Sunday, August 23, 2020

One Thing About Convention Bounces

This first thing that FHQ does every four years -- if not more often -- around national convention time is go back and consult Thomas Holbrook's classic, Do Campaigns Matter? on the matter of convention bounces. Yes, there are more recent treatments of the phenomenon, but this one was integral to my formative thoughts on what a candidate may gain -- or in some cases lose -- in the aftermath of their national conventions.

One such updated notion is that in an era of increased polarization between the two major parties, the bump is much reduced compared to earlier cycles (those during the latter half of the 20th century, say). And that does seem to hold up for the most part. Since the turn of the 21st century, these convention bounces have been more muted. There are far more one and two point changes and far fewer bounces approaching double digit gains, for example.

But as some of the Sunday morning show chatter naturally turned to the impact of the virtual Democratic National Convention this past week, FHQ was left with one seemingly important question: Does a bounce depend at all on where a candidate began? In other words, if a candidate starts convention week at 40 percent, then does that candidate receive a larger bounce than someone who entered their convention at 49 percent, for example?

Hypothetically at least, the higher a candidate's standing heading into their convention, the lower the bounce would be expected to be. At least that was my thought before looking at the data (a limited version of it anyway). The American Presidency Project has the data on the bounces the two major party candidates have received since the 1964 cycle. And while Holbrook has some exploration of a candidate's pre-convention standing in the polls in his book, it relies on the share of the two party level of support. I wanted to look at it based just on where the candidates stood in public opinion polling the day before their convention commenced without that adjustment. That includes support for other candidates and undecideds; some measure of any given candidate's room for improvement (the pool from which they will draw their bump).

For that pre-convention starting point, then, FHQ turned to the archives of the Real Clear Politics averages of national presidential trial heat polls dating as far back as those go. No, going back to only the 2004 cycle is likely not the rigorous examination that everyone would prefer, but this is a quick and dirty (first glance) look at the bivariate relationship between a candidate's starting position and the bounce they received from their convention.

And when one regresses the starting positions of the eight candidates across those four cycles on the bounces they got out of them, well, the results are not terribly revealing. For starters, there really is not that much of a relationship between the two.

In fact, at an R2 0.075, there is next to no relationship at all. But the relationship is in the hypothesized direction: negative. As a candidate's pre-convention share rises, that candidate's expected bounce decreases. More precisely, a one point increase in a candidate's standing in the polling averages the day before the convention starts decreases the bounce for that candidate, post-convention by 0.35 points. On the surface that appears at least substantively significant (without being at all statistically significant). But one has to consider the range of the independent variable. That is the gum in the works here. If not for Trump's 40.6 percent share in the RCP averages before the 2016 Republican convention in Cleveland, the range would have run from McCain's 44.2 percent share in 2008 to Obama's 46.4 percent level of support in 2012. That is seven candidates clustered into a range of just around two points.

And that brings this discussion full circle because the reason FHQ even thought about this potential relationship is due to Biden's standing heading into last week's convention. Over these last four cycles, the former vice president's 50.2 percent share in last Sunday's RCP averages was head and shoulders above every other candidate since 2004. It is nearly double the range in the data discussed above (minus the Trump outlier).

Biden's position could also be considered an outlier in the data. However, using that (admittedly weak) regression model above, the predicted bounce for a candidate with a 50.2 percent share heading into a national convention is 0.25 points. And that is not too distant from the "no bounce" many pundits are talking about when looking at the handful of post-convention polls that have been released.

Furthermore and for what it is worth, Trump's standing as of now -- August 22 data -- at RCP is 42.4 percent. The predicted bounce for a candidate beginning convention week there is right under three points. But take that with a huge quarry-full of salt grains. Again, this was just for fun. But it was interesting on some level.

The key in all of this remains the fact that in the 21st century Biden entered the convention last week at a level that we just have not witnessed. And it is worth questioning exactly how much a candidate in that position would gain in an era when many/most are lock in on their team's candidate.

Recent posts:
The Electoral College Map (8/22/20)

The Electoral College Map (8/21/20)

The Electoral College Map (8/20/20)

Follow FHQ on TwitterInstagram and Facebook or subscribe by Email.

No comments: