Betwetting

[Carnie]

This was prompted by strip #658, but I'm putting it here because it really has nothing to do with the events or subject matter of the comic. Mods can feel free to move it if they disagree.

Bedwetting correlates to crime.

Wait, what?
I was under the impression that bedwetting was a physical/mental maturation delay amoung children and the occasional teen that affects how the body's natural processes function during sleep. Why in the world would that correlate with criminal activity? This makes about as much sense to me as the statement "Cooties correlates to crime", in that it is something young children get teased for and which in fact betrays precisely nothing about the person in question.

The Wikipedia article in question does not shed any light on this; it simply cites a $50 ebook which I've little interest in purchasing just to answer this question.

So... does this make any amount of sense to anybody?

[JustinReilly]

He's probably referring to the Macdonald Triad of behaviors (animal cruelty, fire-starting, and persistent bedwetting after age 5) that supposedly predict violent psychopathic behavior. The theory hasn't seen much support in the intervening decades (especially the bedwetting component), but it was once commonly taught and reached widespread popular exposure in the media.

[Flint_A]

I'd also like to point out that lots of things correlate to lots of things.

Ice cream sales correlate to crime, because both of those correlate to rises in temperature.

Unless someone's talking about causality, you don't really need to take such things too seriously. Bedwetting correlates to crime? Sure, it might. How strong is the correlation? It's obviously not 1.00, literally every bedwetter does not become a serial killer. We don't have the numbers. Even if we did have the numbers and there was a serious correlation, we still wouldn't really know what was going on.

I bet that if we went out and collected data, we would find a small but positive correlation between name length(in letters) and crime. That would not actually be useful information.

[DanielH]

Why would you expect a positive correlation between name length and crime? I would expect no correlation (beyond statistical error), and given that a correlation existed I would actually expect it to be negative (because my mind’s stereotype of a criminal has a short name like “Joe”, not for any good reason).

[Flint_A]

Asian names tend to be shorter; Latin, African and Middle Eastern names tend to be longer. I was also thinking of "full name" rather than "first name"(although the previous sentence kind of works either way) which warns us against unclear constructs in research.

[snowyowl]

Still, that's a bit of a stretch.

[Flint_A]

It certainly is. Hence it being an example of unreliable statistics. I could very well pick a sample and find results that support this silly hypothesis, that's my point. If I only tell you that a study found some correlation, that should tell you ABSOLUTELY NOTHING without looking at how the study was conducted.

[snowyowl]

You're also using a very loose definition of "there's a correlation". I could grab some statistics about... I dunno, number of Facebook direct messages sent in a given month versus emperor penguin populations, and it's unlikely that I'd get a correlation of exactly 0.00. It would be slightly positive or negative - but too small to be statistically significant.

I don't think you could get a statistically significant correlation between name length and crime.

[JustinReilly]

You're also using a very loose definition of "there's a correlation". I could grab some statistics about... I dunno, number of Facebook direct messages sent in a given month versus emperor penguin populations, and it's unlikely that I'd get a correlation of exactly 0.00. It would be slightly positive or negative - but too small to be statistically significant.

I don't think you could get a statistically significant correlation between name length and crime.

Pshaw! You trawl a large enough data set you can get p < 0.02 for just about any damn thing.

[Flint_A]

Exactly.

Yeah, pretty much nothing has a correlation of exactly 0. But we have a saying in the social sciences: "If you torture your data enough, they will confess."

(Well, I usually hear it as "it will confess", but "data" is plural damn it.)