Wednesday, 25 July 2012

Causality vs correlation in military domestic violence

Does this get you pregnant?
Sorry if that title sounds a bit like an obscure scientific paper, but there's an important point to make. I was listening last night to a radio documentary about domestic violence among military personnel and it made the most fundamental scientific blunder. It missed a phrase that should be tattooed on the hand of every broadcaster: 'Do not confuse correlation and causation.'

Let me take a step back with an example that was used on my Operational Research masters course. For a good few years after the war, the pregnancy rate in the UK had a strong correlation with the import of bananas. When more bananas were imported there were more pregnancies. Fewer bananas, fewer pregnancies.

The amusing response is that the bananas were causing the pregnancies. But to accept that at face value is to miss two other possibilities. One is simple reversal. The pregnancies could be causing the bananas. By this I mean that there could be a causal connection between someone being pregnant and increased consumption of bananas. Maybe pregnant women crave bananas. Or maybe young children (a common  result of pregnancy) eat more bananas than older people.

The second, and much wider, option to explain the apparent link is that there is a third factor that causes both the increase in pregnancies and the increase in bananas. Perhaps there was more money around and this caused both. Or one of many other potential third factors.

I am not suggesting any of these alternative causal processes is correct, but that it would be absolutely stupid to make the initial assumption that because banana imports went up and pregnancies went up, eating bananas make you pregnant.

Now let's go back to that radio programme. Because it made just such a stupid assumption. Let's be clear again - I'm not saying what the causal link is, merely pointing out the unscientific way in which correlation was turned into a particular causality.

The topic of the radio programme was essentially that, despite denials from the MOD, there was more likelihood of domestic violence in an army household than a civilian one. The big, bad assumption was that being in the army (and the experiences you had there) made you a more violent person. There was no attempt whatsoever to look at the two alternative causalities. What if being a more violent than average person made you more likely to join the army? Was the causality the reverse of the one they assumed? For that matter could there be a third factor that caused people join the army and to be more violent?

This was sloppy journalism and bad science.

No comments:

Post a Comment