Wednesday, May 20, 2009

If the colors were red and blue …

… would this look like an election map?

(via Map Scroll)

(click to enlarge)

That appears to be well correlated. I'm not Nate Silver (UPDATE: He went in to way more analysis than me) and it is late, so I'll leave the analysis for another time.

Edit: So I wrote some cursory analysis in a comment elsewhere, I figured I'd post it here:

More specifically, states where Obama far outperformed what the HDI would predict were New Mexico, Nevada, Maine and Oregon. He also did in Vermont and Hawaii, but mainly because of his strength there. In addition, amongst the states clustered at the top of the HDI scale, there are two clusters, one where he outperformed (60% of the vote or more) and where he performed around predicted (50-55% of the vote).

This is mainly because there is little variation amongst the HDI amongst high-scoring states. The top 20 states, which are above the US average of .950 (these are larger states, which is why only 20 are above) range from .950 to .962. There is a lot more variation lower down; while the top 20 states fall in to .012, the bottom 30 span .151.

The states where Obama underperformed what the HDI would predict were Alaska, Idaho, Wyoming and Utah (note that he underperformed here by far more than he outperformed in any state). It is the southern states which are clustered with both low HDIs and low Obama vote totals.


  1. That's a pretty low coefficient of determination, and is it statistically significant?

  2. It would be interesting to see a multiple group ANOVA in this case.

    We can see that if we divide states into red and blue (by drawing a horizontal line at the .500 mark), then made a box plot, the difference would be obvious.