There’s been some debate about whether independent conservatarian candidate Aaron Day (former chairman of the Free State Project Board of Directors) cost Republican Kelly Ayotte her U.S. Senate seat at this past election. Skeptics point to the fact that Day and Libertarian Brian Chabot between them about equaled Gary Johnson’s vote percentage in the presidential race (Johnson-Weld got 4.2% in New Hampshire, Chabot and Day between them about 4.1%). They say Day merely siphoned off Libertarian voters from Chabot, not Republican voters from Ayotte.
They’re probably wrong. Here’s why.
First, let’s note that Clinton and Trump were the two least popular major-party presidential candidates in the history of polling. In particular, Trump was (and is) by far the least popular presidential candidate (and now, president-elect) that we’ve ever seen. Under those conditions, you expect an outsized third-party vote share in the presidential race. By contrast, Maggie Hassan and Kelly Ayotte, the two U.S. Senate candidates, were reasonably popular – and about equally so. So we should expect a small third-party vote share in the U.S. Senate race. The fact that the third-party vote share in the Senate race equalled that in the presidential race therefore suggests something else is going on – perhaps a particularly strong independent candidacy. Libertarian Chabot got 1.7% of the vote, while independent Day got 2.4% of the vote – so if one of them was the particularly strong candidate, it was Day.
To try to see whether Day was siphoning off Republican or Libertarian votes, I looked at town-by-town results for all 239 New Hampshire jurisdictions with voters in this race. I then looked at how the number of Libertarian and Republican presidential votes by town correlated with Day’s support. The results are in the figure below.
The coefficient estimate on percentage of the vote for Trump-Pence in a town is 0.05 and is highly statistically significant, as you can see by the tiny confidence interval on the estimate. This estimate means that for every 20 additional Trump-Pence voters in a town, one additional voter cast a vote for Day in the Senate race.
The coefficient estimate on percentage of the vote for Johnson-Weld in a town is 0.08 but not statistically significant. This estimate implies that for every 12 additional Johnson-Weld voters in a town, one additional voter cast a vote for day in the Senate race — but again, we can’t be sure this is really any different from zero.
Day’s vote share was a little more than one-twentieth of the Trump-Pence ticket’s in New Hampshire. In other words, we can be reasonably confident that almost all of Day’s electoral support came from Republicans, not Libertarians.
To verify that Day specifically was the spoiler, not Brian Chabot, we can do the same exercise for Brian Chabot’s vote shares by town. The figure below shows those results.
Look at the difference! Chabot pulled overwhelmingly from Johnson-Weld voters. For every four additional Johnson-Weld voters in a town, Chabot got an additional one vote. Meanwhile, Chabot did siphon a few Republicans. For every 80 additional Trump-Pence votes in a town, Chabot got about one vote. Still, these results suggest that over half of Chabot’s support came from Libertarians, not Republicans.
Now, Kelly Ayotte lost by 1,000 votes, just over 0.1 percentage points. My estimates suggest that without Aaron Day in the race, Kelly Ayotte would have won her race by about 15,000 votes, similar to Republican Chris Sununu’s margin of victory in the governor’s race.
Is it possible that without Aaron Day in the race, those 16,000 or so disaffected Republicans would still have voted for some other third-party candidate or a write-in, or just that race blank? Sure, it’s possible. We can’t rule out that possibility because we can’t do an experiment in which we randomly assign some towns’ ballots to have Aaron Day on them and some not to have him on them. But at minimum these results strongly suggest that Kelly Ayotte disaffected a decisive share of Republican voters who went for Trump and Sununu but not for her.
PILEUS 
Kudos to Aaron day. Not only are we rid of the weak, anti-Trump, RINO Ayotte, but we still handily maintained control of the Senate. It’s a win-win result!
This is a pretty short-sighted outlook, in that Maggie Hassan will be there for at least 6 years, and it would have been nice to have an extra person to deal with other potential RINO’s like Flake, McCain, Graham, Murk, and Collins.
“Too often people make the mistake of confusing prudence and civility for the abandonment of principle.”
(Quote from Charlie Arlinghouse column)
Yeah thanks, Aaron. See if Hassan responds to your letter asking her to vote for a Trump initiative.
I’m sorry, but this statistical analysis makes no sense. First off, you’re using Beta, not correlation coefficient.
Second, by using percentages, the smallest towns will dominate your results. For example, in Millsfield, 4 people voted for Clinton and 16 for Trump. It also happened that Day got 10% of the vote (2 votes). But that 80% for Trump / 10% for Day will have a huge effect because the larger towns like Nashua will have less extreme percentages, just because there are more people (41% for Trump / 2% for Day). In least squares regression, the extreme points dominate the estimates. But obviously the answer to your question cannot be found in Millsfield.
Third, most of your numbers are wrong.
Finally, think through your model. You say your results show that 5% of Trump voters voted for Day, because a 1% increase in the percentage of voters who went for Trump is associated on average with a 0.05% increase with the percentage of voters who went for Day. If voting for Day is a disease peculiar to Trump voters, and strikes independently of other considerations, this could be a reasonable model. But that’s not what’s going on.
