Accuracy of Potomac Primary Delegate Estimates and a Wisconsin Delegate Prediction

We have final delegate estimates in from Virginia and DC, but we are still waiting for final counts from Maryland. Nevertheless, it is certainly not too early to say that the polls underestimated how many delegates Obama would pick up in the Potomac Primary. On Monday, I used pre-election polling to estimate that Obama would have a net gain of about 40 delegates from the Potomac Primary. However, the polls appeared to underestimate Obama's support in each state, which led to an underestimation of the number of delegates he would win.

In Virginia, I predicted that Obama would win 50 delegates to Clinton's 33. In fact, Obama won 54 delegates and Clinton won just 29. Thus, the estimate was off by 4 delegates in that state. In DC, the estimate was that Obama would win 10.5 delegates to Clinton's 4.5. I rounded both up, but that meant I was giving out one more delegate than DC had (oops!). Nevertheless, even by doing that, I was still off by just one delegate for Obama's total, and 2 from Clinton's total.

What I was really interested in seeing, however, was how much better Constituent Dynamics did compared to me. As some of you know, Constituent Dynamics used automated polling techniques to provide some great polling information across 60 congressional races in 2006. Leading up to Tuesday, the organization conducted 14,276 interviews in MD, VA, and DC for the Potomac Primaries. This gave them enough interviews to make predictions by congressional district, which is the level at which a significant share of the delegates are allocated. My method, on the other hand, simply allocates delegates by the statewide polling numbers and assumes district variations will even out when aggregated.

How much better did they do compared to my estimates? Well, they didn't do better at all. In fact, they did slightly worse. In Virginia, they predicted 45 delegates for Obama, 32 for Clinton, and called 6 too close to call. If you divide the too close to call delegates evenly between Obama and Clinton, then my estimates based just on the statewide vote were closer to the final totals than theirs. The only way their estimates come closer than mine is if you assume that all 6 too close to call delegates when to Obama. In DC, their estimates were essentially the same as mine. (Once Maryland's delegates are tallied, I will update this post to tell you how we both did in that state, but I think I'm going to be closer there as well.)

Thus, the Potomac primaries provided for an interesting (though limited) test of whether polling by congressional district allows for a more accurate prediction of delegate allocations compared to having just statewide data. Based on this limited test, statewide polls fared at least as well in predicting delegate allocations as polling conducted by congressional district (and just polling the state is less expensive). This does seem to provide support for the assumption that I make when I use these surveys to predict the Democratic delegate estimates--that is, that variation across congressional districts will even out when delegate shares are aggregated to state totals.

Finally, I've seen no polling from Hawaii, but in Wisconsin there have been several surveys, so I'll use the Pollster.com average to make the delegate prediction. Based on the current average available on Pollster.com, Obama should win 40 delegates on Tuesday compared to 34 for Clinton. While that is the estimate, it will be interesting to see if Clinton can pull a New Hampshire type of upset or if Obama wins by bigger than expected margins once again. Thoughts on this?

(I'll have a new Superdelegate Prediction up in the next few days, with updated data and information on how we've done with our predictions so far.)

UPDATE: Pollster.com added another Wisconsin survey to their average last night which shifted the numbers slightly. I'm now esimating 40 delegates for Obama and 34 for Clinton.