Measuring a Convention Bounce

Polling generally becomes more accurate as you get closer to Election Day. The exception to the rule is in the period immediately after the party conventions, when the polls can be a wild ride.

In some past elections, candidates have received as much as a 25-percentage-point lift in the polls after their convention. A bounce of that magnitude is unlikely this year; in fact, there is reason to think that the change in the polls will be quite modest. Nevertheless, this will make poll-reading more difficult.

The FiveThirtyEight forecast will build in an adjustment for the effects of the conventions. It’s possible that a candidate could gain ground in the polls but lose ground in the forecast if his bounce is below average. The model will take it as a bad sign for Mitt Romney, for instance, if he fails to pull ahead of President Obama in polls conducted in the brief window between the Republican and Democratic conventions — even if he gains a point or two.

Historical Convention Bounces

Let’s start by looking at how the polls progressed in past election cycles in the period surrounding the conventions. (We’ll go back to 1968, which is when our polling database becomes more robust.)

The way we look at this data is going to be a bit different than we do ordinarily. Instead of thinking of the candidates as Republicans or Democrats, we’ll break them down into incumbent and nonincumbent groups.

For these purposes, a candidate counts as the “incumbent” if his party controls the White House at the time of the election, even if he is not the president himself. So, for example, in 2008, John McCain is listed as the incumbent candidate since another Republican, George W. Bush, was president at the time, and Mr. Obama as the challenger.

In the charts that follow, the letter “I” and the color black designate the incumbent party; the letter “C” and the color orange designate the challenger. During this time period, the challenging party always held its convention first. (The challenging party is sometimes also designated as the “out-party” — as opposed to the incumbent party, which is also called the “in-party.”)

For each election cycle, I’ve taken an average of the national polls in our database during five different periods surrounding the conventions. I’m presenting this data in two slightly different ways.

The first table simply reflects the polling average. So, for example, the designation “C +5″ indicates that the challenger was ahead by five points, on average, in polls conducted during that period.

The second version of the table evaluates the bias in the polls, comparing the polling during the convention period against the actual election result. If you see the designation “C +5″ in the second chart, what that means is that the challenging candidate polled five points higher at that time than his actual finish on Election Day. In other words, the polling was biased toward him by five points because of the convention bounce.

All numbers are taken on a net basis: they reflect the margin between the candidates. If the Democrat improves to 47 percent from 43 percent in the polls after his convention (a four-point gain), and simultaneously, the Republican declines to 49 percent from 53 percent (a four-point loss), we’d describe that as an eight-point bounce since the Democrat went from being a net of 10 points behind to just two points behind instead.

The first column in the charts, Column A, looks at the polls in the three weeks immediately preceding the challenging party’s convention. Since the challenging party always held its convention first, this gives us a sense for what the polling baseline was before the conventions. In 6 of the 11 years in our study, the challenging party held the polling lead before the conventions, although it was modest in most of these cases.

The next time frame in Column B is more critical: it consists of the polls in the two weeks immediately after the challenging party’s convention (but before the incumbent party’s convention). This is when it would ordinarily be at the height of its convention bounce.

The only challengers who still trailed immediately after their conventions were George McGovern in 1972 and Bob Dole in 1996 — even Walter Mondale, in 1984, had somehow managed to pull into a tie with Ronald Reagan.

Almost all of the challenging candidates, also, polled higher during their convention bounce than they eventually realized on Election Day. The exceptions? Mr. Obama in 2008, who got a small convention bounce and was ahead by only about four points after the Democratic convention in Denver that year. He eventually won the election by seven percentage points instead. And in 1996, Mr. Dole trailed by nine percentage points after his convention, the same margin by which he eventually lost to Bill Clinton.

On average, however, the polls just after the challenging party’s convention overrated their standing by a whopping 14 points relative to their actual finish. Even candidates who got a lot of momentum out of their conventions, like Mr. Clinton in 1992 and Mr. Reagan in 1984, saw inflated polling during this time period.

The third column, Column C, consists of an interim period: those polls conducted more than two weeks after the challenging party’s convention, but before the incumbent party’s. This category does not apply in cases where the conventions were separated by two weeks or less.

