Thursday, February 7, 2008

How Well Did the Polls Predict Democratic Delegate Allocations on Super Tuesday?

UPDATED on 2/10

I've been referring to the Democratic Convention Watch, the New York Times, and various state party websites for the latest state-by-state estimates of how the candidates fared in terms of delegates on Super Tuesday. We still do not know have full final delegate allocation estimates in CO, GA, and IL. However, we do appear to have a good idea of the delegate allocations in 14 states for which we had polling data before Super Tuesday.

On Monday (2/4), I used polling data from to generate estimates on how the delegates would be allocated in 17 states. The basic point was that it really didn't matter much on the Democratic side who won or lost states since proporational allocation of delegates meant that the percentage of the vote each candidate received was most important. The assumption I was making was that if the polls were right, we could get a pretty good estimate of how many delegates each candidate would receive from each state simply by assuming they would receive delegates proportional to their vote in the state. Of course, some delegates are apportioned by congressional district, which does complicate things. But I was assuming that it would mostly even out so that the statewide polling data would be a good indicator.

While we still don't have information on all of the states, it appears as though the pre-election polls did a good job of predicting delegate allocations in every state except one (MN, which I will discuss later). The polls appear to have been off by just one delegate in AZ, MO, NJ, TN and UT, by two delegates in AL, CT and DE, and by 3 delegates in MA, NY, and OK. If you go by the numbers in the New York Times, the poll-based estimates were off by 6 delegates there (out of 370 delegates up for grabs in that state). This is not a bad showing for the pollsters, particularly given how limited the polling was in several states leading up to Super Tuesday.

The one state where the estimates were way off was MN (underestimating Obama's margin by 15 delegates). I think there are two reasons for this error. First, rather than relying on the average as I did with most other states, there was just one survey in MN leading up to the election. Furthermore, that survey was conducted from 1/18-1/27, which means most interviews were conducted prior to Obama's win in SC. Second, MN held caucuses and polling in caucus states is notoriously problematic.

The chart below shows my Monday estimates for each state, compared to how each candidate actually fared. The "state error" column shows by how many delegates the estimates were off. I have not produced numbers in this column for CO, GA, and IL chart since we don't have full delegate estimates for those states.

Also, keep in mind that I did not make estimates in states where there was no polling, though I do include the actual totals in the table.

I'll update this post as I get updated numbers and see how the estimates fared in the remaining states. However, I think it is important to note that the polls (particularly the averages) did a fairly good job of predicting final delegate allocations. So far they did seem to overstate Clinton's support slightly. The poll-based estimates predicted that she would win 52.4% of the delegates available in those 17 states; but so far, she has won only 50.6%. Nevertheless, the bottom line behind my estimates was that neither candidate was likely to come out of Super Tuesday with a significant lead in delegates, something that can clearly be affirmed by the outcome of the contests. Thus, the polling largely provided a useful way to predict eventual delegate allocations.

No comments: