For the stats text, I’ve been trying to think of/find rules that apply across a wide array of disciplines. Here’s one: control for cyclical effects. It applies here (summer gas spikes):

And here, with voting in elections — presidential elections (a cyclical event) boosts turnout every second congressional election):

 2008:  a bit over 340,000 come out to vote:

 And here is 2010:

The left-most column, incidentally, is Republicans, followed by Democrats in the second column. So did Democratic support collapse, dropping by nearly 50%? Of course not. It’s very difficult to make any comparison of a mid-term to a presidential cycle election, because in mid-terms different sorts of people tend to turn out. In general, midterms tend to pull out more anti-incumbency voters than pro-incumbency, and in this case the incumbency in 2010 was Democratic.

If you want to use the mid-term results to predict the general election you are going to have to come up with some model to account for the difference. Maybe you could look at subpopulations or some other measure. But the point is, like the gas prices, you have to control for cyclical effects.

I’d love to hear about additional examples of cyclical effects from your academic disciplines or the jobs you work. Shoot me an email, or post to twitter, facebook, google…