Usually I don't devote a lot of attention to a single pollster. That's almost as bad news organizations calling the New Hampshire Senate race a toss-up because one poll showed Scott Brown within the margin of error of Sen. Jeanne Shaheen (well, actually two polls--and I was quite ashamed of my native Dartmouth College for reporting the race as such). Or the pundits who suggested that Oregon's Senate race was a toss-up because Monica Wehby led Sen. Jeff Merkley in one poll conducted by a Republican firm. Or the--you get the idea. But as I've said before, I'm paying a lot of attention to this YouGov survey because it's just so ambitious, and just crazy enough to work. Maybe.
To reiterate, I flagged the YouGov survey for a closer look because 1) it was longitudinal, meaning that it would try to track the same panelists in the months leading up to the election, and 2) it was an attempt to poll every race in every state. Even in a state like Utah, where neither Gov. Gary Herbert nor Senators Orrin Hatch or Mike Lee are up for re-election, YouGov assembled a panel of 1087 voters just to ask them the generic House ballot question. I found the sheer scope and ambition of the survey pretty cool.
Unfortunately, as I also wrote, it also sorta fell flat. Too many things--suspicious results, distorted weights, a distinct lack of Asians--seemed off for me to consider any of the results in our average. So I didn't, but I did decide to wait for the next iteration of the poll to come out before passing judgment on YouGov as a whole.
Well, the August iteration is here, and there are definitely signs of improvement. The results are no longer as suspicious--Sen. Cory Booker of New Jersey is now doing better than Sen. Mark Begich of Alaska, as well he should be. Weights have been published in the crosstabs, which are now more readily available here if you just click on each race individually. Most importantly (for me, anyway), the crosstabs now acknowledge the fact that there's more to America than white, brown, and black. Sure, they lump together everything else in an "other" column, but it's a start.
As a reminder, here's what the crosstabs used to look like (I've redone it in Excel):
And here's what they look like now (also redone in Excel):
The most important thing they've changed here is the addition of a "weighted N", which adjusts the column totals to be closer aligned with expected voter turnout. Now, you may be saying, "Hey, those numbers on the bottom don't add to 412!" Don't worry. I noticed it, too, and I really have no idea why. Rounding error may make up part of the difference, but to be honest, it doesn't make much of a difference in the overall percentages on the left. It's certainly not significant enough to make me discard the results entirely, as I did with the last survey.
A more significant concern I'd have with the crosstabs are the weights themselves. From past voting history, for example, I'd expect the electorate to be much more Republican than YouGov has it. In the previous post I used a Democrat:Independent:Republican ratio of 19:42:39 to roughly estimate what weighted crosstabs would look like. In this case, however, I'm going to tell you a secret: these people are paid to do what they do. I'm not. I defer to their judgment and trust that they know what they're doing. They've probably got better ways of estimating voter turnout than I do, and while I may disagree with the methodology, I've got to trust in their methodology if I'm going to trust their results.
Which I am. So congratulations, YouGov. You've made it into my averages. Not that I would assume it's a source of prestige or anything, just as I wouldn't assume that you included Asians in your crosstabs just because I called you out on not having them earlier. But I do appreciate the effort to improve. With any luck, the September iteration will be even better than this one.
No comments:
Post a Comment