Behavioral Hazard in Health Insurance

While health insurance offers valuable protection against the risk of incurring large health expenses when a serious illness strikes, it has long been understood that there is a down-side to insurance as well — by making health care cheaper (lowering its cost to a generally modest copayment), health insurance may induce people to consume more health care. This phenomenon, known as "moral hazard," can result in the inefficient overuse of health care, as individuals opt to consume care that is worth less to them than it costs to produce. The well-known RAND Health Insurance Experiment of the 1970s and other studies have documented that people's use of care does indeed depend on the price they face.

More recently, health care analysts have documented another type of inefficiency in health care — the underuse of (often inexpensive) care that provides large health benefits. Examples of this abound, from diabetics' lack of adherence to drug regimens that reduce the risk of limb loss and blindness, to the underuse of beta blockers in the treatment of heart disease, to the failure of some pregnant women to make use of free prenatal care available through Medicaid.

What can explain this inefficiency? This question motivates a new study by researchers Katherine Baicker, Sendhil Mullainathan, and Joshua Schwartzstein, Behavioral Hazard in Health Insurance (NBER Working Paper 18468). The authors attribute this inefficiency to what they term "behavioral hazard" — misbehavior resulting from mistakes or behavioral biases.

Drawing from the behavioral economics literature, the authors note that there may be more than one source of bias at work. Specifically, "attention matters: choice of care may depend on the salience of symptoms, which is particularly problematic because many severe diseases have few salient symptoms. Timing matters: people may overweigh the immediate costs of care such as co-pays and hassle costs of setting up appointment or filling prescriptions. Memory matters: people may simply forget to take their medications or refill prescriptions. Beliefs matter: people may have false beliefs and poor learning mechanisms about the efficacy of different treatments."

Incorporating behavioral hazard into one's thinking about health insurance alters the standard conclusion that having consumers pay only a fraction of the cost of care (by charging them only a copayment) necessarily leads them to consume too much care. Instead, the reduced cost of care might in some cases improve the efficiency of the health care system by inducing individuals to use valuable care they would have forgone if they had faced the full cost. Indeed, research suggests that higher copays can dissuade the use of high-value care (care for which health benefits are large relative to costs) as much as of low-value care, suggesting that behavioral hazard is quite prevalent.

The authors illustrate the importance of considering behavioral hazard in the context of a recent large-scale field experiment that eliminated copayments for one group of recent heart attack victims, while another group faced a 25 percent copayment. Per patient spending was $106 higher in the free care group. Under the traditional assumption that rational patients only use care that they value at least as much as its cost, this policy generated health benefits of at most $26.50 per patient and a cost of $79.50 per patient. Yet the increased use of prescription drugs in the treatment group was associated with a reduction in mortality rates, which the authors conservatively value at $3,000. Viewed in this way, the policy generated a surplus of $2,894 per patient, or a return of $28 for every $1 spent. In short, a policy that might be viewed as having a modest welfare cost under the standard way of thinking can be seen to generate a large welfare gain once behavioral hazard is taken into account.

More generally, behavioral hazard changes the optimal design of insurance. In the standard model, the more responsive health care use is to price, the greater the efficiency cost of low copayments, as people overuse care to a greater extent. With behavioral hazard, the optimal copayment depends not only on how price affects the use of care but also on how the use of care affects health. For example, it may be optimal to have no copayments on certain drugs where health benefits are large and demand is quite sensitive to price. Interestingly, while behavioral economics has often been associated with the use of psychological interventions or "nudges" such as defaults and reminders, incorporating the behavioral point of view also changes how we analyze standard price levers.

With this new point of view, "insurance no longer provides only financial protection, it can also increase the efficiency of health care utilization by reducing behavioral hazard." However, private insurers may not set prices so as to mitigate underuse of care if naive consumers do not understand and fully value the health benefits that result. This may be particularly true when insurers expect many of their enrollees to eventually switch to another private insurer or Medicare, implying that the insurer will not realize all of the future cost savings that may result from inducing an enrollee to use more high-value care today.

The authors conclude by noting that the areas of the health care system in which we observe substantial un-derutilization, such as management of chronic diseases like diabetes, high blood pressure, asthma, and high cholesterol, are responsible for a large share of total health care costs. Much of the cost of these diseases is incurred in the late stages and likely involves overuse of care following earlier underuse as the disease progressed. "That many of these domains of care seem sensitive both to small changes in copayments and potentially to behavioral nudges - and that many of the treatments affected seem to be of high health value - suggests that incorporating not only moral hazard but behavioral hazard into our models of optimal insurance design may have large-scale implications for public policy."

The authors thank the National Institute of Aging (Grant Number T32-AG000186) for financial support.