In the immediate wake of Joe Biden’s withdrawal from the presidential race, speculation turned to whom Kamala Harris, the newly-minted presumptive Democratic standard-bearer, would tap to share the ticket.
Most reporting honed in on six1 major contenders: Governors Josh Shapiro of Pennsylvania, Andy Beshear of Kentucky, JB Pritzker of Illinois, and Tim Walz of Minnesota, as well as Senator Mark Kelly of Arizona and Transportation Secretary Pete Buttigieg of Indiana—white, male, ticket-balancing running mates for the first woman of color to lead a major-party ticket.
Of course, the vice presidential contest is more of an art than a science, really closer to pageantry than anything else. Even electorally, there’s no statistical model to check our thinking on how Beshear’s Appalachian bona fides might play in Western Pennsylvania, or how Midwestern voters may respond to a Sun Belt-dominated ticket. Polling can’t help much either—near-universally, vice presidential contenders are virtual unknowns outside their home state until the moment they’re named as the pick.
With so little concrete data to go off of, the electoral side of veepstakes has largely relied on trusting conventional wisdom—arguments that, despite their soundness at the time, often prove to be illusory. Take, for example, one such narrative pushed by GOP strategists in 2008. Following a tumultuous Democratic primary in which voters narrowly rejected Hillary Clinton’s bid to become the first woman nominated by a major party for the presidency, Republican nominee John McCain named Alaska governor Sarah Palin to the ticket, hoping to capitalize on disaffected Clinton supporters’ desire to see a woman to the White House. This, of course, was a disaster. By October, three in five voters saw Palin as unqualified to serve as vice president, and McCain-Palin would go on to lose women voters by thirteen points in a resounding defeat for Republicans up and down the ballot.
But even in assessing the seemingly-obvious damage Palin did to the ticket, we fall into the same trap. It’s a compact, logical explanation that Palin’s inexperience and numerous gaffes proved detrimental to the campaign’s odds, but the literature surrounding her exact impact is somewhat contradictory. McCain, of course, was already losing! In every regard, the national environment in 2008 proved incredibly hostile to Republicans. Without a real life counterfactual, there are simply too many confounding variables to conclude what effect Palin truly had nationally. Indeed, in the absence of quality data, we often grow more concerned with what simply appears logical than what’s actually true.2 So, given this inherently-speculative nature of veepstakes, what exactly can we pin down?
One piece of conventional wisdom we can test, perhaps, is the vaunted home-state effect—can a vice-presidential pick bring the ticket an electoral boost in their home state?3 Using a modified k-nearest neighbors algorithm, we approximate the expected trend4 in a vice-presidential home state using states with similar demographics. Comparing this to the actual trend gives a rough estimate of the presence and magnitude of a home-state bump.
We should note that this analysis relies entirely on the assumption that the voting patterns of demographically-similar states are closely correlated. It’s no surprise then that it’s most effective in recent cycles, where the electorate has become especially polarized along racial, educational, and religious lines, and the effects of local politics on the national stage have been all but erased. At the same time, these methods clearly struggle with demographically-isolated states. Try as we might to emulate them in the aggregate (everyone knows Alaska is just Oklahoma with a dash of Washington!), there really is no strong analogue for states like it. In many cases, though a home-state boost may exist, we find that, with the amount of noise inherent to the data, it is not statistically significant at a 80% confidence level.
While fairly crude, our model informs a few important conclusions. As others have noted, the home-state effect appears greatest in smaller states; Delaware, Alaska, and Wyoming5 all trended considerably more than expected towards the party of their respective native vice-presidential nominees. By contrast, Kamala Harris seems to have provided zero boost in California, the largest state in the sample. This is quite intuitive—smaller populations have greater variance, and voters are generally more familiar with their politicians in smaller states. Even just culturally, it’s probably more noteworthy in Cheyenne to have a local voice in the White House than it is in San Francisco.
Harris’ apparent lack of local advantage in California also speaks to the home-state effect’s dependence on crossover appeal. Though generally popular in solidly-blue California during her Senate tenure, Harris’ approval was heavily concentrated among Democratic-leaning voters, most of whom likely would have backed Biden regardless of his running mate. That is, Harris’ selection in 2020 probably didn’t win Biden any otherwise-Trump leaning Californians. By comparison, Sarah Palin’s approvals, though low nationally by November, remained moderately high with Alaska independents, fueling a comfortable win in her home state even as McCain-Palin floundered with Native and rural voters elsewhere.
Though not particularly relevant to the 2024 cycle, it remains worth noting that the home-state effect appears somewhat lessened for incumbent vice presidents seeking a second term. During their reelection campaigns, vice presidents Cheney, Biden, and Pence all saw slight trends against their party in their respective states. This, too, is fairly intuitive—we can expect that the elevation to a national office decreases a vice president’s local influence. The general lack of agency—both on policy and rhetoric—that comes with the vice presidency could also play a part in this.
So then, how does this apply to Democrats’ shortlist? Pritzker and Buttigieg, while both politically talented, likely bring insubstantial home-state effects, based both on their electoral track record and the criteria loosely defined above. Beshear and Walz fare better, showing potential for a modest home-state bounce. Regardless, all four are unlikely to alter the current electoral map. Illinois is a lock for Democrats, Indiana and Kentucky the same for Republicans, and, despite the Trump campaign’s insistence otherwise, Harris remains heavily favored to carry Minnesota’s ten electoral votes, regardless of her running mate. Our analysis supports the general consensus that the candidates with the most valuable home-state upside are Mark Kelly and Josh Shapiro, though even the most generous estimates translate that to little more than a two percent improvement in margin. Of course, it’s not an easy question whether those twenty thousand or so home-state voters flipped outweigh a plethora of intangibles—the public jockeying, the coalition building, the baton-twirling that ultimately define veepstakes. In such a close race, Kamala Harris’ answer may well decide the next president.
Methodology
We used a modified k-nearest neighbors algorithm to estimate the voting pattern of a state; the dependent variable in our algorithm was the trend of each state from the previous election cycle, calculated as the difference between the swing in the particular state and the nationwide swing. By convention, trends towards the Democratic Party use positive numbers; trends towards the Republican Party use negative numbers. The training data used comprised all states and the District of Columbia, but excluded the presidential and vice-presidential home states for the given cycle and the previous cycle to eliminate residues from past home-state effects; for estimating trends in states where the vice presidential candidate was from a different election cycle (e.g. finding the model’s estimated 2016 to 2020 trend in North Carolina, Edwards’s home state), that particular state was also removed from the training data. Distance was calculated as the average Euclidean distance in the proportions of each of the following demographic categories: race and ethnicity (Black, Native American, non-Hispanic White, and Hispanic); college education (graduate degree, all other college); and religion (Catholic, Evangelical, other Protestant, Mormon, and unaffiliated). The number of nearest neighbors selected was set as the number of electoral votes that state had in any given cycle; neighbors were chosen until the sum of the electoral college votes of the neighbors matched that of the target state. If that sum exceeded the number of electoral college votes of the target state, the last state chosen received the difference between the two values. Additionally, if a target state had fewer electoral college votes than its closest neighbor (e.g. Alaska, whose three electoral college votes were fewer than Oklahoma’s seven), the next closest state was included, to receive half the weight of the first state. Multiplying the assigned electoral college votes by the inverse of the rankings of the closest states provided the weighting for each neighbor states’ trend value as a component of the estimated trend; the most similar state received a weighting equal to the electoral college votes, the second most similar state received a weighting equal to half its electoral college votes. These weights were multiplied by the trend and averaged to yield the estimated trend for the target state. Estimated weights were found for all years and each state in order to isolate vice presidential home-state effect from other model error, but years in which the vice presidential candidate was otherwise involved on the ticket (e.g. Biden’s 2020 presidential run) were excluded from this calculation.
1 North Carolina governor Roy Cooper withdrew before the publication of this article
2 In another example, the conventional wisdom says the selection of Indiana governor Mike Pence to the 2016 Republican ticket helped then-nominee Donald Trump appeal to socially-conservative evangelical Christians. This idea, while seemingly obvious, finds the same pitfalls as the Palin argument.
3 Existing research suggests that presidential tickets receive a vice presidential home state bump of about 1.8 points, more than enough to swing a close race—however, the bulk of these studies were conducted prior to the Trump era, when presumably the less polarized electorate may have been more open to switching their votes for a favorite son or daughter.
4 VoteHub uses Dave Leip’s definitions of “trend” and “swing,” where trend is the change in margin between cycles normalized to change in national environment, and swing is simply change in margin between cycles.
5 Though we could not obtain sufficient demographic data for 2000, a cursory look at that cycle’s trends by state shows Wyoming clearly moving dramatically further to the right of the nation than the rest of the Mountain West.