In a change from four years ago, when neither Kamala Harris nor Mike Pence were from anything resembling a swing state1, Harris — now heading the Democratic ticket — is more likely than not to pick a running mate from a state with electoral implications. At Polymarket, Pennsylvania Gov. Josh Shapiro and Arizona Sen. Mark Kelly are cofavorites to be her VP pick, and the fourth-likeliest choice is Gov. Roy Cooper of North Carolina, a state that has proven frustrating to Democrats in recent years but isn’t that far from the electoral tipping point.
But if VP picks bring a home-state bonus to their ticket through “favorite son” (or “favorite daughter”) effects, you usually have to squint to see it. Hillary Clinton chose Virginia Sen. Tim Kaine in 2016, for instance, and carried his state by 5.3 points — but that was less than Joe Biden’s 10.1-point margin in Virginia four years later, part of a sharp blue shift in the Commonwealth. In 2012, Mitt Romney picked Wisconsin Rep. Paul Ryan, but lost the Badger State by 6.3 points. That was an improvement over the Republican performance in 2008, when Barack Obama won the state by nearly 14 points (!!), but underwhelming given that Donald Trump and Pence would win the state four years later.
By crunching this sort of before-and-after data — how did the presidential ticket perform relative to four years earlier and four years later, adjusting for overall national trends?2 — our election model (which we’ll turn back on in the forthcoming week with Harris replacing Joe Biden) estimates the magnitude of these home-state effects for both the presidential nominee and the VP pick. And we’ve found out three things, all of which cut against the hypothesis that a veep choice makes all that much difference:
Home-state effects are only about one-quarter as large for the VP pick than for the presidential nominee.
Home-state effects are roughly inversely proportional to the number of electoral votes in a state. In other words, a VP choice can make a small difference in a large state or a relatively large difference in a small state (say, Sarah Palin in Alaska or Biden in Delaware). I think this makes sense intuitively; in a smaller state, a politician is more likely to feel as though he’s from your proverbial backyard.3
And home-state effects have probably diminished over time as partisanship dominates all other considerations.4
So let me be absurdly precise, given that these estimates are based on relatively small sample sizes and include a lot of intrinsic uncertainty. The model estimates that having a VP from Pennsylvania would improve a ticket’s margin in the Keystone State by slightly more than 0.4 percentage points. In Arizona, the effect would be slightly larger according to the model’s logic — a hair over 0.7 percentage points — since it has fewer electoral votes.
How likely is this, exactly, to make a difference? Since we don’t have the Harris-Trump model up and running quite yet, I pulled up the 40,000 simulations we ran on July 10 — before the assassination attempt against Trump and the GOP convention, when the race really got out of hand for Biden. And here’s what I found:
In 1915 of the 40K simulations, or about 5 percent, Biden lost Pennsylvania by 0.4 percentage points or less — enough that Shapiro would theoretically have turned a loss into a win.
However, this is not sufficient for Shapiro to have made the difference in the race. What matters is both that Shapiro was decisive in Pennsylvania and that Pennsylvania was decisive in the Electoral College, with its 19 additional electoral votes enough to put Biden over the top. In other words, we’re looking for simulations where Biden lost Pennsylvania by 0.4 points or less and Biden had between 251 and 269 electoral votes, enough that winning Pennsylvania would get him to 270+. These conditions were satisfied in 427 of the 40K simulations, or about 1 percent of the time.
So in theory, if Biden had replaced Harris with Shapiro — ignoring all the other crazy reverberations that move would have had — he’d have increased his chances of winning the Electoral College by about 1 percentage point based on home-state effects. That isn’t nothing: if you consider this election to be existentially important, then even a 1 percent improvement is meaningful from an expected value standpoint.
And maybe you could argue that this is on the low side. Unlike, say, Ryan, who only represented a single congressional district, Shapiro is the governor of Pennsylvania and he’s a popular one.
