In the post to follow shortly I'll add a bit more detail to a change in my identification of pollsters' ratings. In each post it will eliminate 538's ranking of each survey while still providing confidence in the cited polls' reliability, and perhaps best of all simplify what I believe is the need for an evaluation of each poll's accuracy. That said, there's this from AtlasIntel:
An exception to my intro is this poll. AtlasIntel is new to the collection of pollsters I have cited to date, and its ranking — 2.7 of 3.0 in 538's listings — is noteworthy because of this post's title: Its findings are peculiar, very peculiar.
Nate Silver writes that the "AtlasIntel poll cut against what was otherwise a strong day of polling for Harris. It’s a highly-rated poll so resist the temptation to unskew."
I do resist "unskewing" polls whose findings are less than pleasing. From an obligation to objectivity I take them at face value, good or bad, though often with explicative commentary. But AtlasIntel's finding is so utterly out of line with a dozen other highly rated polls, my resistance is strong — not that I'm questioning the pollster's objectivity, only that its methodology in this instance was perhaps flawed.
Here's another finding — but in the other direction — for which there is probable cause for reasonable doubt. Notes the newspaper that commissioned it: "The Des Moines Register and Mediacom by Selzer & Co. of Des Moines." Selzer is also highly rated, 2.8/3.0.
"I wouldn’t say 4 points [47-43] is comfortable" for Trump, said the pollster's president, J. Ann Selzer.
Either both polls are broadly inaccurate, or really bizarre fluctuations are taking place within the electorate, nationally and regionally.
Comments