Yeah, you're right! So I guess there are maybe three (or four) data sets? 1. the actual truth (rape or not rape) 2. what a prosecutor decides to prosecute

Yeah, you're right! So I guess there are maybe three (or four) data sets? 1. the actual truth (rape or not rape) 2. what a prosecutor decides to prosecute (reasonably provable case versus not provable case) 3. what is claimed by the accuser (this is always rape, else there is no accusation) 4. what is not claimed by a could-be accuser (a person who gets raped but doesn't report it) In order to get a conviction you need someone from set 3 and set 2. That completely neglects set 4 and should have some correlation with set 1, but this correlation probably isn't going to be amazing because there are both false positives (people who claim to have been raped, but were not) and false negatives (rapes which did happen, but that go unreported or unprosecuted). What really sucks about everything surrounding this particular category of crime is that unlike many other crimes, there's a lot of possibility (not necessarily probability though!) for false accusations. I don't honestly know what the underlying probabilities are, and in fact they might well be unknowable. As an engineer the idea that you can't even measure the error to figure out how well the system is performing at its job is deeply troubling. I certainly HOPE that the error rate is low and that innocents are being jailed and criminals aren't walking freely. But keeping up with the news certainly causes me to have some doubt that the error rate is truly low. I'm not saying that there's a pandemic of false accusations or a pandemic of unprosecuted criminals. But it's hard to hear about a few high profile accusations gone wrong (Duke, mattress girl, etc) and not wonder. Anyhow thanks for pointing out the contradiction in my post!