Last time I talked about an apparent *positive* relationship between state unemployment rates and state support for President Obama in political polls. This despite the conventional wisdom that Americans vote with their wallets. That would suggest that states with higher unemployment rates should be *less* likely to support an incumbent President.

I have to be honest. I started this thinking that the positive relationship was just a fluke. That it was really just a sort of anomaly that would go away if I included the right other variables. So I did what any good data analyst would do in the situation. I played with my data. That is to say, I added other variables, tried this and that, checked correlations here, ran regressions there.

And it didn’t turn out the way I expected.

***

Again, my Economical Oath requires me to point out in advance that there are assorted technical issues with approaching this problem this way. But that doesn’t mean we can’t learn some interesting things from it. Virtually every economic analysis has technical problems with it; it’s mostly a matter of degree.

“Regression” is an intimidating term to many people. It sounds all complicated and statistical. All it really means is trying to find out how a some numbers affect other numbers. Practically speaking, it’s just using a computer to do calculations – you can even do it in Excel.

In addition to correlations, I used simple regression analysis to look at the relationships here. In the end, you come out with an equation that represents the relationships, along with numbers that tell you how good the relationships are, statistically-speaking.

***

Bbefore I tell you about the relationships that I found, let me first tell you about the variables that *didn’t* work.

First, the percentage of the states’ populations that are black. Virtually no relationship to speak of – a correlation of -0.002. Likewise percentage Latino had a correlation of 0.18, but this went away after taking account of other factors.

Percentage urban versus rural, percentage with a Bachelor’s degree, population, population density, even acres of farmland – all of these have some correlation with support for Obama, but all of these apparent relationships proved small or insignificant when other variables were included.

***

Including the state median family income looked really good at first. And in fact it made the state unemployment rate look even more important, statistically. A $1000 increase in a state’s median family income was associated with a 0.5 percent increase in support for Obama.

But it turns out that family income was just capturing something else: the effects of advanced college degrees. States with a larger percentage of people with masters and doctor’s degrees have higher family incomes. And it’s those advanced degrees that really are associated with more support for Obama.

Education really does make you liberal. Or, properly speaking, appears to make you more likely to support the more liberal candidate. In this case, anyhow.

***

So shut up and cut to the chase. What’s the bottom line?

After looking at more than a dozen different variables, here’s the equation that turned out to explain the most differences in state support for Obama in the best statistical way:

By “religiosity”, I mean the percentage of the state population who say that religion is very important in their lives. I used state data from 2007 Pew Research survey for this. My thinking was that many people base their political views on their religious faith, and that this might explain some of the differences in support for Obama.

And it does. For every 5 percent increase in state religiosity, support for Obama decreases by 1 percent, and that relationship is statistically significant. In other words it isn’t just be caused by random factors.

For the curious, the R-squared for this equation was 0.54, meaning these variables explain 54 percent of the differences in state support for Obama — not too shabby for a back-of-the-cocktail-napkin analysis. All three variables were significant at the 0.05 level.

***

In the end, including percentage with advanced degrees and religiosity in the equation made the relationship between state unemployment rate and support for Obama look even *better* than it did without them. In fact, no matter what variables I included, there still was a positive relationship between unemployment and Obama support.

So the relationship that I didn’t think existed, the one that I thought would go away if I included the right variables, turns out to be pretty solid. Higher unemployment rates do seem to be associated with more support for President Obama.

Maybe Americans do vote with the wallets, only not the way you might think.