Some of you may have remembered that I did electoral college projections back in 2008 and I was thinking of doing it again this year. I thought I would start out by doing some of the GOP primaries. My code as it existed in 2008 isn’t really meant to handle primaries, besides I don’t have a site yet to enter the information in yet anyway. But I can still enter the stuff into a spreadsheet and figure it out, which is what I’m doing here.
I may also use the primaries to test out what might be the best way to present data. In 2008, I largely did a “weighted average” of all polls, meaning that every poll, no matter how old, had a say in the average, however small that say was. While the general election may not be as volatile as the GOP primary, the primary has shown the pitfalls of that type of calculation as older polls that have a candidate considerably higher or lower than they are now can sometimes skew the poll. Then again, you don’t necessarily want to rely too heavily on polls just released either, I don’t think, so it’s good to have a balance. So I’ll be besting out two lengths of time to try to balance the polls: a 1-week weighted average and a 2-week weighted average. This is using the same algorithm I used in 2008, except I would only take the polls conducted in the last week or last two weeks (exception is if there are fewer than 5 polls available, in which case I take the most recent 5).
Here are the current “projections” based on each of the three calculations:
Weighted average (every poll taken is used):
2-week Average (only polls from past two weeks):
1-Week Average (only polls from past week)
Clearly the 1-week average takes into consideration the recent Gingrich momentum, but that also means it’s listening to the immediate polls that are out. Is that good? Well, I suppose the most recent polls are the more accurate description of the race as it is now. However, it also reduces the number of polls available for the average, meaning that outliers can unduly influence the calculations. As a result, I think I’ll rely on my 2-week average for Florida and see how it does. The history chart for Florida, based on the 2-week average, is below (click on the image for full size):