Re: Goosing the Polls

Conspiracy, Ezra? No, not as such. There is certainly a cultural bias going on, deeply rooted enough that many pollsters aren't even aware of their prejudice.

But even that is too strong. The science of polling is just like any other science - there is an art to the practice which, if done correctly, will sift the data into wise conclusions, and if done incorrectly, will produce drivel. In general, the art involves the selection of appropriate methodological assumptions.

Most on our side have noticed, for instance, how the pollsters weight their data in accordance with their assumption about the electoral mix on November 6. If you poll 1,000 citizens, you might find that your sample contains 70% Republicans and 30% Democrats. Clearly, this raw data will not reflect the actual electorate on voting day, so if this sample is to yield any meaningful results, it must be re-weighted to approximate reasonable expectations. 

The key is the term "reasonable." Good scientists can differ, and each can make a good case for his position.  Many pollsters have assumed the race in 2012 will reflect the electoral weighting seen in 2008, others have chosen 2004 as the model, and some (mainly the outliers) have assumed an electorate that appeared in the 2010 bi-elections. I myself prefer something closer to the 2010 elections, but recognize that such a choice would be a bit dicey for any reputable pollster. It was not a Presidential election and historically the correlation is not that close between bi-elections and those involving the President.

And there are myriad different electorate assumptions in any sub-group polled, e.g. women v. men voters, black v. white, hispanic turnout, and etc. Good pollsters will have models that deal with all these demographic groups and more, and each model will tend to push the results in one direction or the other.

But given all of this, there is still a problem: given the number of polls (and this election has seen an extraordinary amount of polling information) you would think that the average would cancel everything out and give us a fairly representative forecast. But they don't; the averages, as you note, uniformly and consistently show a tight race, at best a toss-up for Romney. Why should this be if he will win as decisively as I predict?

The answer is David Axlerod. All politicians and campaigns spin the news media to support their candidate. But Axlerod has introduced something new into Presidential politics: he has been assiduously spinning the pollsters to insure their results will mesh with a story of an overwhelming tide of support for Obama that will swamp Romney. He has been so effective at this that even those pollsters he has not contacted have tended to follow the herd and adopt assumptions favorable to Obama. In one reported case, it seems he even invoked the heavy hand of the Justice Department to get Gallup to change its methodology - which it did, suddenly tilting its polls towards Obama after previous polls had shown Romney surging ahead.

Why this man has such credibility, I don't know, but he does. Fortunately, not even his Justice Department can intrude into the privacy of the voting booth, so November 6 should finally give us an accurate poll, and a Romney/Ryan win.