The Republican Party announced today that the party’s 2016 nominating convention will be held in Cleveland. The choice seems driven by the hope that it may influence the presidential vote in Ohio, which is likely to once again be a key swing state in the next presidential election. But if this is the hope, Republicans are likely to be disappointed. There is no evidence of which I am aware that the location of a convention influences the outcome of a presidential race. Indeed, the research – what little of it there is – suggests it will make no difference.
By way of background, some time back I reviewed an excellent paper written by Ben Lauderdale and Drew Linzer that cast a jaundiced eye on the ability of political scientists to forecast the outcome of presidential elections based on the so-called fundamentals, such as the status of the economy or whether the U.S. was at war or peace. Linzer, as some of you might recall, created the Votamatic website that used state-based polling to predict long before anyone (I think Drew issued his first prediction in June!) that Obama would win the 2012 presidential election with 331 electoral college votes. That prediction, as it turns out, was exactly correct. Note that on this website Drew utilized a Bayesian forecast model that essentially established a set of prior expectations rooted in Alan Abramowitz’s “time-for-a-change” forecast model, but then updated the Abramowitz forecast by incorporating state-based polls as they came in. By the time election day came around, Linzer’s model was almost entirely driven by state-based polls – it had largely discarded the Abramowitz forecast – and it nailed the final tally right on the nose.
Despite (because of!) this success, Linzer and his coauthor are skeptical regarding the accuracy of fundamentals-only based forecast models. In the Lauderdale/Linzer paper I reviewed, the authors argue “that it is a mistake to take any of these fundamentals-based model predictions too seriously.” Elsewhere I will discuss why they are skeptical of most forecast models. However, having cast doubt on the ability of traditional forecast models to predict the national vote with any degree of precision, the authors then proceed to present their own Bayesian forecast model, based on the 16 elections during the period 1952-2012. This model deliberately incorporates many of the elements in the most popular structural forecast models, but in a way designed to allow “the data to reveal which among a potentially large set of candidate variables are most predictive, letting forecasters remain agnostic about the economic and political factors that ‘really’ matter for predicting elections.” I will leave it to you to peruse their findings (although I find it somewhat reassuring that some of the usual suspects, such as economic measures, do seem statistically correlated with outcomes, albeit with a high degree of uncertainty).
I hope to discuss their results in a later post. For the purposes of today’s announcement by Republicans regarding the convention location, however, one of their findings is particularly relevant: Lauderdale and Linzer include a variable designed to measure the impact of a party’s convention on the presidential vote share! They conclude, with what surely will disappoint Republicans, that “There is little evidence that the location of the national convention is predictive…” To be sure, the authors are careful to note that their forecast model also contains a great deal of uncertainty. Nonetheless, there’s nothing in the paper to suggest that choosing Cleveland will help Republicans win Ohio.
If there’s no evidence that convention location has much discernible impact on the presidential vote, then why choose Cleveland? The short answer is – why not? Doubtless party officials comforted themselves by noting that no Republican candidate has won the presidency without winning Ohio. Of course, there’s a whole list of election-related nostrums like this that are of dubious relevance. Still, there’s no evidence it will hurt the party’s chances and there’s always the chance that, the Lauderdale/Linzer finding notwithstanding (and I have no evidence Republican party officials are even aware of the paper!) that it might help at the margins. In short, the decision was likely driven by the same reasoning that leads candidates to kiss every baby and shake every hand – they don’t know if it will help their electoral prospects, but it probably can’t hurt.
[UPDATE 12:03 Wednesday. Harry Enten has some data that comes to the same conclusion as Lauderdale/Linzer: convention location doesn't matter. See here.]
For now, however, I’m sticking with political science. Cleveland Rocks – but probably not when it comes to influencing the presidential election!