Alexander Agadjanian and Brian Schaffner
With the 2020 general election season on the horizon, anticipation continues to grow over who Joe Biden will select as his vice president. As the Biden campaign has narrowed down potential running mate options to a shortlist, consensus is still lacking over what Democrats themselves want and what might constitute the most optimal electoral strategy.
Public opinion polls have given some indication of how to make sense of the VP decision process. A recent Monmouth survey finds Democratic voters prefer Kamala Harris by a wide margin over their next choice, but even then, Harris is selected by just 28% of voters. Another poll, one from Morning Consult, shows that when voters are asked whether various potential VPs would make them more or less likely to vote Biden, Warren has the biggest positive impact. Yet, as a recent academic paper details, data based on self-reported attitude change like this is not reliable and likely overstates the potential influence of such a decision.
New Experimental Evidence
To shed better light on the question of whether the choice of vice president impacts electoral prospects, we fielded an experiment on a survey of 4,927 American adults. The poll was sponsored by Data for Progress, a progressive think tank, and was conducted online from June 14th to June 16th using respondents recruited from Lucid (a firm that provides demographically diverse samples to researchers and polling firms).
We focus on two prominent VP options currently being considered: Kamala Harris and Elizabeth Warren. The procedure was simple: in place of the usual vote choice question, we randomly assigned half of respondents to see a hypothetical matchup that showed Harris as Biden’s vice president and another half showing Warren as Biden’s vice president. Respondents were asked how likely they would be to vote and who they would vote for in an election with Donald Trump and Mike Pence running against Joe Biden and either Kamala Harris or Elizabeth Warren.
While the VP selection was randomly assigned to respondents, we also ensured that the two groups were equivalent on a range of factors, including 2016 vote choice, partisanship, ideology, gender, race, education, income, and age. The analysis was also weighted to ensure that it was representative of American adults.
Results
The first graph shows the difference in the Biden - Trump margin when he was assigned Elizabeth Warren rather than Kamala Harris as his running mate. On the whole, the VP selection appears to make little difference. Biden’s margin over Trump drops by just three-tenths of a percentage point when Warren is his running mate, but the 95% confidence interval means that we cannot rule out the possibility that picking Warren over Harris would mean either a larger reduction in Biden’s margin or a significant net-positive effect.
Subgroup results are shown below this top estimate, and they mostly tell a similar story. We cannot be confident that the VP selection would make a difference in the vote margin among any of the groups and in most cases the estimate is quite close to zero. The one exception is for Americans who identify as Latino or another race/ethnicity (other than black). Among that group, Warren’s selection produces a fairly sizable narrowing of Biden’s margin, shifting his advantage over Trump from 57%-22% to 52%-26%. Nevertheless, the fairly large confidence interval for this group indicates that we cannot confidently rule out that there would be no meaningful effect on the margin among this group.
Of course, VP selections are sometimes seen as a way of energizing certain parts of the coalition to help with turnout, rather than as a way to change vote preferences. For this reason, we also asked people to rate on a scale ranging from 0 to 100 how likely they would be to vote (where 0 is a certain non-voter and 100 is a certain voter). While people generally over-report their intention to vote on items such as these, they are still useful for at least understanding whether people become more motivated to vote under a particular condition.
Once again, no significant differences emerge, and the magnitudes of VP effects remain very small. A Biden-Harris ticket increases vote likelihood by just one-quarter of a point compared to a Biden-Warren ticket, but the effect is not distinguishable from zero. The same pattern holds for groups like Independents and Hispanics/Other race identifiers: there is some trace of Harris boosting turnout among these groups, but these changes are not statistically significant.
But the key metric for judging the impact of a particular VP selection is the combination of these two metrics; that is, will the VP selection affect the Biden-Trump margin among people who are likely to vote. To calculate this effect, we weighted the vote choice questions based on how likely an individual said they would be to vote. Utilizing this approach, we find that Warren’s selection slightly reduces Biden’s margin from 10.4 points to 9.1 points, but the shift is not statistically significant.
We conduct a few robustness checks on these main results and find further null results. Most importantly, we presented the same questions to respondents much later in the survey but with the other VP option they didn’t see (e.g. if they were assigned Harris at first, they saw Warren later, and vice versa). When using a within-subjects approach that grants more statistical power, we continue to find null effects across the board despite the gains in precision.
Conclusions
Amid so much attention towards Biden’s VP selection and its possible ramifications, our results offer something of a reality check. Given how stable vote preferences are over time and within election periods, it shouldn’t come as much of a surprise that new peripheral information fails to move intended voting behavior, much of which is already set in stone from now until November. That isn’t to say VP selections can’t have any impact--past research shows these choices can produce home state advantages, though this of course is less relevant for the potential VPs we study as California and Massachusetts are already Democratic strongholds.
As with any research, this study comes with some caveats. First, we stress null effects on intended vote behavior; how the campaign (and the VP debate) plays out for Biden’s selection could ultimately factor into how people vote, though it’s unlikely. Second, our design does not reveal whether any reactions are more about Warren attracting/repelling votes or Harris having this impact. We don’t include a pure control of no VP listed as there’s no world where Biden does not select a VP, but this still somewhat impedes our understanding of the treatment effects. With all of that being said, these results provide as good a current piece of evidence as any on the question of how voters will immediately react to two of the highest-profile VP options Biden is weighing.