In all years where this data is available, the bounce had faded somewhat after a couple of weeks. But it hadn’t faded completely, as six of the seven challenging candidates still polled better during this period than their Election Day finish.

The fourth category, Column D, is an average of polls in the two weeks right after the incumbent party’s convention. The incumbent party led after its convention in 7 of the 11 years; it trailed in 1968, 1976 and 1992, and Jimmy Carter and Mr. Reagan were roughly tied after the Democratic convention in 1980.

But it gets a little tricky when we want to describe the size of the incumbent party’s convention bounce, since it’s not clear what baseline we should be measuring it against.

As compared with where the polls stood at the peak of the challenging candidate’s convention bounce, the incumbents had gained back a lot of ground — 15 percentage points on average.

However, they had gained much less ground — only three or four percentage points — relative to where the polls stood before either party’s convention.

Moreover, the polls conducted during this period did not exaggerate their standing all that much. The incumbent was ahead by an average of five percentage points in polls conducted just after his convention, but eventually won the election by an average of three points, only a modest difference.

This may simply reflect the fact that the incumbent’s convention is often held at a time when the challenger’s convention was recent enough to still have some effect on the numbers. That is, the incumbent’s convention bounce is partly canceled out by the aftereffects of the challenger’s. By contrast, the challenger gets a window of time when he pretty much has the stage to himself, until the incumbent holds his convention.

Finally, Column E consists of the polls in the period two to five weeks after the incumbent party’s convention, when the effects of the convention bounce should be fading.

The polls during this period are not that much different from those in Column D. Nor were they much different from the eventual election results on average. The biases that the conventions introduce into the polls seem to have worked their way out of the system by this point in time.

Does the Convention Bounce Predict Election Results?

So it’s really the period just after the challenger’s convention, but before the incumbent’s convention, when the polls can be the most skewed.

Do the polls conducted during this time frame have any predictive value? Or are they so far from the mark that it’s better to stick with preconvention polls, then wait until both parties have had their conventions before you start paying attention again?

Actually, the answer is a little unclear.

If you run a regression analysis, it suggests that the postconvention polls probably have a bit of predictive power, but ought to be discounted fairly heavily. You should still be looking at the preconvention polls as well.

However, it’s hard to tell because the postconvention and preconvention polls are fairly highly correlated with each other. When variables are highly correlated, regression analysis is less useful, especially with only 11 elections to evaluate to differentiate between them.

On a more anecdotal level, there have certainly been cases, as in 1980 and 1992, when the challenging party got a huge convention bounce and rode to victory. But there are also exceptions to the rule: Mr. Mondale’s fairly big bounce in 1984, which was very short-lived, for instance. In 1988, Michael Dukakis was already ahead in the polls before the conventions and then got a pretty big convention bounce, but wound up losing. In 2008, conversely, Mr. Obama got a rather small convention bounce — Mr. McCain and Sarah Palin got a bigger one — but Mr. Obama ended up on top.

Is the Size of the Convention Bounce Predictable?

In this section, I’m going to focus on the challenging party’s convention bounce. As I mentioned, it is much more straightforward to evaluate the challenger’s bounce than the incumbent’s, since what we think of as the incumbent’s bounce may really just reflect the challenger’s bounce fading. Are there patterns in which challengers tend to get bigger or smaller bounces?

One hypothesis is that the challenger tends to get a bigger bounce when the economy is poor. But in my view, that’s not really reflected in a consistent way in this data. Mr. Clinton in 1992 and Mr. Reagan in 1980 got huge bounces amid a bad economy. But Mr. Obama got very little bounce despite a bad economy in 2008. And some challengers, like Mr. Mondale in 1984 and Mr. Carter in 1976, got big bounces despite a good economy. There just doesn’t seem to be much of a consistent relationship.

Another theory is that the challenger tends to get a bigger bounce when he is less well known to voters. This idea fits well with the examples of Mr. Carter in 1976 and Mr. Clinton in 1992, both of whom were fairly obscure figures before winning the Democratic primaries that year, and both of whom got very large bounces.

But there seem to be just as many exceptions. Richard Nixon in 1968, and Mr. Mondale in 1984, were extremely well known to voters in those years, having been former vice presidents. Both got pretty big bounces despite being known commodities. John Kerry, in 2004, was not all that well defined to voters before the Democratic convention that year, but he got almost no bounce despite a lot of emphasis on his biography at that convention.

