I saw a reference to a Marquette Law School poll of Wisconsin which found that 80% of Trump supporters expected Trump to win and 80% of Biden supporters expected Biden to win. Of course, people tend to be optimistic, so it's not surprising that expectations differ by party, but that's a big difference. It's also interesting that it's symmetrical--since polls show Biden with a pretty substantial lead, you might expect this supporters to be more optimistic than Trump's.
Questions of this type have been asked of national samples since 1952. I've written about them before, but just looked at the overall distribution. I didn't have time to do a comprehensive analysis, but I went back to 1992 (except 1996--I couldn't find one for that election) and broke it down by party identification (another possibility would be to break it down by candidate you planned to vote for, but the 2020 data just showed it by party ID, so that was what I used).
us them us them log odds
1992 44 47 78 9 2.1
2000 72 17 56 27 2.2
2004 83 7 57 26 3.3
2008 67 31 85 14 2.6
2012 71 19 86 8 3.6
2016 55 40 93 6 3.0
2020 90 9 73 24 3.4
For example, in 1992, only 44% of Republicans expected their candidate (Bush) to win and 47% expected him to lose, while 78% of Democrats expected their candidate (Clinton) to win and only 9% expected him to lose. The log odds in the last column is a measure of the strength of the tendency to expect one's own candidate to win. There are some signs that it's increasing, although with the small number of cases it's hard to be sure. The greatest confidence in one's own candidate occurred for Democrats in 2016, which probably explains something about how they reacted to the election. The lowest level of confidence was for Bush supporters in 1992. As I recall, that campaign had quite a few ups and downs, and that poll may have been taken when things were looking especially bad for Bush. Still, it's remarkable that less than half of Republicans expected him to win (of course, Bush did lose, but not by a landslide--the margin was a little smaller than 2008, when two-thirds of Republicans expected McCain to win). This year, Democrats are less optimistic than Republicans, which is probably because of the memory of 2016.
While I'm on the subject, I didn't offer a prediction on this blog for the 2016 election, but on Nov 5 (the Saturday before the election) I sent an e-mail to a friend giving one:
Nov 5, 2016 Predictions Actual
Clinton 50.5% 48.0%
Trump 46.5% 45.9%
Johnson 2% 3.3%
Jill Stein 0.5% 1.1%
[others] 0.5% 1.7%
At that time, the 538 average of polls showed 45.1% for Clinton, 42.3% Trump, and 4.4% Johnson, leaving 8.2% undecided or other (they didn't break it down further). I thought that a lot of the "other" support would evaporate and that most of it would go to Clinton. In fact, it held up better than I expected, and it seems like the people who decided at the last minute broke for Trump. The current 538 average is 53.1% Biden and 43.2% Trump, leaving 3.7% undecided and other (including the Libertarian). In addition to Biden's bigger lead, the smaller other/undecided group reduces the uncertainty about possible late shifts. I'll say that the other vote will be about 2% (it was about 1% in 2004, 1.5% in 2008, and 2% in 2012), and that things will tighten a little*, producing:
Of course, what matters is how that translates into the Electoral College vote, but with an 8% lead Biden would almost certainly get a majority.
*Also, the Economist model shows a slightly smaller Biden lead.
[Data from the Roper Center for Public Opinion Research]