The Poll Averages

[KingSweden]

Looking at the Upshot got me thinking - which of the poll averages do you think are closest to having it right?

538 has been the most bullish on Trump lately, but Nate includes some polls (i.e. all-landline Emerson and way-outdated Dem internals) that seem to be data for the sake of data. I've looked at HuffPost Pollster as well, seems a little less bearish on Clinton than, say RCP.

The two models that hype their correctness are Sam Wang's PEC and Pollyvote, which I discovered for the first time today. Both have shifted in Trump's direction but still strongly predict a Clinton win.

Anyways, what are some of your thoughts? Which averages are the most trustworthy?

[Buffalo Bill]

She was right where McCain was at in 2008 a week ago.

[Erich Maria Remarque]

As it was not enough to unskew the polls, Dems started to unskew 538

[KingSweden]

As it was not enough to unskew the polls, Dems started to unskew 538

538 is, contrary to what you may believe, a site I visit multiple times a day. Just seems the various major averages (including RCP and HP) have some pretty wide divergences

[Young Conservative]

I trust RCP and 538. They're both fairly accurate.

[GeorgiaModerate]

As it was not enough to unskew the polls, Dems started to unskew 538

538 is, contrary to what you may believe, a site I visit multiple times a day. Just seems the various major averages (including RCP and HP) have some pretty wide divergences

I think the main differences are that the 538 models include more uncertainty, and they're also quicker reacting and thereby "noisier" (now-cast more than polls-only, which is in turn more than polls-plus).  Whether this is better than other models such as Upshot is an open question.  Personally I think it's useful to look at models from multiple sources, on the same principle as looking at multiple polls.

[100% pro-life no matter what]

As it was not enough to unskew the polls, Dems started to unskew 538

538 is, contrary to what you may believe, a site I visit multiple times a day. Just seems the various major averages (including RCP and HP) have some pretty wide divergences

Huffington Post puts a few assumptions in, I think, and it also keeps more polls as part of the average, so it's slower to move.  It always does seem more Democratic, though.

[Erich Maria Remarque]

As it was not enough to unskew the polls, Dems started to unskew 538

538 is, contrary to what you may believe, a site I visit multiple times a day. Just seems the various major averages (including RCP and HP) have some pretty wide divergences

Good!

I look only at 538 and Upshot. I really like how 538 weights and adjusts for trendslines, house effects etc; they often "guess" the results of new polls just by adjusting for trends by nationall polls and polls in states with similar elections. But their overall prediction is too noisy.

[KingSweden]

As it was not enough to unskew the polls, Dems started to unskew 538

538 is, contrary to what you may believe, a site I visit multiple times a day. Just seems the various major averages (including RCP and HP) have some pretty wide divergences

Good!

I look only at 538 and Upshot. I really like how 538 weights and adjusts for trendslines, house effects etc; they often "guess" the results of new polls just by adjusting for trends by nationall polls and polls in states with similar elections. But their overall prediction is too noisy.

I like the house effect adjustments too. I think the drawback on Upshot is that it's including projections from Cook, Rothenberg and Sabato that may not be entirely numbers based, though I like the idea of an "aggregate of aggregates." RCP and HuffPo are not my favorites because they both strike me as having implicit biases (RCP more R-friendly, HuffPo too D-friendly, particularly with their non-polling related articles)