The Anatomy of a Bounce (updated)

We've been hearing a lot about "bounces" in polling lately. But I don't think anyone is being very specific about what they mean. So I thought it was time we talked about it.

First of all, calling a change in election polling a bounce indicates - at the outset - that we don't know whether or not it will last. Most of them don't.

Secondly, during an election, bounces usually occur following a big event, ie conventions, debates, etc. Those are the kind we're pretty sure won't last.

Many people assume that a bounce indicates one of two things: (1) a group that was previously undecided made up their minds or (2) people who had previously supported a candidate changed their minds.

Lets take a look at these two things as they relate to the latest Pew Research Poll since it got the most attention due to the huge bounce for Romney from their previous poll. Among likely voters, the race went from Obama leading 51/43 in mid-September to a Romney lead at 49/45 in early October.

Looking at the percentage of people who said they were undecided, there was no change from Sept. to Oct. It was 6% in both polls.

In terms of movement from one candidate to another, we see that Obama lost ground with Republicans (from 13 to 8 percent) and Romney picked up ground with them (from 88 to 91 percent). Otherwise with both Democrats and Independents, there was no change.

So you have to ask yourself where that 12 point swing from Sept. to Oct. came from. We see no decrease in the number of undecideds and almost no change from one candidate to another.

To answer the question, you have to look at how polling is done.

Polling starts off with calls to a random sample of the population. What makes that a bit more complicated is that only about 10-15% of the people they call actually answer and/or complete the survey (automated polls are even less at about 4-5%). As Nate Silver explains, this is where bounces following major political events occur.

Polling trends can sometimes be odd in reaction to news events. One factor is that supporters of a particular candidate may be more enthusiastic, and more inclined to respond to surveys, after he gets a favorable development in the news cycle...

There is another type of polling bias, however, which is potentially more relevant when there is polling after a major development in the news cycle. Namely, polls are very probably biased toward high-information voters who take more interest in the news and are more likely to respond to political surveys.

In other words, short-term bounces happen in polls after major political events because the sample of people who respond to them changes. That's why conservatives had a point in complaining that the percentage of Democrats included in post-convention polls was too high. Contrary to their conspiracy theories, it wasn't because the liberal media/polling outfits were trying to rig the results. It was just that Democrats were more enthused and therefore, more likely to engage when the pollster called.

The reverse happened after the first debate. You see high levels of white, older, Southern Republicans included in the polling with a bounce for Romney.

Most pollsters will tell you to beware of putting too much weight in polls conducted immediately after a major event. They almost never tell you why. I'd suggest that its because if they did so, we'd recognize that those polls are fairly meaningless in their predictive capacity. Sure - they tell us that Democrats were pumped up after the conventions and Republicans after the first debate. But we pretty much knew that already. Do they tell us much of anything about the overall state of the electorate? The answer to that would be...NO.

In saying all this, I'm not being completely dismissive of polling. I think there is a science to it. Watching aggregates of polls and trends over time can give you information about the electorate you have no other way of knowing. But as I said at the outset...they call them "bounces" for a reason.

UPDATE: If you're interested in a comparison of the party ID numbers in Pew's September and October polls, here's a breakdown. In terms of Democrats, they went from 38% of those surveyed in September to 33% in October. Republicans went from 31% in September to 36% in October. That's a 10% point swing from D to R in the respondents and pretty much covers the 12% point swing from Obama to Romney in the results.