demand equal to –0.19 among wives and–0.08 among husbands, suggesting that own-employer
take up among husbands is less sensitive to price than own-employer take up among wives.
Other research has found candidates for family coverage to be more price-responsive than single
workers (Blumberg et al. 2001). Estimating worker-level coverage (not only direct take up, but
coverage from any source), Bernard and Selden (2003) found significantly greater price
responsiveness among workers in small firms as well as among workers with lower income,
lower health risk, or both.
As with studies of employer offer, most studies of worker demand for employer-based
coverage are observational, not experimental. As a result, they usually must deal with the
absence of price data for workers who decline coverage, making it necessary to impute price for
these workers based on their characteristics.
Here again, the imputation of prices introduces potential endogeneity bias from two sources.
First, workers who select jobs with an affordable offer of health insurance may have a greater
demand for insurance that may not be observable in terms of the measures that are available
(Monheit and Vistness 1999). In short, their health plan premiums may be correlated with
unobservable components of their demand for insurance. Second, the quality of insurance plans
offered by employers is usually unobservable, and it may be correlated with the price. If so, the
price elasticity estimate is biased toward zero.
While instrumental variables may be used to address these endogeneity problems, finding
the right instrumental variable (one that is correlated with the endogenous variable but
uncorrelated with the dependent variable) can be very difficult. For example, Cutler (2002b)
used the state’s average marginal tax rate as the instrumental variable for the employee premium
share, but it may be inadequate as an instrumental variable: it may reflect the demand for