Presidential Predictions 2012 Headline Animator

Thursday, June 26, 2008

What My Electoral Predictions Are About

Hello all,

My name is Chris Comninel and I am extremely interested in this year's presidential election, as I believe many other American citizens are. Last cycle (2004), I was only 17 and was not able to vote; however, I was very in tune with the political process and the particulars of the outcome of the electoral college.

This sincere interest in the process, and a particular candidate, I decided to be as informed as possible about the election, and state polling, for it would be state polling that would most accurately gage the possible outcomes of the election. I searched for some type of averaging formula, but at the time could not think of a good empirical formula. Of course I did give more weight to more recent polls, and gave more weight to polls which had a larger sample size. Given the lack of such a particular formula, I was pretty accurate in noting that Ohio would be within about a point (I have an excel spreadsheet to prove it).

So this election cycle, I thought I would get specific and try to figure out a proper method of averaging the state polls, which are coming out in great numbers due to increased interest. So I followed my weighting stated above, with regard to recentness, but it still had much to be desired so, I did some research and found a remarkable idea, which is currently used at 538.com, a half-life formula of 30 days. Also, I am using a weighting number, given to each poll, based on primary accuracy, with the most accurate organization receiving a number of 3, and decreasing at a rate according to pollster ratings on 538.com.

So here is my methodology, and I am crediting 538.com, for a lot of help in this process. First, my base poll, is a SurveyUSA poll conducted from 2/26-2/28 in all 50 states. Then, there is a jump for most states to mid May, when it was pretty clear who each party's nominee would be, to the most recent poll I could find in a given state. For each poll, I multiply the 1) recentness number, 2) poll accuracy number, 3) the sample size number, and 4) the candidate's percentage received. This gives each candidate a "Multiplication Number." For each candidate, the "Mult Nums." are summed up and divided by a number created by multiplying the 1) recentness number, 2) poll accuracy number, and 3) the sample size number. This gives each candidate an average percentage. The numbers change each time a new poll is introduced.

This is different from 538's predictions because I do not factor in regression data, because although we can understand something from prior elections, I believe each election is unique and has factors which make it such. For example, in '96 and '04, we were dealing with incumbents and any election is going to favor an incumbent because of his great amount of resources. Also, my predictions change daily and are solely based on my averages, and not trend lines. This may change in the future, but because it it so early trends may be exaggerated in a particular direction.

I am doing this solely for my own interests and I am posting it just so its out there. If you find it interesting, thats great, if not, I still do. In coming days, I will post current averages in varying states, and have a little bit of commentary on each average.

Thanks for reading.