Here’s a much better model. Look at the “missing Senate voters”; the people who voted DEM / GOP / LIB for President but not for Senate. Regress these numbers on Day’s votes. You’ll find that every Day vote is associated with 0.84 (0.79 to 0.89 95% confidence interval) missing Chabot voters, so clearly this is the pool from which Day drew nearly all his support.
Each Day vote is associated with -0.25 (-0.43 to -0.08) missing votes for Ayotte. But that doesn’t mean Ayotte benefitted from Day’s candidacy. Where Day got a lot of votes, there were also more people voting GOP for Senate but not President. But these people would likely have voted for Ayotte with or without Day in the race. There’s no credible causation story for Day support translating to more Ayotte support.
Each Day vote is associated with -0.05 (-0.16 to +0.05) missing Hassan votes. This could easily be zero. If it’s truly negative, it’s likely a similar story to Ayotte. Day got votes where Clinton was unpopular. That doesn’t mean dissatisfied Democrats voted for Day, just that places where Clinton was unpopular are the same places where Trump was unpopular.
So what would have happened if Day had not run? Clearly Chabot would have taken most of his votes, but the statistics suggest 2,800 of them might have gone to other candidates, with Ayotte more likely than Hassan to collect them. But since they are associated with places where people didn’t like the major party candidates, my guess is most of them would have not voted, or written in a name. If about a quarter went for Ayotte, she could have won. I’d rate that about a toss up.
It’s true that this year Johnson/Weld mostly overperformed down-ballot Libertarian U.S. Senate candidates; but in no other state by as much as in New Hampshire. In some states, the Libertarian candidate for Senate actually over-performed Johnson, and in most others they got at least two-thirds of the Johnson vote.
I think it’s true that Day’s presence in the race probably pulled at least 1,000 votes from Ayotte; just because of what a razor-thin margin that is compared to Day’s 17,742 votes. I think it’s also true that at least 1,018 Day voters also voted for Trump & Sununu.
But I think 16,000 of those votes going to Ayotte, or 16,000 of them who voted for Trump/Pence/Sununu/Day, is an overestimate. It could be a fraction of that and still be enough, but a fraction of that is closer to the truth I think. There were probably well over 1,000 people who voted for either Johnson and/or Abramson but chose Day over Chalbot.
It’s also true that when the margin is that close, a lot of plausible causes are but-for explanations for the result, up to and including random noise. Maybe if Ayotte had a marginally more effective or better-funded GOTV operation. Maybe if it had been snowy or rainy and depressed urban turnout slightly more than rural turnout. Maybe if Ayotte had embraced Trump and lost fewer Trump voters, or maybe if Ayotte had denounced Trump more and gained more anti-Trump voters.
It’s very hard to point to one single factor and say that explains a 1,018 vote margin to the exclusion of other explanations. I think Day can claim credit for his scalp, while it *also* being true that he got most of his votes, or at least a substantial chunk of them, from votes that otherwise would have still gone to Chalbot.
A more likely estimate is that about two-thirds of Day’s support come from otherwise-Chalbot and about one-third came from otherwise-Ayotte… which is still enough to let Day claim his scalp; but also leaves true the point that most his votes came from splitting the third-party Libertarian vote.
I don’t see any way to know for sure just how many votes Day took away from Ayotte as opposed to nonvoters or other candidates. What we do know, if the town-level results are a good guide to individual-level vote choices (obligatory ecological fallacy disclaimer), is that he got votes from about 16,000 Trump voters (plus or minus 1,500) and about 1,000 Johnson voters (plus or minus 1,500), plus some indeterminate number of presidential nonvoters.
Another factor that makes me think that without Day Ayotte would have won by about the same amount as Sununu is the fact that she and Sununu were doing roughly as well in preelection polls and in relative favorability compared to their opponents. In fact, if anything, Ayotte’s net favorability was noticeably closer to Hassan’s than Sununu’s was to Van Ostern’s. And both faced Libertarian challengers. True, Sununu had a name recognition advantage over Van Ostern, but it’s not clear that benefited him much because of his net favorability disadvantage.
But since you already know it, I should haven’t to explain to you why the ecological fallacy is a fallacy.
This sort of geographical inference isn’t good methodology to begin with, and the result you claim is highly implausible. What reason do we have to think that many times more Johnson voters picked Ayotte than Day? That doesn’t make any intuitive sense.
Sometimes it’s the only kind of inference we have available, and I think the result is highly plausible. Most Johnson voters around the country were NeverTrump Republicans.
I think it’s true that minus Day, Ayotte would have likely won. I just think your breakdown of which presidential-candidate voters voted for which Senate candidates is highly implausible and isn’t really proven by your methodology.
Yep. Ayotte caused her own loss because she had a 16% conservative/small government/non interventionist voting record!! Sorry Kelly you hung out with the wrong people (Graham McCain etc.)
It is not so much that Aaron Day ran and caused Kelly Ayotte to lose the seat; anyone can run for any office under any designation and people can choose to vote for those people or note. However, the undue influence of Day’s campaign comes into play when someone or some org illegally spent potentially tens of thousands of dollars to lie to voters and influence them into voting for Day: https://patch.com/new-hampshire/concord-nh/illegal-political-mailers-aim-depress-ayotte-vote-promote-indie-challenger
Yeah, I saw that. It wasn’t Aaron – everyone seems to think it was the Dems.