The Phantom Chasm

April 5, 2008
Posted by Jay Livingston
An article by David Sirota in In These Times (a solidly southpaw monthly) has been getting a lot of attention lately, mostly for this graph.

Sirota says, “when you chart Obama’s margin of victory or defeat against the percentage of African-Americans living in that state, a striking U trend emerges.” Sirota calls it “the race chasm.”

Now Brendan Nyhan has offered a much more detailed look at this question using more refined data. His blog offers a critique that might serve as a unit in a methods course. For one thing, if you look closely, you’ll see that the X-axis plots the states according to rank order on percent-black. If you use the actual percentage, the U-shape becomes much less U-ish.

Second, Sirota’s graph makes it tempting to talk about “white voters” in these states. But as I hope my students remember, to use state-level data to draw conclusions about individuals is to commit the ecological fallacy. So Nyhan uses exit polls to estimate the percentages in each state of whites voting for Obama. The scatterplot is not U-shaped at all. In fact, a straight regression line yields a correlation of -.53.

Not a U-shape at all, but a straight line: the greater the percentage of blacks in a state, the less support Obama gets from that state’s white voters. Nyhan has much more analysis and more graphs. You can find them all here.

No comments: