Explaining the RAND Experiment

At Overcoming Bias, David Balan ponders the counterintuitive results of the famous RAND health insurance study. The RAND study gave one group of patients more generous coverage, and a second group of patients less generous coverage. Not surprisingly, the former group consumed a lot more health services. Surprisingly, the two groups had indistinguishable health outcomes. This is bizarre because, although we would expect health services to have diminishing marginal returns, we would not expect them to have zero marginal returns.

But Balan has a different problem with the diminishing returns explanation:

The problem is that some of the other evidence from the RAND study is not really consistent with this story. It seems that the marginal care consumed only by people with more generous insurance is not just low-value stuff. The marginal treatments consumed only by those with more generous insurance, in the opinion of expert doctors, looks a lot like the infra-marginal treatments consumed by everybody. But if that's true, doesn't it have to mean that all health care is of little value? If the marginal care looks just like the infra-marginal care, and the marginal care is of little value, then doesn't the infra-marginal care have to be of little value too?

I don’t find this objection compelling. The concept of diminishing marginal returns does not require that marginal and inframarginal units differ from each other, only that they differ in their effects. If my firm experiences diminishing returns to labor, for instance, that does not mean the additional workers I’ve hired are not as good as the previous workers in terms of their strength or skill. I could hire perfectly identical workers and still experience diminishing returns because of overcrowding, insufficient capital support, and so on. Similarly, perfectly identical medical treatments might have diminishing returns because the most good is done by the initial treatments, little good by subsequent treatments.

What requires an explanation is not that marginal returns to (even identical) health services fall, but that they seem to fall to zero. For that, I’m guessing three factors are at work.

1. As many have noted (most recently here), additional health services can have adverse consequences. Additional hospitalization can mean greater exposure to disease, for instance. If health services are characterized by diminishing marginal health benefits and increasing (or even constant) marginal health costs, the net marginal effect eventually goes negative – even ignoring the cost of providing the service.

2. We’re only looking at two different groups, one of which consumed much more health care than the other – about 40% more as I recall. If the first additional 20% generated positive net returns, and the next 20% generated negative net returns, then it might appear that the whole extra 40% had close-to-zero impact. We don’t know how a hypothetical third group receiving an intermediate amount of care would have fared; they might have done better than both groups. (If the RAND study did look at more than two groups, I haven’t heard about it.)

3. There are health outcomes that aren’t captured in standard health indicators like mortality and disability. Inconvenience and discomfort, for instance. Say that patient A and patient B had the same irritating skin rash. In both cases, the rash would have gone away on its own after a month, but patient A went to the doctor and got a treatment that made the rash clear up in a week. At the end of the year, patients A and B might well have the same measured health outcomes. But that doesn’t mean A’s additional treatment had zero marginal benefit.