Wednesday, September 26, 2012

Skewed and Unskewed Polls

I mentioned this morning that I'm not going to engage in conspiracies about how all the polling is biased toward Obama and the Democrats. Some polls are biased because any poll will have a systematic error as a feature of the methodology. Whether there's an actual demonstrable left-wing tilt to a survey will depend on a lot of factors. Polling organizations will adjust their poll samples to reflect larger census data for various demographic groups, for example. A survey can adjust samples to reflect partisan identification trends for some earlier time period, like the last election. A lot of conservatives, Ed Morrissey comes to mind, likes to compare polling samples to the level of partisan turnout in previous elections. That makes sense but such analyses rely on conjecture and estimation. The factors that contributed to turnout in a prior election may or may not recur in the upcoming election, so estimating polling bias against present trends such as voter enthusiasm is basically unscientific. There's some excellent learned guesswork going on, but conclusions are unsystematic.

So again I'm interested in all this right-wing angst at the current polling trends. My recent piece at PJ Media piqued my interest especially, since the comments there were decidedly unhappy with the suggestion that Romney was imploding, an argument based on actual campaign flubs as well as a growing gap in most of the recent polls. The fact is, President Obama's bump after the conventions has been significant and sustained. That should be non-controversial. That said, any poll putting Obama up by 9 or 10 points over Romney in the swing states is likely an extreme outlier, as Reliapundit argued today, "MORE BS POLLS FROM THE LEGACY MEDIA: NYTIMES HAS OBAMA NEARLY 10% UP IN OHIO AND FLORIDA." Other polls show those states as closely fought, within the margin of error. I expect at this point Obama is up in both those states by just a couple of points, and if so, GOP voter enthusiasm could indeed put Romney ahead on election day. So it ain't over until its over, no doubt.

That said, let's check in with Jay Cost, at the Weekly Standard, "Are the Polls Tilted Toward Obama?"

I can't get a pullout quote from that, so read it all. Cost first touches on some of the points I've raised above about weighting and estimation of previous voter coalitions. But the more interesting thing is Cost's historical analysis of partisan retention. The GOP has a stronger record of keeping partisans in its column on election day. If those trends hold for 2012, then as we get closer to election day, the Obama convention bounce should largely evaporate and Romney should hold a slight advantage if partisan defection rates are factored in. All of this assumes that survey samples are not biased towards Democrats, and that "true partisanship" is teased out with follow up questions for those who first identify as independents. The Monkey Cage has a worthwhile post on that, "Polling Biases and Their Potential Impacts."

So basically, all this big hubbub about 10-point Obama margins in Ohio and Florida is indeed bullshit. But those polls showing Obama with a slight edge are probably accurate, especially when compared to polling trends over the last few months and when compared to contemporaneous surveys. As I said this morning, I'll be more excited when I see some polls with Romney holding a slight edge rather than Obama. So far Romney hasn't been able to pull up neck-and-neck in the horse race. That's what should be worrying conservatives --- Romney's year-long failure to secure an edge in the polls --- not the so-called huge left-wing bias to the entire polling universe. I just don't think there's that big a leftist bias, and I'm surprised to see some on the right entertaining conspiracies. On that point I don't normally refer to Erick Erickson, who I personally think is an asshole, but on this he's right, "On Polls and Polling":
I do not believe the polls are all wrong. I do not believe there is some intentional, orchestrated campaign to suppress the GOP vote by showing Mitt Romney losing. I actually believe that Mitt Romney trails Barack Obama. I think Republicans putting their hopes in the polls all being wrong is foolish. But I also believe the polls are reflecting a bigger Democratic strength than is really there.
More at the link.

And check Allahpundit at Hot Air for the backlash he's getting for posting polls critical of the GOP ticket, "Which polls are, or aren’t, legitimate?" Allah's describing the kind of harsh response I got at PJ Media last week.

No doubt there'll be more to this story, but I'll close with some red meat for the hyper-partisans looking to utterly destroy the Dems on November 6. It's Dick Morris with an extremely bullish take on the race, which has even Sean Hannity saying, nah (via Nice Deb):