Well, tonight we get the Selzer/DMR/CNN poll out of Iowa, perhaps our last best marker of the race heading into the caucus. A few thoughts heading into it
1) Whether the poll is 'right' or wrong, I would certainly expect Sanders to lead this poll if the race hasn't changed much over the last week.
Last time it was in the field it found Sanders ahead, and Sanders has generally gained in polls since then.
One easy marker: in early Jan., it was a net-11 pts better for Sanders in the matchup v Biden than Monmouth; the more recent Monmouth poll was Biden+2
One easy marker: in early Jan., it was a net-11 pts better for Sanders in the matchup v Biden than Monmouth; the more recent Monmouth poll was Biden+2
Similarly, it's methodologically quite a bit like our poll, which found Sanders+7.
Like our poll, the frame includes the whole RV universe and it's using self-reported turnout. That's been a good combo for Sanders, whether in our poll, DMR, or elsewhere
Like our poll, the frame includes the whole RV universe and it's using self-reported turnout. That's been a good combo for Sanders, whether in our poll, DMR, or elsewhere
Beyond that, the stringent self-reported likely voter screen and the absence of education weights are particularly bad for Biden, who has been at just 15 percent and no better than tied for fourth in the last two DMR polls
Now, the DMR poll is not perfect. It's a poll; polls aren't perfect. It hasn't been perfect in the past. And these methodological choices are unusually sensitive in this caucus, and none of the options are prefect. If Sanders leads, it doesn't mean he's going to win
But I point this out as a matter of benchmarking what to expect here. If Biden managed to lead this poll, I think that would be a really good result for Biden, since it generally makes choices that have tended to reduce his support (rightly or wrongly!)
I would even go so far as to say that if Biden led this poll, it would be fairly strong evidence of actual movement in his direction over the last week, given the larger sample size and the fairly clear evidence that a poll like Selzer would yield a Sanders lead as of a week ago
2) The Selzer poll is not perfect. It's good! Things it has going for it: no herding; a low design effect from a fairly simple set of weights; a really large sample for the caucus; the frame is the whole RV universe. But that does not assure a perfect result in this setting
That was true in the '16 caucus--see, Trump lead. It was even true in the '18 general--see, a Dem lead in the governors race.
I only point this out because the Selzer poll has been put on a one-of-a-kind pedestal, not because these misses are especially alarming or something
And there are real tradeoffs involved in elements of the Selzer approach. A simple set of weights creates some bias-variance tradeoff risks, especially on education (though not the biggest deal in IA), and may just straight up add variance on party reg if she doesn't quota on it.
And self-reported likely voter screens will help you nail a surprising turnout, but there's no guarantee that those irregular voters turn out (perhaps the cause of Trump's overestimate in '16)
Bottom-line: it's a very good poll. It's not perfect, and in this particular caucus, poll results are sensitive to methodological choices. We do know that this set of choices has generally yielded good results for Sanders, and the result can be judged with that in mind
