There's some discussion here and here about Rasmussen's new North Carolina poll, conducted entirely on Wednesday night, and whether that could have skewed the results in a Democratic direction because those who go to church that night are more likely to be Republicans.
I doubt it for two reasons:
The first is that Rasmussen, unlike PPP and Survey USA, weights for party identification. So no matter when they do their poll, it's going to have the same distribution of Democrats and Republicans that it would at any other time.
The second is that I just haven't seen much evidence that Wednesday night polling really does skew in a Democratic direction. The last time I paid careful attention to it was when we did a South Carolina poll in July. I started it Wednesday night and Obama got hammered. The subsequent polling was much better for him. So that particular time McCain actually polled at a higher percentage when his base was supposedly at church than when it wasn't.
The real story in NC polling right now is that Bev Perdue has on average a five point bigger lead in August than she did in Rasmussen, PPP, and SurveyUSA's July polls, and Kay Hagan has pulled an average of six points closer to Elizabeth Dole in PPP and Survey USA's most recent surveys than she was in the ones before.
Just some back of the paper number crunching, but it appears that the partisan makeup of Ras' poll is somewhere in the neighborhood of 40% Dems, 37% Reps, and 22% Ind. I have no access to Ras' cross tabs, but that is what it looks like. Of course, one could shift the numbers a couple points here and there and get pretty close to the top line number, but this is what my quick analysis shows.
ReplyDeleteTom,
ReplyDeleteThe problem is even if they weight up Republicans, they miss an important segment - Wednesday night church goers. That may be good or bad for McCain. They could be upset with him on abortion or they could be overwhelmingly for him. In either case the exclusion of such voters from the overall total would affect the outcome of the Republican sample prior to weighting which could have effect on the poll results.