Woodchopper wrote: ↑
Thu Jun 25, 2020 7:57 pm
Vertigowooyay wrote: ↑
Thu Jun 25, 2020 5:28 pm
Apologies - I can’t look back at the 2016 thread on the old place where this was dissected, but why were the polls wrong in 2016, and is anything different now?
I don't think they were that wrong. As far as I remember, the polls had Clinton slightly ahead in Pennsylvania, Wisconsin and Michigan, but within the margin of error. Complacent people interpreted that as being Clinton ahead, but it would have been better to portray it as being too close to call.
I think the last Fivethirtyeight forecast gave Trump a 30% chance of winning, which shows that the polls hadn't predicted a sure Clinton win.
As for why the polls consistently showed Clinton to be slightly ahead when she was slightly behind, as far as I remember the problem was due to the difficulty in getting a representative sample through calling random phone numbers. Call screening now means that normal people very rarely take part in a survey.
The polls were pretty close on the national popular vote, they just missed part of a swing away from Clinton in a particular demographic in those three states. And because those three states have highly correlated voting patterns, a miss in one became a miss in three. Add in the electoral college, which can over-emphasise the effect of a line-ball result in a single state*, and the polls got the binary call wrong, while being numerically very close. Fivethirtyeight were the only people who were accounting for correlations between states, so even the other psephologists were expecting a Clinton win, and the non-numerical pundits were very confident of the Clinton win.
There was also a very high undecided vote late in the campaign, and those undecideds broke for Trump the challenger, rather than Clinton the pseudo-incumbent/known quantity or one of the high profile Others. This year, Trump is very much a known quantity so likely won't get benefit-of-the-doubt voters and third party campaigns tend to do badly the year after they've seemingly affected the outcome.
In Fivethirtyeight's polling average (which weights for pollster quality, so is a little more sophisticated than other averaging tools), Biden currently has an average of 50.6% and Trump gets 41.1%, with an undecided/other vote of 8.3%. In 2016, Clinton's highpoint in the 538 polling average was 46% and the Libertarian Johnson polled 5-10% through most of the campaign, which added uncertainty.
So, Trump needs to peel off voters who have actively chosen to support Biden, rather than scoop up undecideds, and that is much harder. Also, the last few polls that I've seen crosstabs for have the undecided votes concentrated in younger age brackets, which isn't exactly productive ground for Trump. He has a particular problem in that 4 years of usual mortality (plus the pandemic ones) disproportionately affect his voters because they're older, and the younger generations aren't becoming Republicans as they age. So he *has* to find new voters to replace the dead ones, even if he hangs onto every group that he got last time. There is one path to a second Trump presidency and it requires hanging on to every state he got last time, despite demographic change, a base-only campaigning strategy, an economic crash, a mishandled pandemic, and a black lives matter that is actually changing white people's minds about racism in America. In contrast, Biden has many possible states that he could pick up and he only needs a couple of them. The political environment this year is highly volatile but something is going to have to go extremely strangely for Trump to get back in again.
*At the moment, the EC helps Republicans. If/when Texas goes blue, it will almost certainly become a Democratic advantage. Watch all the pundits switch positions on whether it should be abolished...