First, a tiny dose of cognitive science. Our brains are really good at letting in information that agrees with our prior views – and we look for reasons to reject information that is disagreeable. In a complex media environment, this tendency is deadly. It probably underlies our deep political divisions: getting the agreeable information is very easy. Witness the echo chambers in which dumps of fairly anodyne email from Hillary Clinton take on sinister significance.
People are the same way when they interpret polls. Two past cases come to mind:
- In 2004, state polls were dead-on in giving a snapshot of the close race between Senator John Kerry and President George W. Bush. However, Kerry supporters suggested that undecided voters would break toward the challenger. This was a pretty small break, but it was enough because the race was so close that year. As longtime PEC readers know, I made this error.
- In 2012, “poll un-skewers” on the Republican side took it upon themselves to correct polls that they felt were demographically unbalanced. The king of the un-skewers was Dean Chambers.
This year, I have heard multiple possible objections to interpreting polls at face value (for example, see this PEC comment thread). Here they are, with my answers.
1. Brexit. It is said, wrongly, that polls missed Brexit. However, that is not true. Pre-election polls indicated that Brexit was too close to call – and there were 9% undecideds. It was pundits and conventional wisdom that failed.
2. The 2014 midterms. This is fair – on average, there was a 5 percentage point error that year. However, as I have analyzed, it is generally the case that in midterm years, which represent a low-turnout condition, polls are terrible. However, general election polls did extremely well in 2000, 2004, 2008, and 2012. I wrote an article about it.
3. “We didn’t think Trump would win the primaries, either.” Erm…speak for yourself. Once again, pundits were wrong – but a poll-based approach was correct before the primaries began, and continued to be correct all the way through the primaries. Go through PEC’s archives and you’ll see.
4. Known unknowns. Will Sanders supporters show up to vote for Clinton? Will Trump bring new voters out of the woodwork, particularly people who don’t necessarily answer the phone for pollsters?
Most of these reasons are already captured in polling data, at least at the state level (national polls aren’t as good). For example, if Sanders supporters have come home to the Democratic nominee, they will be captured in polls. The same is likely to be true of Trump supporters; for example, primary-season polls did very well in predicting Trump support.
The most unmoored form of this argument is that somehow, tons of Trump supporters are simply not captured by polls. Evidence contradicts this speculation. This has not been apparent in voter registrations, so there is no clear place for these voters to come from.
Unpersuaded? Maybe you like that one poll from USC/Dornsife, which has serious weighting problems. If you want to imagine what would happen if your side got a few percentage points, the “Clinton +2%” and “Trump +2%” links at right will show you that.
5. Undecideds and Gary Johnson/Jill Stein supporters. In the past, nearly all of these voters end up supporting one of the two major candidates. For example, Johnson and Stein supporters combined got less than 2% of the vote in 2012. What will they do this year? Below, I estimate the impact.
Currently, undecideds and Johnson/Stein supporters constitute about 14% of voters. Drew Linzer has pointed out that undecideds/minor party supporters are unusually high this year, about 6% ahead of 2008 and 2012. This is a legitimate source of uncertainty about the eventual outcome. However, there is information about how these voters will eventually fall.
Undecided voters usually break somewhat evenly (sorry, 2004 Sam!). Data from SurveyMonkey suggests that Johnson supporters break about evenly between Clinton and Trump, while Stein supporters tilt strongly toward Clinton. This is consistent with many state polls that show Clinton doing the same or slightly better when the matchup is Clinton/Trump compared with Clinton/Trump/Johnson/Stein. So the net expected effect is, on average, slightly toward Clinton*.
The topic of undecided/minor-party support requires more unpacking. I would rate it as the most legitimate concern about prediction. In my estimation it alters the probability by a tiny amount at most.
Anyway, if there’s some hidden reason why polls are all wet, it hasn’t come up yet. If you’re a Trump supporter, it would be more productive to focus on downticket Senate and House races, where your side has a better chance of surviving November.
*This can be quantified. Based on recent data, let us assume 6% undecideds, 6% Johnson supporters, and 2% Stein supporters. By November 8th, they might break as follows.
Undecideds: Let’s say these voters end up somewhere between 4.5%-1.5% favoring Clinton, to 4.5%-1.5% favoring Trump. Using the rule of thumb that the range is 4*standard deviation, this leads to a net change of 0.0 +/- 1.5% (mean +/- SD).
Johnson voters: Assuming 1% stay with Johnson, the same logic leads to a net change of 0.0 +/- 1.3%.
Stein voters: Assuming 0.5% stay with Stein and an average 3-to-1 split for Clinton, the effect is an increase in Clinton’s margin by 0.7 +/- 0.5%.
The overall combined change is a net increase of margin in Clinton’s favor by 0.7 +/- 2.3%. Given current conditions and based on the uncertainty of 2.3%, in 3 out of 100 cases enough voters would switch to Trump to close his current deficit.