We apply our ideas to a Medicaid data set to illustrate how to calculate distortion incentives, and we conduct policy analyses of risk adjustment and carve out options add comment.
Our paper is related to other recent research in applied industrial organization that begins with an explicit characterization of conditions for profit maximization and information constraints in the market. That literature has explained phenomena such as the inefficient choice of the number of product lines (Klemperer and Padilla, 1997) and entry and exit in hub and spoke networks (Hendricks, Piccione and Tan, 1997). These papers use the profit maximization conditions to explain observed equilibria that appear to deviate from simple market models.
From the a practical standpoint of in health policy, our paper shows how the incentives to distort services depend in a relatively straightforward way on means and correlations among predicted values of health care services in a population. Several interesting findings emerge from the small data set we analyze. The most striking is the importance of individuals’ knowledge and their ability to forecast their health expenses.
This factor has been appreciated in abstract terms in earlier writing, but the dramatic effect that information has on incentives has not been fully appreciated. As we figure it, if people know what they are sometimes commonly assumed to know (age, sex and prior spending), selection incentives would be very severe and beyond the power of existing risk adjusters to deal with. Study of what individuals can forecast is a very key area of empirical research.
We are forced in this paper to analyze hypothetical cases in which individuals are not allowed to know “too much.” Within this limitation, we illustrate how in this data set, risk adjustment can be assessed and decisions about the efficiency effects of carve outs can be made. Two proposed risk adjustment systems have significant and similar effects in terms of cutting the magnitude of distortion incentives. Carve outs too can help, especially when one service seems to be the major distortion instrument, as mental health and substance abuse is in one of our scenarios.
Another point to emphasize is that the specifics of the results will vary according to the underlying patterns of use in a population. We have analyzed one relatively small data set of young women on welfare. The distortions likely to arise for the elderly or for an employed population may be quite different.