5 Keys to Campaign Strategy #3: Undecideds in Polls
Or, why a 4-point lead could mean one of two very different things
In the first two entries in this series on campaign strategy, we’ve seen how politicians can more efficiently use the tools in their toolkit by understanding concepts like elasticity, expected margin impact, and the persuasion-GOTV matrix. We’ve also learned how everyday citizens can become smarter consumers of political news by understanding that there are two very different kinds of swing states in America.
To continue down that path, I want to use this piece to explain how we miss a key factor when we read election polls and how that leads voters, election forecasters, pundits, and politicians to make crucial mistakes.
Not all 4-point leads are created equal
Toward the end of the 2016 campaign cycle, Hillary Clinton was leading Donald Trump in key swing states by 3–6 points, per FiveThirtyEight’s polling averages. She famously went on to lose the “Blue Wall” states of Michigan, Pennsylvania, and Wisconsin by a hair.
Joe Biden is now leading by slightly wider margins in these swing states, but I hear a lot of worried Democrats chastising anyone who dares to claim that Biden is in a better position than Clinton was. “Biden is up in the polls, but so was Hillary!”
That analysis misses the mark because it overlooks a key factor: undecided voters.
Suppose your candidate is up 4 points in the polls. It could be that they’re up 47–43, or it could be that they’re up 51–47. There is a huge difference between these two scenarios:
- If you’re up 47–43, a full 10% of voters are undecided. If 8 of the 10 swing against you, you’ll lose 49–51.
- Meanwhile, if you’re up 51–47, just 2% of voters are undecided. You’ll still win 51–49 even if all the undecideds move to your opponent at the last minute.
In other words, there’s a hidden number in polls: the percent of undecided voters. The more undecideds there are, the shakier any lead is. Depending on the number of undecideds, a 4-point lead could be either brittle or rock-solid.
Polling sites often report just the partisan advantage: e.g. “The Republican is up by 2 points.” But that’s only half the equation. You won’t know the true state of the race unless you figure out how many undecideds there are.
2016 versus 2020
The big difference between 2020 and 2016 is that 2020 has far fewer undecideds than 2016, and thus Biden’s leads are much more solid than Clinton’s were, even though both leads are of similar sizes. In 2016, Clinton’s leads were of the 47–43 variety: she hit the mid-40s, but there were enough undecideds that she couldn’t break the crucial 51% threshold.
In 2020, Biden’s leads are of the 51–47 variety: similar lead size, but with far fewer undecideds. Biden’s lead is thus a lot more solid than Clinton’s. By one estimate, only 6.4% of voters in battleground states are undecided or planning to vote third-party, compared to 17.8% in 2016.
More surprisingly, Biden has 50%+ in the polls in nine battlegrounds: Maine, Virginia, New Mexico, Colorado, Michigan, Pennsylvania, New Hampshire, Wisconsin, and Nebraska’s 2nd Congressional district. Clinton had 50%+ in exactly zero battlegrounds.
What everyone got wrong in 2016
Knowing this, we can also push back on the narrative that “the polls were terribly off in 2016.” Pollsters did make a mistake by failing to weight for education, which led them to systematically underrate Trump’s support. But if you look at the data, you see that pollsters captured the state of the election quite accurately: Clinton led among decided voters, but there were a ton of undecided voters.
If you compare the 2016 polls to the 2016 results, you’ll see that both candidates gained support on Election Day as undecideds finally made up their minds. It’s not like the polls overestimated Hillary’s support. The shocking thing was that undecideds swung incredibly hard toward Trump, probably due to Comey’s letter, giving him a net gain of about 4 points in several key swing states. That was enough to put him over the top.
In 2016, the polls weren’t necessarily wrong. They just didn’t account for undecideds breaking toward Trump. Hillary’s 3–4 point edges couldn’t withstand the last-minute shift.
The mistake of even breaks
The polls told us that there were a lot of undecided voters, so we should have understood that they could easily break toward Trump, overcoming Clinton’s polling edge.
Instead, most forecasters (and casual election-watchers, too) just assumed that the undecideds would break evenly, preserving Clinton’s slight edge.
For instance, FiveThirtyEight saw that Clinton was up 44.8–40.8 in Michigan polls, with 5.4% for the Libertarian and 9% undecided. Nate Silver and co. guessed that 3.5% of voters would break for Clinton, and the same number would break for Trump. That would lead Clinton to maintain her 4-point lead on Election Day.
That wasn’t a bad assumption, but as we saw, it ended up being wrong. Instead of breaking 3.5–3.5, undecideds broke something like 1.5–5.5, enough to give Trump his narrow victory.
Some caveats
Hillary’s loss in 2016 was due to a combination of many, many factors. I argue that undecideds breaking against her in the last few weeks was a major factor, but there were doubtless many more. Many complacent Democrats stayed home; pollsters didn’t weight by education and thus underestimated Trump’s support in the Rust Belt; there may have been “shy Trump voters” who didn’t tell pollsters they supported him; Russia hacked state election systems; and so on.
What’s more, just because Biden is above 51% in many swing states doesn’t mean that his victory is a done deal. Polls don’t count; only votes do! Remember to visit vote.org to find out where, how, and when you can vote.
Conclusion
Both the theory and the 2016 example show us that we can’t just look at the top-line results in the polls (like “Democrat +4”). If we ignore or downplay undecided voters, we might miss what’s really going on.
This is significant for state parties, national parties, and folks looking to donate as well. If your candidates are leading in the polls, you need to figure out whether they have a 47–43 edge or a 51–47 edge. Or, if they’re trailing, you need to figure out whether they’re behind 43–47 or 47–51. The more undecideds there are, the more financial and other support a candidate needs.
In part 4 of this series, we’ll zoom out to explore why gerrymandered states can suddenly become competitive, how you can predict these events, and why certain gerrymanderers are bad at their jobs. Stay tuned.
