Quote:
Originally Posted by Cotton Hill
I think their models actually respect statistical variance way more accurately.
Some people will say "Oh Hillary has been ahead in every poll of this state for the past 6 months, she's got it 100%" without necessarily considering the very small, but non-zero possibility, that those polls were all off.
538 are not baking in a "very small" probability that all the polls are off though. In New Hampshire, where Trump has not led in a RCP-aggregated poll EVER and the aggregate is C+4.7, they nonetheless have Trump at 26.9% to win the state. (Trump has led in a few polls that 538 have aggregated, their methodology was too dubious for RCP to include them).
I'm not buying that the poll average is not just 5 points out, but 5 points out in the correct direction, over a quarter of the time. That sounds way wrong to me. Like that is outside the MoE of
single polls, let alone aggregates.