Princeton Election Consortium

Innovations in democracy since 2004

Outcome: Biden 306 EV (D+1.2% from toss-up), Senate 50 D (D+1.0%)
Nov 3 polls: Biden 342 EV (D+5.3%), Senate 50-55 D (D+3.9%), House control D+4.6%
Moneyball states: President AZ NE-2 NV, Senate MT ME AK, Legislatures KS TX NC

In the Washington Post

November 12th, 2020, 2:58pm by Sam Wang

In today’s Washington Post, my take on the sources of polling errors in Senate races – and the consequences for resource allocation when we don’t take those errors into account.

I have more to say to PEC readers on this, since it was a major feature of what we did. If we knew then what we know now, North Carolina and Georgia Senate races would have gotten more emphasis. I will dissect that soon.

Tags: 2020 Election · Senate

14 Comments so far ↓

  • xian

    I like the idea of tracking how close candidates get to 50%. This time around there was a bit more awareness about vote totals and not as much on spread (which misled so many of us with Hillary), but I still felt that a challenger with 48% was likely to beat an incumbent struggling to hit that number.

    There does seem to be a pattern where polls often get the core vote right but misallocate the “unverbalized” voters. Even “very off” polls often have underestimated the winner vs overestimated the loser.

  • Bette Yamada

    Could you maybe post this article somewhere in front of a Paywall?

    • Sam Wang

      You already get this entire site for free! On the other hand, they paid me a small amount for that piece. Surely the least you can do is pay for it.

    • 538_Refugee

      WP is cookie based. I have my browser set up to ‘toss its cookies’ when I close out. Sometimes I simply go to the settings and do it manually. At some point there is going to have to be an intelligence based clearing house that can distribute payments to sites you visit. There has to be a weeding of click bait.

  • Ellen Carey

    Sam, really appreciated reading your perspective and ways pollsters can better weight responses. How do you think adding a more qualitative research approach would have helped here? I’m biased as a user researcher but I can’t shake the feeling that we only look at what the numbers are, then we all debate other’s opinions – however well-founded – to determine the why of it. There is a better way to do this.

  • Harvey Moseley

    The much larger electorate must have been a been a significant factor in the uncertainty and systematic error. How accurately did the “likely voter” category in polling capture all the new voters since 2016? Was there a differential in the likelihood of new Democratic voters being identified in polling as opposed to Republicans?

  • Hari Mallidi

    Dr Wang – This is a wonderful resource and I appreciate all your hard work in making this site as unbiased as possible. However, at what point are systematic polling errors evidence of election manipulation?

    A senate polling error > 2.5 times historical average error suggests a major flaw in the conduct of the election. Could the behavior of the post-office, delaying the delivery of mail-in ballots, be an explanation for the systematic error of all the polls towards the Republican candidates? Essentially, it appears that there was a 5% shift towards Republicans up and down the ticket all across the country. Could this shift be explained by problems that affect one party’s voters more than the other? Couldn’t this be seen as evidence of effective voter suppression?

  • A New Jersey Farmer


    It appears that the campaigns also had issues with their internal polling, which led to their misallocation of resources, especially Trump in Michigan and Biden in Florida. My impression was that the campaigns had more accurate information than the polling firms. Is this generally not true anymore?

  • ArcticStones

    Sam, this morning you re-Tweeted this by election security writer/advocate Jennifer Cohn:

    …the USPS “processed 94.5 % of ballots on time (the expected rate is 97%),” but in the “9 postal districts spanning 5 states — AZ, NV, PA, NC & GA — that have yet to decide the presidential race, the on-time rate was 84.6 %.”

    I do wonder what impact the 15.4 % undelivered ballots in these key states, and 5.5 % (or more?) in other states have had on Senate control, down-ticket races, and even the Presidential race.

    If the Postal Service had delivered all/more ballots on time…

    — Could Ossoff or Warnock have won outright in Georgia? Cunningham won in North Carolina?

    — Would the Democrats be set to take control of the Senate?

    — Might this have had consequences for fairer redistricting / anti-gerrymandering? And if so, where?

    — Would Biden have won North Carolina as well?

    As Jennifer Cohn comments: This seems to have been surgical!

    • ArcticStones

      Also: Might the high rate of undelivered mail-in ballots merit a re-examination of how much the polls were actually off?

  • 538_Refugee

    Shortly after this piece appeared I read it and typed up my take. Then deleted it. It may be some time for this to be properly dissected but presume it will still come back to what I initially deleted. Cult Trump drove turnout on both sides.

    Given the turnout I think we now have a better idea where the electorate is. Our current situation screams for ranked choice to bring a little more democracy to our republic.

  • pjc

    On something unrelated:

    I seem to recall a post in one or your blog suggesting that GA voters had to have voted in the Nov 3 election to participate again in the Jan 4 runoff.

    This should be double-checked, and, if necessary, corrected in the appropriate place on the blog.

    • Pechmerle

      As has been noted on this site several times previously, the deadline to register for the Jan. 5 Georgia runoffs is DEC. 7. And the voter does not have to have voted, or even registered and not voted, in the Nov. 3 general election

Leave a Comment