Predictability and Predictiveness in Health Care Spending

We argue that predictability (how well spending on certain services can be anticipated) and predictiveness (how well certain services contemporaneously co-vary with total health care spending) both matter for understanding adverse selection incentives. Three main points are made. First, the predictability of specific services rather than of total spending matters for problems of adverse selection. Second, it is generally what the plan (not the individual) predicts that matters, since it is the plan (not the individual) that chooses actions such as excluding applicants or distorting services, which create adverse selection. Third, among those services that are predictable, health plans have incentives to underprovide those that are positively correlated with total spending and overprovide those that are negatively correlated with total health spending. We examine predictiveness and predictability using a 5% sample of Medicare eligibles from 1996-1997. Spending disaggregated by type of service is generally more predictable and more predictive than spending by diagnosis, place of service, or provider specialty. Certain types of disaggregated spending are significantly more predictable than total spending on medical care, such as spending on lab tests, durable medical equipment, psychiatry, emergency rooms and dialysis services. Among these services that are more predictable, spending on certain types of service -- emergency rooms, dialysis, office visits, and mental health services, to name a few -- are more positively predictive of contemporaneous spending than other services and hence are most vulnerable to service underprovision stemming from incentives related to adverse selection. JEL Keywords: Health plans, Adverse selection, Risk Adjustment, Medicare, Managed Care