John Roemer’s formulation of the problem of equality and responsibility is characteristically illuminating; it both focuses and advances debate about distributive justice. But I am not convinced by his proposal, for two reasons.[1]

First, Roemer’s proposal conflates issues about how responsible people are and what they are responsible for. Are different people equally responsible for the results of their choices? Even if they are, they can make different choices, with different results. This distinction is blurred by Roemer’s talk of “comparable responsibility.” But if we draw this distinction clearly it will cast doubt on Roemer’s proposal to equalize the prospects of people at the median of different groups.

People can be equally responsible for very different outcomes. Within each of Roemer’s types, people are by hypothesis equally responsible for their position. The types are defined so as to ensure that people are responsible for the differences within the type; all differences for which they are not responsible result from the factors used to distinguish one type from another. But because of this, it is arbitrary to use the median within each type to establish comparable degrees of responsibility across types. Within a type, the person at the median — for example, someone who has smoked for 10 years — is just as responsible as the person at the 80th percentile — someone who has smoked for 15 years. They simply make different choices. To the extent different results flow from those choices, they are responsible for different results.

But across types, differences of persons’ positions are due partly to choice and partly to circumstances beyond control, or brute luck. (Henceforward for short I’ll simply say “luck” for factors people are not responsible for: circumstances beyond control and brute luck.) How, then, can we correct for differences across types that are a matter of luck, while preserving differences within types, for which people are responsible? We cannot eliminate all differences across types while respecting differences within types as due to choice. So we’d need to know, when people in different types are not responsible for actual differences, what other, hypothetical differences they can be regarded as responsible for. This seems to me fundamentally indeterminate; our judgments of responsibility simply may not reach this far. Any general answers to the question, such as Roemer’s idea of equalizing median points, will therefore be arbitrary. But how well-off people are after redistribution will turn on which arbitrary resolution is adopted, and that seems unsatisfactory.

Schematically, consider A and B in Type 1 and C and D in Type 2; assume that A is better off than B, who’s better off than C, who’s better off than D. (A may be at the median for white female college professors and B at the 80th percentile, while C is at the median for black male steelworkers and D is at the 80th percentile.) By hypothesis, the difference between A and B is not a matter of luck, so should not be eliminated. Ditto for the difference between C and D. But the differences between types are a matter of luck, so require compensation. Should we compensate to equalize A and C’s positions, or B and C’s? Is C responsible for the same position as A or the same position as B?

It doesn’t help to ask how responsible these people are: we need to know what positions they are responsible for. Suppose all Type 1 persons are more responsible, because circumstances constrain their choices less, than all Type 2 persons (within type, as we’ve seen, all are equally responsible.) This supposition doesn’t answer our question: that A and B are more responsible than C and D doesn’t tell us whether to make C’s position the same as A’s or the same as B’s. We cannot do both, or we would make A’s position the same as B’s, and that would fail to respect the difference between A and B, which is due to choice. Since A and B are by hypothesis equally responsible for their respective positions, either decision seems arbitrary. There’s no reason to think C is responsible for the same position as A as opposed to B. We don’t need a correlation in degrees of responsibility: A and B are equally responsible, but simply make different choices and hence are responsible for different results.

To solve our problem, then, we need to find a reason for a judgment such as this: “A and C are responsible for the same position and B and D are responsible for the same position,” where the first position is better than the second. This particular judgment, of course, is only an example of the type of judgment that is needed to solve our problem. But if we do implement this particular judgment, we will equalize A and C and equalize B and D. Notice that this will reverse the positions of B and C: B was better off than C, and is now worse off. Presumably this difference reflects a judgment not only that it is a matter of luck for B to be better off than C but that it would not be a matter of luck if C were instead better off than B. But it is hard enough to judge that B’s actually being better off than C is a matter of luck. I see no non-arbitrary general basis for specifying in addition that a different possible state of affairs in which C was better off than B would not be a matter of luck.

When actual differences between people are due to luck, there is no reason to assume we can in general identify what the differences between them would be in the absence of luck. It is hard enough to judge whether people are or are not responsible for what they have actually done or its results. There is no reason to believe we could in general also specify, even in relative terms, different possible outcomes for which people are responsible, when they are not responsible for their actual outcomes. There may be intuitive answers to this question in some particular cases involving limited departures from actual circumstances. But any principle claiming to answer this question in relation to all actual circumstances due to luck seems bound to be arbitrary. In many cases, our real-world judgments of responsibility simply do not extend this far. Once we disconnect the idea of what people are responsible for from the consequences of their choices, given actual or close-to-actual circumstances, it is underconstrained and indeterminate. For this reason I think it is unwise to make egalitarianism hostage to applications of the concept of responsibility: the needed applications outrun the determinate content of the concept.

A feature of Roemer’s smoking and income examples disguises the points I’ve been making. Both are one-dimensional: the effects of smoking are assumed to be bad (bad health), and those of education good (more income). So we can assume that people should avoid smoking and get more education if they can. It is this feature that lends itself to the talk of comparable responsibility or degree of will power exercised — talk that blurs over the distinction I’ve been emphasizing. Within one type, the person who smokes less is no more responsible than the person who smokes more; he just makes a different, though better, choice. But we cannot in general presume that behavior at one end of a range within a type ought to be encouraged or resisted.

For example, consider the number of children people have within two different types. Suppose the median for Type 1 is one child, while the median for Type 2 is three children. We cannot assume that having more children is either to be encouraged or resisted; it is a matter of free choice within natural constraints, and people within each type are equally responsible for the results of their different choices. But it is evident that occupying the median provides no reason to suppose that Type 1 persons with one child should be compensated to the same extent as Type 2 persons with three children. Occupying the median within a type is mere happenstance. But when we see this point clearly in this case, we can see that it also applies in the one-dimensional cases (for the reasons I explained above).

My second objection concerns Roemer’s claim that his view is not committed to any position about what factors are beyond persons’ control; each society is left to decide that for itself.[2] This disclaimer notwithstanding, Roemer’s view does seem to be committed to a controversial assumption about responsibility. He appears to endorse a version of what I call the regression principle: that in order to be responsible for something, a person must be responsible for its causes (and so in turn responsible for the causes of those causes, and so on). This controversial assumption arguably makes responsibility impossible, whether or not determinism holds: nothing could possibly be responsible for the causes of anything, all the way back. Similar points apply to control: if control of something required control of its causes, control would be impossible too. For this reason, building a regression requirement into the idea of equal opportunity may make the idea look a Trojan Horse to friends of responsibility. To eliminate this threat, we’d have to admit that responsibility for certain results does not require responsibility for causes, all the way back. But this may be what libertarians, among others, are urging.

This admission is not, however, incompatible with Roemer’s equality-of-opportunity procedure (despite Roemer’s comments on the libertarian view): it simply alters the view of responsibility a given society might apply in order to determine the types it would use to implement Roemer’s procedure. Indeed, to get any scope at all for differences for which people are responsible, and hence for Roemer’s notion of variation within a type, it is arguably necessary to reject the regression principle, since it would make responsibility impossible. Alternatively, we can try to develop a version of egalitarianism that is not tied to ideas of responsibility and luck and hence avoids these difficulties. I think this is the better alternative, but that is a topic for further discussion.[4]