Poor Mr. Shonubi! He went to prison for quite some time. But -- just to keep the record straight -- Judge Jack B. Weinstein did not propose to refer to a statistical reference class to determine whether or not Mr. Shonubi committed a crime. Judge Weinstein proposed to use statistics about other smugglers to determine the appropriate punishment (under the federal sentencing guidelines) for Mr. Shonubi. Mr. Shonubi had already been found guilty of narcotics trafficking. For details see P. Tillers A Statistical Oddity.
But the question Mr. Colyvan asks is, in truth, not inappropriate -- if it is rephrased. One might put the question Colyvan asks this way: Should people's membership in reference classes -- particularly involuntary membership -- affect the allocation or incidence of social burdens, particularly severe burdens such as those imposed by the criminal justice system?
To avoid reducing this question to a technicality (i.e., to a question about statistics and statistical inference), one can restate the riddle this way: Is it just, appropriate, or necessary to allocate social burdens (e.g., criminal punishment)on the basis of inherited (or, perhaps, more simply, "non-criminal") attributes such as race, class, gender, height, and national origin?
The question so stated obviously has a connection with current controversies about racial "profiling." So the question about the criminality of belonging to a reference class is related to controversies about matters such as DWB -- the "crime" of driving while Black.
While I certainly do not propose to criminalize being Black (or, for that matter, being Latvian), driving by Black people, etc., I do wish to note that the question of whether it is possible to draw inferences about individuals without burdening individuals with adverse inferences that result at least in part from such individuals' inherited (or, just, non-criminal) attributes is both an old question and a hard one. See, e.g., Adrian Zuckerman's PRINCIPLES OF CRIMINAL EVIDENCE (1989), which both identified this general question and addressed it directly and at length.
Zuckerman's answer: it is improper to burden individuals with inferences that arise from individuals' membership in reference classes, and it is possible to avoid this evil.
I am not persuaded that it is possible, even in the "practical" activity of inference, to avoid saddling people with disabilities resulting from the situations that individuals happen to find themselves in, through no fault of their own; I believe even in this sphere there is more than a grain of truth in the old adage that life is unfair.
I suspect that the correct general answer to this puzzle -- the puzzle of the extent to which society can burden individuals for conditions that they are not responsible for creating -- must have a Kantian aroma: principles are principles if and only if we ("society") are prepared to pay a price for adhering to principles, and this benchmark of morality and justice applies to the social activity of social inference as well as to other social activities -- and thus it is the case that sometimes we (society) will & should forego certain inferences -- simply because in some circumstances we (society) don't like the social consequences that ensue if we draw some inferences on the basis of matters such race, age, height, weight, wealth, and so on.
It is of course true that many inferences that rest on matters such as race, age, and so on, are just rotten inferences, epistemically-unwarranted inferences -- but the potential of such weakness, while it must always be carefully considered (since many inferences really are just inferentially lousy), the evil of socially unfair or unjust inferences -- e.g., methods of social inference & proof that result in unfair or unjust allocations of benefits and burdens -- is different from the evil that can occur when untrue inferences are drawn. (But it is also the case that it is hard -- it is intrinsically hard -- to maintain a separation in inference between questions of truth and questions of value: we are generally happier if the inferences that generate consequences that are distasteful to us also happen to be false. [There is much more to be said about the relationship between preferences and truth in uncertain inference -- but I have learned -- but only recently! -- that it is never possible to say everything at once. So, Gentle Reader, forgive me!])
I should not end this post without noting that Colyvan's principal focus is on epistemological, inferential, and statistical questions rather than on questions of morality, legality, and justice. But the title of Colyvan's paper invites ruminations of the sort that I have recorded here. (Perhaps Colyvan's comments deserve a more extended response? What say you, dear Reader?)
