Wednesday, November 7, 2012

Improving election night one graph at a time

I watched the election last night mostly on CNN. We flipped around a bit, but we mostly watched John King in front of his magic wall, zooming in and out of election maps.

King did a good job. He clearly knows the political geography of the United States, especially in the swing states. As he showed the results in one or another county, he would toggle the map from 2012 to 2008 or 2004, so as to put the current results in context. And as returns were reported in Ohio, Florida and Virginia, he would note where the votes yet to be counted were coming from, and how those counties tended to vote. It was informative stuff.

It could have been better.

Maps are a nice visual. But they are not meant to make comparisons over time. Counties can be toggled from red to blue, but that washes away all the information about how red, or how blue. And anyway, winning a county doesn't matter. Winning states does.

The natural way to make comparisons is with a scatterplot. Like this one:

The figure shows county-level vote in Florida in 2012 and 2008. Current data was taken from CNN's website. They are not official, but CNN reports 97% of the state's vote has been counted. Last cycle's data from the Precinct-Level Election Data Project. The diagonal line shows the scenario in which the vote shares in 2012 are the same as in 2008. This is not generally the case. Romney did better than McCain against Obama, as we would have expected him to. This was a closer race. 

But not everywhere. 

Obama actually appears to have done better in Miami-Dade in 2012. That's important. As the vote was coming in, Miami-Dade (and Broward) were counties where the vote was not completely in. Projections that Obama would do well were based in part on the fact that we would expect this area to favor Obama. And yet, if Romney had done as well there as he had in other counties (relative to 2008), he might well have won Florida. Meanwhile, it's interesting to ask why Romney didn't have a similar improvement vs 2008 in Miami-Dade.

The other problem with maps is that they show territory, not voters. But land does not vote for the president, people do. We can distort maps to be representative of the population, but you can also represent size on a scatterplot. Here, the dots are weighted by the total votes cast in 2012:

And you can see why Obama wins, even though so many counties would have been colored in red on the map. The larger counties are where Democrats do better. Several times last night, John King had to remind viewers that the vast areas of red had far fewer voters in them than the small pockets of blue.

Maybe I'm being overly optimistic about how well a television audience could understand a scatterplot. We're starting to see better data presentation in print and online, but television might be a studio too far. Still, most of election night, especially early on, is spent filling airtime, waiting for news to report. Surely some pretty scatterplots could add some variety. 

So, OK networks, you have two years to try to pull this off. I'm pretty sure John King could handle this. So could newscasters at other networks. I think he was trying to tell us things that scatterplots would have made easier. Next time, maybe give him better tools to work with. 


  1. Yes scattergrams with size-weighted dots! But rather than directly plot 2008 against 2012, I'd rather plot 2008 against swing. That way you can exaggerate the Y axis and see how things are changing - and draw a horizontal line for what the overall swing has to be for the state to change hands.

  2. Swing against 2008 would be an improvement. The simple scatterplot is a powerful tool, and one that is simple to interpret. But plotting swing would be perhaps better in this case.

    The horizontal line would probably not work, because it would depend on the size of the counties. A small swing in Miami-Dade would have more consequences than a larger swing elsewhere.