The long months of speculation before the Iowa Caucuses abound with headlines about the “momentum” that a candidate has, including nearly half a million Google results for “Buttigieg Momentum” and one atrocious portmanteau in “Buttimentum”. Certainly winning Iowa or New Hampshire can boost a Presidential hopeful’s campaign, whereas losing early states can destroy a candidate’s standing in later states — just ask not-President Rudy Giuliani, whose strategy of avoiding early states ruined his chances at becoming the GOP’s 2008 sacrificial lamb. But do campaigns really experience momentum before the early states vote? Does improving your standing portend further growth in the future? And does a decline in polls suggest that your campaign’s future is dim?
First, I’m not sure what journalists and pundits mean when they say that a candidate has momentum. They might mean that a candidate has recently improved her standing and is going to maintain that improved standing, or that a candidate is likely to improve in the future, or they could have nothing precise in mind at all. (I think the latter is often the case with pundits.) But we’ll look at how much a change in a candidate’s polling in one week tells us about how that candidate’s polling is likely to change next week in a couple of ways.
I grabbed all the data I could from RealClearPolitics’ (codename: RCP) national polling averages for the contested Presidential primaries since 2008 and looked each candidate’s standing each week in the six months preceding the Iowa Caucuses. If “momentum” (in something analogous to the classical physics sense) were the only force governing the trajectories of candidates’ polling, then we would be able to use the change in a candidate’s polling over one week to predict the change in that candidate’s polling over the next week. But the actual story is far, far messier. In fact, there doesn’t seem to be much of a correlation between how a candidate’s polling changed one week with how his polling changed the next week, as we can see when we plot candidates’ change in national polling in a week versus their change the following week.
The numbers confirm what the graph makes obvious — the correlation between a candidate’s polling shifts over one week and the next is very minimal. The linear correlation coefficient suggests that a candidate’s national polling change one week explains less than 3% of her polling change the next week.
We might at this point be tempted to conclude that there really isn’t anything like momentum governing the polling changes in the pre-Iowa months. But there are some things we should note. One is that any polling average is going to be noisy and a polling average that doesn’t normalize based on pollsters’ house effects (RCP doesn’t, though Fivethirtyeight does) will be particularly noisy. This may drown out real momentum effects. Another is that the straightforward correlation between a candidate’s polling changes one week and the next might not actually be what we should be looking for. We usually don’t think of very small changes in polling standing as signifying that a candidate has momentum. Maybe there is a threshold of change in national polling average after which we can say that a candidate really has momentum (or has “lost momentum”).
Most weeks for most candidates come with fairly minimal polling changes. In over 80% of the candidates’ weeks of national polling I examined, a candidate’s standing changed by less than 2%, and in under 5% of cases did a candidate’s polling change by at least 4% in the span of a week. The general picture of pre-Iowa primary campaigns is typically one of short-term stability and few sudden changes.
But what happens after a candidate’s polling undergoes a sudden shift? Looking at the changes in a candidate’s polling after the week following shifts of at least 1%, 2%, 3%, and 4%, this is what we find:
We see that when a candidate’s polling shifts by a notable amount over the course of a week, it is more likely to shift in the same direction than in the opposite direction the following week. For instance, if a candidate’s polling drops by 2% one week, we have some reason to expect that her polling is more likely to drop a similar amount the following week rather than rebound to what it was before the drop. In fact, a candidate whose polling shifts at least 2% one week is more likely to experience a shift of at least 2% in the same direction the following week than a randomly selected candidate is to experience a shift in polling of at least 2% in either direction.
So there seems to be some evidence that there really is something like momentum in pre-Iowa national polling, in that short-term growth in national polling should make us think that future growth is more likely than it would otherwise be (and similarly for short-term decline). But most of the time, short-term growth is followed by relative stagnation. So while there might be something to “momentum”, most of the time momentum — in the sense of continual national polling movement in the same direction — doesn’t play out. But there are a lot of possible ways that we could define and measure momentum, and looking at week-by-week changes in national polling in this way is only one of them. Perhaps there is a more useful way to define momentum that is more predictive or better captures what people mean when they talk about momentum. But looking at week-by-week changes in national polling, there does seem to be something real behind at least some discussions of candidates’ supposed momentum, whether it be the result of media coverage, effective campaigning, or something else.
There are also a lot of considerations that I didn’t take into account that might be worth looking at to get a really good idea about momentum in the pre-Iowa phase in primaries. There is of course the inherent noisiness of the available polling averages for past national primaries. National polling might not even be the best place to look; voters in early states are likely more engaged with the primary at this early stage and might be a better place to look for trends. Weeks may the wrong time increments to find these trends as well, especially early on in the primary when there often isn’t as much news or attention put on the candidates. Additionally, higher polling candidates naturally experience larger swings in the polling average, which might be worth factoring out; for instance, a candidate polling at 5% will experience about half the polling movement in a week on average as a candidate polling at 30% (about .76% compared to 1.46%). There are also plenty of statistical tools that we could use to evaluate the data more thoroughly that I didn’t bother implementing. Certainly there are many other things worth noting…