>To have 2 of them (that we know about) stemming from prosecutions by one individual, I would have to say, is pretty remarkable. While it could just be a coincidence, the one common denominator in these two statistically rare incidents is Stephen Heymann.

Careful. You are subtly reusing evidence in an invalid way.

The reason that Heymann's record is under examination in the first place is that Swartz committed suicide while Heymann was prosecuting him. That's the observation that generates the hypothesis "Heymann is more aggressive than the average prosecutor, and this leads to an increase in suicides". So then we test that hypothesis by looking at Heymann's record and seeing if we find more suicides.

So far so good, but what we can't do is then reuse Swartz's suicide to support that hypothesis. Appropriately enough, this is an example of the prosecutor's fallacy (http://en.wikipedia.org/wiki/Prosecutor%27s_fallacy). The key is that we have to keep an eye on the context of the original hypothesis. If someone had looked at Heymann's methods and said "Wow, that's going to lead to people killing themselves", and then looked into his record and said "Willikers! Get a load of this body count!", then they'd be right to count both Swartz' and James' suicides.

But that's not how we got here, so the question has to be "Is the number of suicidal defendants prosecuted by Heymann — other than Swartz — significantly out of line with expectations?"

And that number is one. To answer the question would require some data. I've tried gathering data from different angles, but I'm not sure it's out there. Maybe someone else can take a whack at it.

The point of the prosecutor's fallacy isn't that you should ignore evidence, it's that you must include all the evidence in calculations of probability. Swartz' suicide is inductive evidence for the hypothesis "Steve Heymann caused Jonathan James' suicide", you can't just ignore it.

I'm not saying to ignore it. I'm saying not to double-count it.

>Swartz' suicide is inductive evidence for the hypothesis "Steve Heymann caused Jonathan James' suicide",

That's actually a separate hypothesis.

Can you describe the point at which double counting is occurring? My hypothesis is, "Steve Heymann's defendants are more likely than others to commit suicide". I propose to determine with what likelihood a criminal defendant will commit suicide, determine with what likelihood Steve Heymann's defendants have historically committed suicide, and compare the two numbers. How does excluding Aaron Swartz from consideration make my conclusion better reflect reality?

I a court of law, or over a dinner with a mathematican, you are absolutely right.

In real terms -- in getting the prosecutor removed from office, in stopping these kinds of mindless overuse of federal reach, it is best to count both, and use the nickname 'Suicide Steve'.

I think it warrants investigation. If we have sufficient outside evidence to support the hypothesis, then of course go after him. But if the truth is that this is par for the course, and all prosecutors are driving defendants to suicide, then it would be very bad to crucify this one guy. Nothing stops progress like a good scapegoat.

Is it quite rare though? It would be good to see some figures on that.

It could also be sampling bias playing a part.

http://en.wikipedia.org/wiki/Sampling_bias

Two does not necessarily make a trend, to be sure. However, if the odds are 1 in 10,000, a small number of incidents quickly become a significant indicator that the way this guy handles cases dramatically increases the likelihood of suicide.

I don't know if any reports have been done on the subject that would be relevant here (specific to federal defendants, as they generally are looking at more time and are more likely to be experiencing jail for the first time than state defendants). However, anectdotally, I know a fair number of criminal defense attorneys, and they all say that it's actually quite rare. It seems that most defense attorneys have never had a criminal client commit suicide.

You assume way too much. I highly doubt that federal defendants are 10 times less likely to commit suicide than the general population in USA.

For me personally two is still waaay to low for it to be considered anything but a fluke.

The general population suicide rate is about 1 in 10,000, which is what the GP quoted. What's your source for saying it's 1 in 1000?

I apparently read his post as 1/100.000.

Two is a pretty low sample for it to be statistically significant.

Anecdotaly, a friend of mine committed suicide and I later found out he was going to jail for robbery.

As I said in the original comment, it could just be a coincidence. However, if the bodies continue to pile up over the years (or more likely, that there are far more suicides already on the guy's resume that we simply haven't heard about), that will become less and less likely.

Two is not a sample size. The sample size here is the number of defendants prosecuted.

Two events can be very significant if the prior probability of the event is very low.

It's also possible hackers tend to be more depressed. That would not surprise me considering the stereotypical hacker's isolated lifestyle. Aaron certainly had a history of depression and doesn't really strike me as stereotypical or average in any way, though.