South Carolina Prediction Time

Time for me to get out my SC prediction so that later on tonight I can come back and post about how and why I got it wrong.

There have been 12 publicly available polls taken in the week prior to today. For the vast hordes of new readers, let me quickly recap my feelings about polling. Polls are excellent tools for generating a snapshot of how the electorate feels at any given time. They are somewhat less useful for predicting results, especially in a close, volatile race with lots of undecideds. Moreover, when each poll is considered all by itself, without regard to other polls that were taken in the same general timeframe, then we are further reducing their predictive power. Each poll has its own margin of error (usually about 3.5% to 4.5% depending on the size of the sample), so that introduces another element of uncertainty.

I think it is a big mistake for the media to report on new polls as if they indicate real movement in the electorate. In most cases, that "movement" is too small to be statistically significant, which means that we can't say with any certainty that candidate A really has gone up over the last time we polled.

But, if all of the polls that were taken in a similar time-frame are combined into one mega poll, we can get a much better sense of what's going on and be more confident in our results.

Getting back to South Carolina, these twelve polls range in their results. One has Clinton as low as 24, another has her as high as 36. Those kind of disparate results are what make people throw up their hands in disgust when they hear about polling. But, if we take all twelve polls, and instead of thinking about them as individual samples, we think of them as one big 6 day sample with 7939 people polled, then we get these results:

Obama - 41.6%
Clinton - 27.3%
Edwards - 18.2%
Undecided - 12.9%

Before we stop right there and say, "that's how the SC electorate feels," let's remember that using this methodology missed something back in the early states: late movement. As I said earlier, one poll showing Candidate X up two points is usually not enough to convince me that said candidate is really gaining ground. But, just to be sure, I took the last six polls conducted this week and combined them together, and then I took the first six polls conducted this week and combined them together, and compared the results. Interesting outcome. Take a look:

First Six

Obama - 41.7%
Clinton - 27.6%
Edwards - 15.7%
Undecided - 15%

Last Six

Obama - 41.5%
Clinton - 27%
Edwards - 20.7%
Undecided - 10.7%

There has been a statistically significant increase in Edwards' level of support. Put plainly, he's jumped five points, while Obama and Clinton have stayed pretty much the same. That, of course, leads to the conclusion that the undecideds are breaking to Edwards in a big way.

So, my prediction.

Obama - 43%
Clinton - 29%
Edwards - 28%

I think Obama will win, and Clinton was come in second, but I think it will be a closer race for second than was originally predicted. With undecideds breaking so heavily for Edwards, he could really challenge her for second.

Happy South Carolina Democratic Primary Day!