Darn that Lindsey Graham! (or, “Mr. P Predicts the Kagan vote”)

On the basis of two papers and because it is completely obvious, we (meaning me, Justin, and John) predict that Elena Kagan will get confirmed to be an Associate Justice of the Supreme Court. But we also want to see how close we can come to predicting the votes for and against.

We actually have two sets of predictions, both using the MRP technique discussed previously on this blog. The first is based on our recent paper in the Journal of Politics showing that support for the nominee in a senator’s home state plays a striking role in whether she or he votes to confirm the nominee. The second is based on a new working paper extending “basic” MRP to show that senators respond far more to their co-partisans than the median voter in their home states. Usually, our vote “predictions” do not differ much, but there is a group of senators who are predicted to vote yes for Kagan with a probability around 50% and the two sets of predictions thus differ for Kagan more than usual.

The other key factors that enter into the models (which build on the work of Cameron, Segal, Songer, Epstein, Lindstadt, Segal, and Westerland) are senator and nominee ideology, party and partisan control, presidential approval, nominee quality, and nomination timing.

The bottom line? The older model predicts nine Republican defections (votes for Kagan) but the newer model breaking down opinion by party predicts only five. Ten Republicans straddle or push against the 50% mark for point predictions.

Median state-level support for Kagan is approximately that for Alito, and about nine points higher than that for Sotomayor. Median state-level support for Kagan among Republicans is about 12 points higher than for Sotomayor. On the other hand, Obama’s approval is definitely lower. So far, we have only one national poll to work with (which we thank ABC for), but we will update our data and “predictions” later when other poll data become available. We do not yet have Jeff Segal’s official scores for quality and ideology so are currently fudging these a bit (using the same scores as for Sotomayor).

First, here is the distribution of opinion across states by party groups (using an extension to the MRP technique to generate not only opinion by state using national polls, but opinion by party by state):

View image


Next, here are the predicted probabilities of a positive confirmation vote for Republican senators (Democrats are all predicted to vote yes):

View image


If only Lindsey Graham would do the right thing and vote no…

9 thoughts on “Darn that Lindsey Graham! (or, “Mr. P Predicts the Kagan vote”)

  1. But, just looking at the graph, shouldn't about 2 of the ten senators between, say, Hatch and Cochran vote yes? (Graham is one of those two) If all of them voted no, wouldn't that be evidence against your model?

  2. Interesting point. It depends on whether one assesses the prediction for Graham specifically — in which case the model is supported more by a no vote than a yes — or whether you assess the predicted probabilities across a set of senators… in which case we would indeed expect 2 of 10 senators (each with a 20% chance of voting yes) to vote yes… So if all senators below 50% vote no… is that support for the model or against? There must be a name for this sort of prediction problem… Andy?

  3. It's called calibration, and yes, you'd expect to see 20% of the "20%" cases to vote no. On the other hand, with p=.2 and n=10, it's not so unexpected to happen to see y=0.

  4. Sure… mild evidence, but evidence nonetheless. (And I think you meant to say vote "no," in the comment above. And, by the way, according to newspaper reports today Ben Nelson says he'll vote no. If the Democrats were all at 100 percent, this is a huge (infinite?) error in logit space and, I suspect, a pretty big one in whatever space you're using.

  5. It's also a form of multiple-comparisons problem. If you were just looking at Graham ex ante, his failure to follow the prediction might be interesting, and perhaps statistically significant. If you look at all the Republican senators with a probability below 50% of voting yes and then focus on the small number (perhaps one) with a result different from the prediction, that is much less likely to be statistically significant or meaningful.

  6. Nebraska Is All That Counts for a Party-Bucking Nelson
    Dem Senator On Blowback From His Opposition To Kagan: 'Are They From Nebraska? Then I Don't Care'

    Fine, but 62% of Nebraskans with an opinion favor confirmation… 91% of Democrats, 39% of Republicans, and 61% of Independents. So I guess he only cares about Republican Nebraskans…

  7. Also, from a June 11 ABC poll, so perhaps things have shifted. We don't have more recent raw data to play with yet…

Comments are closed.