meet the criteria, of whom 2,814 are observed to be unemployed at some point and 7,307 are not.

Model: Continuous time search model with idiosyncratic match values and firm/searcher bargaining over match surpluses.

Empirical implementation: Simulated maximum likelihood estimation with 1000 repetitions of the simulation. Alternative specifications of the model (no heterogeneity, firm heterogeneity, and both firm and worker heterogeneity, firm and worker heterogeneity with worker heterogeneity.)

An Equilibrium Model of Health Insurance Provision and Wage Determination
Matthew S. Dey and Christopher J. Flinn

Conference paper presented at ERIU Research Conference, July 2003

The final version of this paper appeared in Econometrica 73 (March 2005): 571-627

ERIU Working Paper #19 (Adobe PDF)

Back to top

Funded by The Robert Wood Johnson Foundation, ERIU is a five-year program shedding new light on the causes and consequences of lack of coverage, and the crucial role that health insurance plays in shaping the U.S. labor market. The Foundation does not endorse the findings of this or other independent research projects.