Tuesday, April 22, 2008

Tuesday Statisticz: The Dude would have voted Libertarian

"Smokey, this not 'Nam. This is bowling. There are rules." - The Big Lebowski

I've always associated bowling with Miller Lite, cigarettes, and the 50s. And when I visualize this picture, I can't help but think that everybody who goes bowling also votes Republican. It's like candy corn and carnivals. But is there really something about throwing a ball down a lane that makes you want to give your soul to the devil? Just kidding, but seriously, that's what this Tuesday Statisticz is set to find out.

I figured that the best way to do it would be to compare states. So I found the number of bowling alleys in each state (here), and the number of people that voted for Bush compared to Kerry for each state in 2004 (here). First, here's the states with the most bowling alleys per capita:

There are 4 levels here: the states with less than 1.5 bowling alleys per 100,000 people (like California) are clear, the states with 1.5 to 3 bowling alleys per 100,000 (like Maine) are grey, the states with 3 to 4.5 bowling alleys per 100,000 (like Wisconsin) are dark grey, and the state with more than 4.5 bowling alleys per 100,000 people (like North Dakota) are dark green. I know it's hard to tell the difference between dark grey and grey, but squint for me, this map was hard to make and I can't figure out how to fix it.

Next, we have the presidential voting by state in 2004. We've all seen this graph more than we've seen Jamie-Lynn Spears's boyfriend, so I tried to spice it up a little bit:

There are technically 5 levels here: dark blue states that voted more than 57% for Kerry (NY, a few other in the Northeast), light blue for states that voted between 57 and 52% for Kerry (ie, Hawaii), white for states that were "toss-ups" between 48-52%, darker red for states that voted more than 52% for Bush (like Florida), and light red for states that went predominantly for Bush (like Texas). I know that it is hard to tell the difference between dark blue and light blue, but I wanted somehow to differentiate between states that really went for a candidate and states where the populace didn't have a serious preference.

Looking at the maps, do you think that there is a correlation? Think of your answer now, I don't want any hindsight bias on my conscience.

Here's the scatterplot for votes versus bowling alleys, along with a best fit line:

It's tempting to look at the r squared value and assume that there is a pretty good correlation. But we can do better than that, and use an actual measure of significance. I found that the result was indeed significant (p < .01, t = 3, two tails, df = 48). The effect size at 0.148 isn't humongous, but it is almost certainly nonrandom. The amount of bowling alleys per capita in a state does account for a positive correlation with the number of people that will vote Republican in that state.

There are a lot of different possible third variables that could be at play here, including average temperature (notice how the bowling alleys are mostly in the North?) and distance from a major ocean (where there may be other distractions). But somebody else will have to do those regressions. I'm off to drink a White Russian.

(Thanks to Andy Eggers for providing the R code for these maps.)