Although Marquis' article is much broader than the statistical argument that bothers Feige, Marquis' basic point on the numbers is this: a study on exonerations claimed to find 340 exonerations nationwide druing a 15-year period. Marquis grants that and raises the number by a factor of 10 for purposes of argument: hence, he assumes the real number of wrongful convictions of actually innocent defendants is 4,000. Given the 15 million convictions obtained during the time frame of the study, the resulting error rate is 0.027%, or stated differently, the percentage of correct convictions is 99.973%
Feige claims that Marquis is disingenous because he uses the incorrect comparison. According to Feige, what Marquis should compare is the number of exonerations to the number of cases where guilt was actually contested. In the other cases, guilt was not at issue, so those cases should be excluded from the analysis. Feige goes on to argue that of the 15 million cases referenced by Marquis, 14.25 million were pleas, not contested trials.
Using the 750,000 figure (contested trials where guilt was at issue), instead of the 15 million figure of all convictions, Feige reasons that the error rate becomes not 3 in 10,000, but 50 in 10,000.
Feige also complains that the exonerations usually only come in murder or rape cases, not in the far more routine cases of drug possession or distribution, bad check cases, or the like.
A couple of things jump out immediately: first, the exoneration study cited by Marquis found exoneration for all types of felonies, not just murder and rape. Secondly, Marquis quite generously inflated the number of exonerations by a factor of ten to account for this potential problem, even though there is really no empirical evidence whatsoever that the rate of exoneration would even come close to Marquis' assumed level. So Marquis' generous assumptions about the true rate of exoneration (a number, after all, that can never be known with precision) bolster the strength of his argument.
Moreover, Feige's complaint that the total number of 15 million convictions is not the correct figure to use for comparison is faulty. If the question at hand is: "how well does the justice system identify and hold accountable the guilty," then in fact we want to know of all convictions, how many are untrustworthy. The guilty pleas are relevent to answering that question, because they tell us that the system has correctly identified and held accountable the guilty party.
If however the question is "how often do factually innocent people who contest their guilt or innocence get convicted," then Feige's adjusted numbers are the relevant ones.
So what do we conclude from this exercize? Even if we accept Feige's numbers as he posits them, here is the percentage of wrongful convictions: