The above headline (without the “Maybe, but I’m doubtful”) is from a BBC News article, which continues:
One of the world’s leading political scientists believes Donald Trump most likely won the US presidential election for a very simple reason, writes Hannah Sander – his name came first on the ballot in some critical swing states.
Jon Krosnick has spent 30 years studying how voters choose one candidate rather than another, and says that “at least two” US presidents won their elections because their names were listed first on the ballot, in states where the margin of victory was narrow. . . .
“There is a human tendency to lean towards the first name listed on the ballot,” says Krosnick, a politics professor at Stanford University. “And that has caused increases on average of about three percentage points for candidates, across lots of races and states and years.” . . .
When an election is very close the effect can be decisive, Krosnick says – and in some US states, such as Pennsylvania, Michigan and Wisconsin, the 2016 election was very close.
What about the other close states? In Minnesota, it looks like Trump was first on the ballot, and he did almost come from behind to win that state.
Florida and Pennsylvania appear to list the candidate of the governor’s party first, which would put Trump first in Florida and Clinton first in Pennsylvania. New Hampshire I can’t quite tell, their rules are confusing. Nevada uses alphabetical order so I think this means Clinton went first. In Maine, I’m not sure but it looks like Clinton might have been listed first.
So, suppose ballot order gave Trump the win in Michigan, Wisconsin, and Florida. That’s 16 + 10 + 29 = 55 electoral votes. On the other side, maybe ballot order helped Clinton in Maine (at-large) and New Hampshire, that’s 2 + 4 = 6 electoral votes, for a net gain of 49 for Trump. Take away 49 of Trump’s electoral votes and he no longer has the victory (assuming all electoral voters voted as pledged; I guess that will be our next constitutional crisis, come 2020). We tend to think of all these little things as averaging out, but they don’t have to. The number of swing states is small.
So, yeah, maybe Krosnick is right on this one. It all comes down to Florida, I guess.
Could ballot order have been enough to cause a 1.2% swing? Maybe so, maybe not. The research is mixed. Analyzing data from California elections where a rotation of candidate orders was used across assembly districts, Jon Krosnick, Joanne Miller, and Michael Tichy (2004) found large effects including in the 2000 presidential race. But in a different analysis of California elections, Daniel Ho and Kosuke Imai (2008) write that “in general elections, ballot order significantly impacts only minor party candidates, with no detectable effects on major party candidates.” Ho and Imai also point out that the analysis of Krosnick, Miller, and Tichy is purely observational. That said, we can learn a lot from observational data. Krosnick et al. analyzed data from the 80 assembly districts but it doesn’t look like they controlled for previous election results in those districts, which would be the obvious thing to do in such an analysis. Amy King and Andrew Leigh (2009) analyze Australian elections and find that “being placed first on the ballot increases a candidate’s vote share by about 1 percentage point.” Marc Meredith and Yuval Salant (2013) find effects of 4-5 percentage points, but this is for city council and school board elections so not so relevant for the presidential race. A Google Scholar search found lots and lots of papers on ballot-order effects but mostly on local elections or primary elections, where we’d expect such effects to be larger. This 1990 paper by R. Darcy and Ian McAllister cites research back to the early 1900s!
So, putting all the evidence together: what do I think? As I said above, it all comes down to Florida. In 2000, Florida was extremely close—best estimates has Gore winning by only about 30,000 votes (according to Mebane, the votes were lost “primarily due to defective election administration in the state”), and had ballot order been randomized he could well have won by even more, enough for the state to have counted in his favor in the electoral college.
In 2016, maybe, maybe not. Based on the literature I’ve seen, a 1% swing seems to be on the border of what might be a plausible ballot-order effect for the general election for president, maybe a bit on the high end given our current level of political polarization. So I think Krosnick is overstating the case, but it is just possible that the ballot order effects were large enough that, had the ballots been randomized, Clinton could’ve won Florida, Michigan, Wisconsin, and thus the electoral college.