I applaud John Roemer’s effort to defend egalitarianism and to define it in a way that incorporates a role for personal responsibility. Although, as I discuss below, I have serious reservations about the details of the particular scheme he has laid out, his broad objectives strike me as compelling.

Roemer proposes to classify people according to the characteristics that they cannot control. Within a class of people sharing the same uncontrollable characteristics (a “type”), individuals are to be held accountable only for their differences. He then goes on to compare responsibility across types. To modify one of his examples slightly: if the median White male steelworker smokes for thirty years, whereas the median Black female college professor smokes for eight years, then the former should be judged, according to Roemer, as “comparably responsible” to the latter, even though he has smoked for twenty-two more years.

This judgment is based on the assumption that it would be as “hard” for the steelworker to keep his smoking down to thirty years as for the professor to keep hers down to eight. But is this assumption correct? As justification, Roemer offers the principle that the range of things that people of a given type actually do accurately mirrors the range of choices that are realistically open to them. So he asserts that, if it turned out that all sixty-year-old steelworkers smoked for exactly thirty years, then they could have had no option but to do so. But, in my view, this principle is flawed; correlation — even perfect correlation — among different individuals’ choices does not imply that they were in any way restricted to those choices.

Let us imagine, for example, that smoking were an important part of the “bonding” that goes on among steelworkers. A worker might take great pleasure from this camaraderie. Indeed, his pleasure might be so intense that he would choose to smoke even knowing that it jeopardizes his health. Smoking would then be a risk that he and his fellow workers deliberately take on, not one that is in any way forced on them. And yet, the observed outcome would be one in which they all behave in much the same way.

I think that this sort of problem would plague any attempt to make indirect inferences about responsibility from empirical distributions. There seems to be no substitute for making direct judgments about whether or not people’s behavior is beyond their control.

Nevertheless, to reiterate what I said at the outset, I strongly concur with Roemer’s more general thesis: it is proper to compensate people for adverse circumstances beyond their influence but to hold them responsible for the choices they make.