Princeton Election Consortium

A first draft of electoral history. Since 2004

Brief update

September 26th, 2014, 8:45am by Sam Wang


In current polling conditions, the Senate Meta-Margin has left Democratic territory for the first time since mid-August. Not much to say about that for now, except that the sharp movement was driven by fresh polls in Alaska and Colorado. If you  have questions, read about Meta-Margins here.

The prediction is described here; it integrates information from June until now. If current conditions persist, it will start to move. Also, starting next Tuesday it will start to emphasize current conditions more, as those conditions start to be more strongly predictive of Election Day. More later.

Tags: 2014 Election · Senate

6 Comments so far ↓

  • Brian

    The unofficial PEC Google Chrome extension has been updated to work properly when the GOP is in the lead:

    https://chrome.google.com/webstore/detail/princeton-election-consor/mopbhdfigdcejolaebakppgnffbhlamd

  • Tony Roberts

    I spoke to a person from the Orman campaign, and it’s clear as day that if the Republicans end up with 50 members, Orman would be the King Maker and caucus with the Democrats. This, he said, would give Orman the greatest leverage to effect the change he wants. Reports that he would agree to caucus with the Republicans in exchange for Milton Wolf’s endorement are false. Orman will NOT, this representative said, would not make any pre-election decisions about with whom he would cuacus. So the Democrats magic number is 49. Thus, they can afford to lose six seats, three in addition to WV, MT and SD. They must win two of the following: IA, CO, AK, AR and LA (assuming they do not pick up GA). I am exluding NC becuase of recent trends showing Hagan pulling away. At present, I do not see us winning CO, AK, AR or LA (in a runoff). I do think we’ll prevail in IA. Where will that second victory come from to get the Democrats to 49?

  • Amitabh Lath

    I heard your interview and found myself nodding in agreement when you described this analysis as a problem of extracting signal from very noisy (and possibly biased) data.

    In that vein, at what level does the signal get so small (evenly split) that pollsters, even in aggregate, cannot measure it? I’m thinking of some sort of Uncertainty Principle, just as you cannot take a large number of “tape-measure and ruler” style length measurements of a thing and hope to get micron-level precision. I would guess somewhere around a 1% difference.

  • SFBay

    Of the 5 recent polls favoring Gardner, the three that polled previously already had him ahead. So I’m not convinced that the voters have made this seemingly gigantic shift over the last week or so. I expect to see a check by Udall’s name come election night. I guess time will tell.

  • Karl Hudnut

    Ironically, 538′s 52 R seats highest probability now has a chance. Since Orman says he’ll caucus with the majority if R=51 in Jan then they’ll get Orman and it will be 52. Bummer.

  • NP

    It’s not really 3 good polls in a row. It is one outlier – Quinnipiac – and all the others within a few percent.