The appearance of an exceptionally favorable Pew poll to Mitt Romney has many of you in a tizzy. You have asked me to comment further. I am really not good for that kind of thing. I respectfully submit: neither are you! Here is why.
Human beings engage in motivated reasoning: we look harder at evidence that disagrees with our existing (or desired) beliefs. From a statistical standpoint this makes sense: any one piece of information that we receive might be wrong. We carry a model of the world in our brains, and rely on that to interpret everything that comes in. This is sometimes called Bayesian inference.
This even permeates to sensory perception: in the above phenomenon demonstrated by Prof. Patricia Kuhl, the McGurk effect (click here if the video doesn’t work), we rely on visual cues (in this case mismatched) to help infer what is being said.This principle also underlies most visual illusions (cool site!).
If motivated reasoning succeeds, a closely related consequence is biased assimilation, in which agreeable facts get in, but disagreeable ones don’t. This can account for the coincidence that people who don’t like President Obama’s policies also mistakenly believe he was born in Kenya. Sandra Aamodt and I wrote about this in 2008 (“Your Brain Lies To You,” NY Times). in Nobody is immune to this kind of thing, including me. This is why I devise statistical measures that attempt to resist such biases – automatically.
Left-learning readers of this site are sniffing around the internal details of the Pew survey. None of your suspicions change the fact at that present, President Obama has taken a 4-point drop in Popular Meta-Margin. It looks a lot like the effect of the Ryan VP nomination, which lasted for a while; or the Palin VP nomination, which did not.
Detailed comments on an individual poll (i.e. poll-dissection or poll-sniffing) are well-suited for discussions at FiveThirtyEight (and will be heavily moderated here). A sabermetrician like Nate Silver thrives on those details, and when it comes to analysis he is pretty good at suppressing his own biases. The approach here at the Princeton Election Consortium is to compile lots of polls to remove noise. It’s a bit bloodless, but it has worked well in the past. Anyway, give it all some time.