TLC’s Poll of “Poll of Polls” – EV and Popular Vote Analysis

I am beginning a new feature that will continue through the end of the election, which is The Liberal Crab’s Poll of Poll of Polls.

For those not familiar with the concept of “Poll of Polls”, it’s when a pollster utilizes several polls to come up with a result – such as averaging.  However, many of the pollsters, particularly 538.com, use very sophisticated algorithms to determine that ‘average’.  They will try to eliminate outliers and poll bias to improve the likely result.

My concept is to take it one step further.  Since the “Poll of Polls” is done differently by different pollsters, I am taking those polls and trying to come up with an even better estimate.  None of these “Poll of Polls” is 100% accurate, but they are much more accurate than one poll.  So, in my theortical – non-provable world, I believe if you do a “Poll of Poll of Polls”, it should give you an even better answer.

Several notes, first:

1) The  current methodology is simply to average the results of all polls.  I believe for the sake of this exercise, this will be good enough.

2) I will use any polls I find and will add additional ones as I find them.

3) For the Electoral Map projection, if given the choice, I will use the map from a pollster with no toss-ups.  If the map only includes toss-ups, I will apply the electoral votes to the candidate ahead in their state poll of polls.  If they don’t provide the detailed polls for a state or if there is a tie, I will split the votes equally.  If a State has an odd number of electoral votes, I will give the additional vote to the candidate’s party that previously won the state.

The following sites are included in my first iteration: 538.com, The Princeton Election Consortium, Real Clear Politics, Pollster.com, Open Left, Electoral-Vote, Talking Points Memo.  Talking Points Memo only had a popular vote poll and Princeton, Open Left, and Electoral-Vote only had electoral vote polls.

Popular Vote

Obama – 50.6%  +7.0%

McCain – 43.6%

Electoral College

Obama – 351.8

McCain – 186.3

Some interesting outcomes of my analysis.  5 of the pollsters had the same 353/185 electoral count.  538.com and Electoral-Vote were the two very slight outliers (both within 5 votes).  However, the popular votes polls spread between a 5.4% and 8.3% lead for Obama.  538.com was the biggest outlier having Obama at 52% and McCain at 46%.  For those not familiar, 538.com is a little different as they give you their projection for November 5th.  The other sites are only a snapshot of today.  Nate Silver’s (of 538) algorithm is showing what the vote will look like with no ‘undecideds’.

So there you have it.  Take it for what it’s worth.  I have degrees in Business Finance and Political Science and definitely not a stat’s junkie.  I think the take-away is everyone is relatively comfortable with the electoral college being at 350 for Obama.  That’s great news as protection if McCain picks off Ohio or Florida (or even Pennsylvania).  The problem is that much of this is a house of cards and Obama could win big or not at all.

Open Left has a neat analysis that shows Obama with a solid lead of 7% or better in states accounting for 280 electoral votes.  Chris Bowers (of Open Left) says that’s significant because no major race poll has been off by more than 7% if taken the final week of the campaign.  I like that stat because if the ‘maximum’ Bradley Effect were to occur, pundits seem to think it might account for 5-6% of an impact.

I am open to suggestions and concerns, please feel free to post.