Undecideds (Still) Don't Break Predictably

Nate Silver and Galen Druke presenting Model Talk Wit’ Da Boys, October 28th.

In the FiveThirtyEight Model Talk podcast the other day, they answered a lot of interesting questions about the midterms, including this one:

GALEN DRUKE: Next question. "In 2010, you wrote [that] the rule that, quote, 'undecideds break toward challengers' was false." We have some-- we have a listener who is digging back into your 2010 archive. Appreciate it.

NATE SILVER: Jesus Christ.

GALEN: "However, you noted that it had been true back in the 1980s. Has that changed at all since 2010? Do undecided voters break in any empirically predictable pattern?"

NATE: As far as I can tell, no. I mean I think it used to be people were reluctant to, um, say they'd vote for somebody who they don't know, and now, in a more partisan era, uh, you know: Galen Druke, Democrat; Galen Druke, Republican—I don't care about Galen Druke; I just care about the party label by your name.

GALEN: You don't care about me?

NATE: I mean, like, very early, right? We turn the model on in the summer basically, right? If you're looking like a year in advance there might be something to that. I, um, but this is more myth than fact. Look at the margin and not the—I mean, you know, looking at the share of the vote can be informative, too. Certainly, like, um you know, some of these races, it's less that Democrats have lost ground than that Republicans have gained it, which is typically what happens as fewer people are undecided the we get closer to the election? So it's kind of relevant context? But still, to make a forecast, uh, the margin is the best number.

Silver cites no sources here, but I don’t think he needs to. He’s Nate Silver. Have you seen his Brier scores? So we’ll take his word for it: more undecideds in a race creates more uncertainty about the race’s outcome, and (as he says) some undecideds predictably “come home” to their normal party (which is one reason why polls tend to converge with race “fundamentals” as the election nears)…

…but margin is more important than vote share, because the remaining undecideds have no particularly strong history of “breaking” against the incumbent—or for the incumbent, for that matter. Undecideds break unpredictably, and their last-minute decisions about whom to support can be an important source of polling error. This happened in 2016. (Moreover, committed voters can and do change their minds! This means that even a candidate who has secured 51% of the vote isn’t very safe if the margin is close!)

This is interesting for a couple of reasons:

1. Lots of people still believe that undecided voters tend to end up supporting the challenger, even though that has not been true since the 1980s and perhaps early 1990s. This error leads some people to expect an even larger Republican victory on Election Day than is currently forecast. This does not appear to be the case. The few remaining “true” undecideds could break toward the GOP… or toward the Democrats! Or they might split even-stevens. We’ll find out on (or slightly after) E-Day.

2. I was the one who submitted this particular question.¹ So when Mr. Druke said that stuff about “a listener” digging around in the 2010 archive, and Mr. Silver said “Jesus Christ!”, I was the one Nate Silver was swearing about! I think that’s going on the De Civ About page.

3. It wasn’t so much that I was digging around in the archive as that I’ve read everything Nate Silver has ever written about politics, and remembered reading this one. It did take me a couple Googles to find it, though.²

4. I have a bad feeling about undecided voters this year. Nothing empirical, nothing I’d even put money on, just a bad feeling. I have no money in PredictIt or Kalshi election markets so far. This is partly because the range of possible outcomes seems very large, paralyzing me with indecision. It is also partly because I try to avoid betting against my desired outcomes in races I’m emotionally invested in. If I were to place a bet, however, I think that Democratic stock in the Senate race is greatly undervalued. It’s selling for $0.32, and I think it’s worth $0.50-$0.55. Conventional wisdom is correct that a big GOP win (reaching as far as New Hampshire and Oregon) is very much within reach… but conventional wisdom has completely lost touch with how few things need to break against the GOP for the Democrats to not only retain control of the Senate, but the House as well.

1. Thanks to Mike B. for a discussion the other day that raised the question in my mind.

2. Yes, I tried Bing first, obviously, but Bing doesn’t recognize the “site:fivethirtyeight.com” search syntax, and so is useless for a lot of advanced searches.