On the other hand, in the Biden-Trump version of our model, Pennsylvania was disproportionately important, with a 40 percent chance of being the tipping-point state. The number is likely to be lower when we turn the model back on for Harris, for two reasons. First, the overall amount of uncertainty in the forecast is going to be higher5, meaning that it will be harder to pinpoint the tipping-point state with as much precision. And second, Harris looks as though she’s going to give Democrats more viable paths to victory through states like Georgia instead of forcing them to depend on the Pennsylvania + Michigan + Wisconsin “Blue Wall” trio.
In the Biden-Trump version of our model, a VP from Arizona would have improved Biden’s chances of winning the Electoral College by only 0.2 percentage points (in other words, he would have converted a loss to a win about 1 time in 500). But it’s possible that Pennsylvania vs. Arizona will be a closer call in the Harris-Trump forecast.
Now, I happen to think Shapiro would be a strong VP choice for other reasons. There might be some objections from the Online Left, but I don’t think this is a good excuse to avoid Shapiro: the Online Left is smart about some things, but (like other types of ideologically-minded folks) they tend to extrapolate too much from their personal political bubble when analyzing political strategy. Harris, perhaps unlike Biden, is clearly going to generate a lot of enthusiasm from the Democratic base: the question is whether she can win over enough voters in the center.
But the home state of a VP pick is a worthwhile tiebreaker if you’re otherwise indifferent toward whom you choose— and not much more than that. If there are other electoral reasons to pick Shapiro or to avoid him, or other reasons involving the line of succession and the future implications for the Democratic Party, those are an order of magnitude more important.
Although don’t neglect that Barack Obama won Indiana in 2008, one of the more unlikely electoral outcomes in recent years.
For instance, Obama won by a narrower national popular vote margin in 2012 than in 2008, so that makes any gains the Romney-Ryan ticket made in Wisconsin less impressive.
My Congressional model has also found that incumbency effects are more powerful in smaller states: think Joe Manchin in West Virginia or Susan Collins in Maine.
This is actually one small change we’re making in the Kamala Harris version of the model that we’ll unveil next week. Most parameters in the model are adjusted for rising partisanship, which tends to make electoral margins closer and diminish the importance of “candidate quality”. I hadn’t been applying this adjustment to home-state effects, but I also hadn’t really bothered to look at it, since the presidential matchup had initially been a rematch of 2020 (so whatever effects Trump and Biden had were already priced into the partisan lean of their respective home states) and since most of the veep picks Trump was considering weren’t from swing states. But we’ll be adjusting home-state effects for partisanship going forward. For instance, although Biden notably improved Democrats’ margins in Delaware in 2020 relative to the Clinton-Kaine ticket in 2016, it was by less than the (rather large amount since Delaware is a small state) improvement the model expected.
This is not to imply that we’re going to inject additional uncertainty in the forecast just for fun. It emerges organically from the factors that the model already considers, like that we have far fewer Harris-Trump polls so far than Biden-Trump polls.
Question: given how some of the highly-online left feels about Nate, will there be any sort of counter-effect? "Oh, Nate's saying it doesn't make a big difference - he's trying to talk us out of it! We must do it!".
College was so long ago, but wasn't that the Heisenberg effect, that we can't measure something without affecting it? The Silver effect: Nate can't opine without causing a reaction.
Totally irrelevant hypothetical. Would Al Gore have won in 2000 if he'd picked Bob Graham from Florida? Would a net-net of 1,000 voters in Florida have changed their mind based on that pick?
On the balance, I agree the odds of it being a deciding factor are small, but Shapiro is also a Governor and it seems that might count for a little more than being a Congressman or a Senator who is usually in Washington and may only represent a small part of the state (in the former case). He's apparently relatively popular and has been a pretty public figure in PA recently. Like, I wouldn't be surprised if Shapiro was worth 0.75% instead of 0.4%, which is starting to be enough to be significant.