Conservatives Set Up Their Own Polling Website
The late senator Daniel Patrick Moynihan once told an opponent that he was entitled to his own opinion but not to his own facts. It is doubtful he would have approved of people generating their own data. Still, Republicans unhappy with the national and state polls showing Obama leading Romney by an increasing margin have created a website with polls that weight the demographic distribution of voters differently, supposedly to reduce skew. Democrats counter that there is no reason to think that blacks or Latinos will turn out in lower numbers than in 2008 so there is no reason to reweight the polls with fewer of them.
In addition, while some pollsters use a demographic model to make sure that Democrats and Republicans are weighted according to their numbers in the electorate, other pollsters, such as Quinnipiac University, don't do that. If their sample has 37% Democrats and 35% Republicans, for example, they just assume that is the correct distribution and do not apply any corrections. So for these pollsters, accusations that the 2008 turnout model will not apply to 2012 are irrelevant; they aren't using it. Here is a good article on partisan identification and its impact on polling.
http://www.washingtonpost.com/blogs/the-fix/wp/2012/09/19/how-party-id-became-partisan-and-why-it-shouldnt-be/
The unskewed polls site has been getting a lot of play on right-wing blogs but its numbers are more imaginary than the square root of -1. The average of all the national polls there has Romney leading Obama 52% to 44%. Unless the voter ID laws prevent massive numbers of minorities and students from voting, there is no way that will come close to being true. While individual pollsters may have bias, our technique of averaging all the nonpartisan ones should get rid of most of any residual bias. For example, in our current map, Colorado is an average of four polls by four different pollsters. Florida and Virginia are an average of six polls each, with six different pollsters per state. The map and numbers above include every poll Scott Rasmussen has published this year, even though earlier studies have shown he has a bias of about 4 points towards the Republicans.
Nevertheless, as a thought experiment, suppose all the pollsters had a bias of +3 for the Democrats, what would the map look like? It is easy to run this experiment. All that is necessary is to add 3 to every Republican score in every poll in the database and subtract 3 from every Democratic score. Here is the resulting map.
http://electoral-vote.com/evp2012/Pres/Maps/Sep26-special.html
The electoral vote score is 237 for Obama and 273 for Romney, with Wisconsin and Ohio ties. If we redo the experiment assuming only a 2% bias, Obama leads 284 to 248 with Iowa tied.
What does this experiment show us? Roughly, that Obama is about 2.5% ahead. While this is a completely different way of measuring public opinion than a national poll, in a way it is probably better since it uses data from more than a dozen state pollsters and has a vastly larger sample size than any national poll does. The conclusion is that if Romney does well enough in the first debate to shift public opinion in his direction by 3%, he will have a lead in the electoral college.