Economists often talk about "moral hazard," the idea that people's behavior changes in the presence of insurance. In finance, for instance, investors may take more risks if they know they will be bailed out, the subject of ongoing political controversy. When it comes to health insurance, the existence of moral hazard is a more matter-of-fact issue: When people get health insurance, they use more medical care, as shown by research including a recent randomized study on the impact of Medicaid, which MIT economist Amy Finkelstein helped lead.

Such evidence helps explain why insurers and policymakers looking to reduce overall costs have become increasingly attracted to the concept of "consumer-directed medical care," in which consumers pay for a larger share of medical expenses, sometimes in the form of high-deductible insurance plans. If people have to bear greater initial costs (the deductible（可减免的） is the amount consumers must pay before coverage kicks in), they may be less likely to seek insured medical care for seemingly marginal health issues.

But a new paper co-authored by Finkelstein suggests that forecasting the likely spending reduction associated with high deductibles requires a fine-grained approach, to account for the differing ways consumers respond to incentives in the health-care market. The research indicates that consumers select insurance plans based not only on their overall wellness level -- with people in worse health opting for more robust coverage -- but also on their own anticipated response to having insurance.

By scrutinizing a large data set based on choices made by employees of Alcoa, Inc., the researchers found that consumers selecting a new insurance policy who expect to reduce their use of medical care by a lot if they have to pick up additional costs shy away from high-deductible plans; by contrast, the people who opt for high-deductible plans are the ones who expect to change their health-care use the least.

Thus insurers, or at least those who expect that offering more high-deductible plans will lower their expenses, may experience smaller spending changes than they might expect if a random group of people were assigned to such plans.

"What we [find] is that if you base your forecasts on random assignment, you would substantially overestimate the spending reduction you can get by introducing high-deductible plans," says Finkelstein, the Ford Professor of Economics at MIT. "The people who select these plans aren't randomly drawn from the population -- they tend to be people who have a lower behavioral response to the [insurance] contract."