Poll Frustrations

If you are a political junkie like me, one of the biggest things to play with is polls. Unfortunately, polls are also so easily manipulated based on the pollster's inherent bias, with the most common form being the weighting sample.

Weighting sample is a way that pollsters manipulate the data to make it conform to their expectations. For example, a pollster may get a set of data and his raw data says that the sample was 40% Democrat, 30% Republican, and 30% Independent. If he believes that he has oversampled Democrats and undersampled Republicans, he will apply some factor that will bring the sample more in line to his own expectations and thus be more believable.

In 2008, exit polls showed that the split among voters was 39% Democrat, 32% Republican, and 29% Independent. Mr. Obama won nationally by 7-points. However, in the 2010 mid-terms, voter split nationally was 35% Democrat, 35% Republican, and 30% Independent. Democrats, by in large, were defeated, with Republicans capturing many state offices and the House of Representatives.

From this we can see a significant swing in party affiliation in only two years. Independents remained stable (only shifting by one point) but Democrats dropped by four and Republicans increased by three. The conclusion that one is drawn to, is that predicting the final outcome is a shot in the dark, although one can expect the numbers to fall within a set of norms.

It is unlikely that Mr. Obama will be able to regenerate the same level of voter enthusiasm that he had in 2008 so the 39/32/29 sample represents an extreme Democratic end. Likewise, with a national race to motivate Democrats, it is unlikely that they will be as docile as they were in 2010 so the 35/35/30 sample represents an extreme Republican end. If I were a pollster, I would probably settle on something close to a 36/33/31 model until I saw something that prompted me to shift it to either the left or right.

Of course, most news outlets don't report the polling sample in the headline, they just give the overall number (Obama and Romney tied at 47-47, for example). Hence the frustration with trying to look at the public mood from polls. It does teach one to be skeptical of polls until you can dig down into the raw data and draw your own conclusions, which is a good life lesson in general.