There are a handful of polling firms (e.g. Trafalgar and Rasmussen) that rather explicitly operate on the premise other polls are skewed against Trump. Including them in your polling averages (as 538 does) means the averages *are* accounting for a smidgen of a "shy Trump" effect.

People neglect the degree to which polling averages are themselves sort of a market-driven consensus. Their power of these averages comes as much from the fact that they incorporate different views about the electorate as that they provide for a larger sample size.

This also helps to explain why prediction markets have been largely unable to beat publicly-available polling-based models (e.g. 538's) over the years. The polling averages already account for many of the desirable characteristics of markets.

Market participants don't really get this IMO, so the markets sometimes wind up doing a sort of double-counting. They'll say "polls were skewed against Trump in 2016, so they might be again in 2020". Not at all crazy! But the polling averages *may already be accounting for this*.

If Clinton had won in 16, you wouldn't have as many Trafalgars publishing Trump +2 in every swing state. Pollsters wouldn't be doing as much education-weighting. Etc. Biden would likely have a larger lead *in the averages* even if his true standing in the electorate was the same.

Rasmussen and Trafalgar are not highly-rated pollsters. I would not bet on them being right this year. But I think our *averages* are likely stronger for including a *dose* of them because it reflects a fuller consensus; that's what this thread was about. https://twitter.com/NormOrnstein/status/1301892738573635584?s=20

p.s. This only applies to 538, not RCP. If you *are* going to arbitrarily pick-and-choose which polls to include, then there's no good argument for including Ras & Trafalgar while excluding many other higher-rated pollsters.

538 has actual rules, though: https://fivethirtyeight.com/features/polls-policy-and-faqs/