Nate Silver has a podcast up in which he says the race may have tightened from 10+ to 9+ points at a national level, but that this is not reflected in the statewide polls which were always more like 9% in the first place.

But he made an interesting observation that there may be some "herding" going on where pollsters are reluctant to report polls too far outside the mainstream (say Biden +11 or +4) to avoid being embarrassed by the final outcome. The exception being obviously partisan polls like Trafalgar and Rasmussen which seem to be trying to skew the averages by releasing lost of polls.

His other interesting observation was that this "herding" phenomenon is not evident in congressional district level polling because they do not influence national or statewide averages and no one is going to remember a district poll afterwards anyway, so pollster embarrassment is less of an issue.

538 haven't done a systematic analysis of Congressional district level polls, but anecdotally they appear to favour Biden (and Democrat congressional candidates) more than national or statewide polls.

When you add in the early voting data, what evidence we do have appears to suggest that Democrats are more likely to out perform rather than underperform the polls this year, in stark contrast to 2016.


Index of Frank's Diaries