Princeton Election Consortium

Innovations in democracy since 2004

Nov 03: Biden 342 EV (D+5.3% from toss-up), Senate 53 D, 47 R (D+3.9%), House control D+4.6%
Moneyball states: President AZ NE-2 NV, Senate MT ME AK, Legislatures KS TX NC

Senate poll error: GOP undecideds came home

November 9th, 2020, 12:14pm by Sam Wang


The polling errors for Senate candidates were quite large this year. In 13 races with final-ish results, the margin of the outcome was a median of 7.9 points more Republican than the last week of polling. That’s an amazingly large difference. In the past, the median error was no more than 3 points or so.

A big reason seems to be voters who were “undecided” about their Senate choice came home to their party. The other day I showed a graph that showed this effect (click to enlarge):

However, this didn’t show up for Democrats.

Overall in the above graph, Republican candidates had a considerable reservoir of support that was not apparent in their named polling. We’ve all been living in The Trump Show – perhaps some Republican voters were not paying such close attention to Senate races until it was time to vote.

(The exception to the overall pattern was Maine. In that case, Collins did about 4 points better than her last week of polls (black symbol), and 7 points better than Trump’s vote share (red symbol). I would guess that her last-minute vote against confirming Amy Barrett had an effect.)

The discrepancy between polls and vote share isn’t visible for Democratic candidates.

Overall, Democrats did a median of 1.4 points (SD 2.6 points) better than their final polling numbers. They also matched Biden’s numbers pretty well.

Could this possibility have been detected in advance? I was focused on the two-candidate margin, as was the FiveThirtyEight tracker, which showed that quantity as a main column in their data tables.

In the past there’s been a lot of talk about undecided voters. In Senate polls, the median share of voters not expressing a preference was 5%. It appears that nearly all of these voters were Trump voters who were unable to state a preference in the Senate race.

I’m pretty sure this is not a “shy Graham” vote or whatever. Generally I believe it’s time to retire or de-emphasize the idea that there are shy voters. Why would a voter be more willing to express Trump support than for the Senate candidate?

I suggest another possibility: voters don’t always have the capacity to verbalize what they want. There’s a fair bit of social science research demonstrating that “undecided” voters have hidden preferences that can be inferred by asking them other questions. Josh Gold and I wrote about this in 2008.

In this case, a few more follow-up questions could have helped, such as “Who do you support for President?” or “Are you favorable/unfavorable toward [insert candidate name here?” or even “Will you vote by mail or in person?” This general concept could be tested by looking at this year’s answers to such questions for “undecided” voters. And in the future, pollsters or aggregators could use such information to help assign “undecided” voters to their top-line Senate result.

We used the two-candidate margin for overall calculations. That could be changed to include undecided voters, and in Presidential years to take into account partisan preference. We also used the margin to identify the best opportunities for donation. With undecideds taken into account the highest voter-power states wouldn’t be that different, but their order would change. Broadly, your donations were mostly well-placed – but taking into account Presidential preferences would have highlighted North Carolina and Georgia more.

In any event, it was not a good year for showing the power of donations. I apologize for any role I had in misdirecting your efforts.

Tags: 2020 Election · Senate

8 Comments so far ↓

  • Joseph Bland

    Is it possible to break the move of the “undecideds” by state? I’d be curious to know if it had to do with President-Elect Biden’s comment in the last debate that was used by the Trump campaign to attack him in those states dependent on the hydrocarbon energy industry. My wife has a strong feeling that was a major contributor, and I concur.

    • 538_Refugee

      I think there is a large ‘undecided’ that is looking for an excuse to do what they really want to do anyhow before publicly committing. Last cycle I heard one Trump supporter claim she disliked Hillary Clinton because of some of her supporters. I forget the exact reason, but the point is, she was looking for something to validate her opinion and couldn’t actually find anything to fault Hillary for. Ironically, I always thought of the Clintons as “Republican Lite”. I think the GOP would have actually fared better long term had she won.

  • Joe ByeDon

    Republicans have been using fracking to try and get new voters for years in Pennsylvania. We were told before fracking started here that it was going to bring tons of jobs and be a huge boost to the economy and it never materialized. All that happened was that some people with family farm land cashed in with having a well put on their land, but it wasn’t many people. As far as the jobs go the gas companies brought so many workers from out of state that the big hiring bonanza that they promised never happened. I had to laugh when watching all of the republican political ads spouting the same lies that we were told from the gas companies before fracking came here. Fracking isn’t a big part of our economy in Pennsylvania. Much like big business in other parts of the country the corporations and Republicans and centrist Democrats promise to lift areas out of poverty with their promises of jobs, but just come in and take advantage of the people and leave with most of the money. Thankfully not enough people fell for the lies again in PA to keep Donald Trump in office. I wish people would realize it for the state legislature and quit voting Republicans into office. PA has a ton of issues that could be solved with a left leaning legislature.

  • ArcticStones

    Sam, would it be an idea to redo this analysis with district-level polling, which offers far more granularity?

    That might well better reveal precisely where the polling deviance is greatest, and the multiple factors causing it.

    For research purposes, perhaps you might be able to obtain the complete set of district polls carried out for the Democratic senatorial campaigns! Or even (less likely) also for the Trump Campaign.

    Just a thought!

  • Elise

    I’ve been trying to understand what the election results will mean for redistricting (is it better or worse than in 2010 and in what ways), but I’m not seeing a lot of good information out there. Can we expect a write-up from you soon or am I missing it?

  • Sam Wang

    Martin wrote: “Funny how the polling errors are smaller in the states that decided the election. What are the odds..?

    I wonder how Democrats justify their actions. The electoral coll[e]ge? Gerrymandering? Trump is a ‘Nazi’ and we need to stop him? All of the above probably.

    Anyways. This is not going to end well. They overdid it this time.”

  • jonsax

    I think this Pew data tells an important tale. Personal income went up in every state because of the federal assistance package. Is it once again, “The economy, stupid?”

    https://pew.org/3mFOxVd

Leave a Comment