One thing that seems to be clearer is that the convention bounces are getting smaller. Mr. Kerry and Mr. Obama got very small bounces. And the challenging candidate in 2000, George W. Bush, got only a modest-size one.

By contrast, six of the seven challengers between 1968 and 1992 got double-digit bounces, the exception being Mr. McGovern in 1972.

This likely reflects some ways in which the political climate is changing. With news coverage being what it is, we are living more and more in a “perpetual campaign.” There is certainly a decent share of voters who will start paying a lot more attention to the campaign once the conventions take place. But it’s not going to be quite as night-and-day as it was 20 or 30 years ago, before the advent of the Internet and 400-channel cable lineups.

Perhaps more important, partisanship has steadily increased during this period. There are fewer and fewer true swing voters, and it is harder to move the polls for any reason, with so many people locked into voting Democrat or Republican from the get-go.

Stable Polls, Smaller Bounces

In fact, there is a relationship between how much the polls move before the conventions and the size of the convention bounces. The more stable the polls are before the conventions, the smaller the magnitude of the convention bounce.

In the chart below, I’ve listed the standard deviation of all national polls conducted between November of the pre-election year, and the challenging party’s convention. The higher this number, the more volatile the polls are. Then I’ve compared it against the size of the challenging party’s convention bounce.

The relationship is not perfect, but it’s reasonably strong, explaining about 40 percent of the variance in the convention bounce. The four years with the most volatile polling before the conventions — 1980, 1976, 1988 and 1992 — all produced large convention bounces. The two years when the polling was the most stable, 2000 and 2008, produced fairly small ones.

But what about this year? The national polls have been exceptionally stable, dating back to late 2011. Except for a period in February and March when Mr. Romney was struggling in the Republican primaries, they’ve pretty much always showed something on the order of a two-point lead for Mr. Obama.

In fact, as measured by the standard deviation of national polls since November, this has been the least volatile polling year ever. The standard deviation in the national polls has been 3.7 percentage points, as compared with a historical average of 6.7 points.

What this means is that we should probably expect fairly small convention bounces. The polls have remained quite steady through events like the Supreme Court’s ruling on Mr. Obama’s health care bill, various discouraging and encouraging economic reports, and Mr. Romney’s selection of Representative Paul. D. Ryan of Wisconsin as his running mate. The conventions will probably produce some effect, but it would be an upset if it were all that dramatic given how immovable the polls have been so far.

Specifically, the benchmark is for a convention bounce of about four points for each candidate. Since Mr. Romney trails Mr. Obama by about two percentage points now, this implies that he’d move ahead of him by about two percentage points instead in polls conducted immediately after this week’s convention in Tampa, Fla.

If Mr. Romney gets a larger bounce than that — say, pulling ahead of Mr. Obama by perhaps four to six points in the national polls — that would be a bullish sign for him. This wouldn’t reflect all that large a bounce as compared with the historical average. However, it would be pretty impressive given how hard it has been for the candidates to move the numbers for any reason at all this year.

By contrast, if Mr. Romney only pulls into a tie with Mr. Obama, or still trails him in polls conducted this weekend, that would be a troubling sign for his campaign. The track record of challengers who failed to lead after their conventions is not good.

When Two Bounces Overlap

In 2008, I modeled the convention bounces as a function of the number of days that had passed since each party’s convention began. The convention bounce seems to peak a day or two after the party finishes its convention, then fades slowly but steadily.

This year, I’ve modified the analysis to reflect the factor that I described above: that we expect smaller-than-average convention bounces because of the low volatility of the polling up to this point in time.

However, this is further complicated by the fact that there are only a few days separating the party conventions this year. We’ll barely have had time to benchmark Mr. Romney’s convention bounce before the Democrats start their convention in Charlotte, N.C., which could counteract its effects.

Our approach to this problem is to assume that each candidate will get a convention bounce of the same size and shape. (It once was the case that the challenging party seemed to get a larger bounce than the incumbent party, but that wasn’t true in 2000, 2004 or 2008.) Those bounces might look something like this:

Note, however, that the bounces overlap with each other. We’d expect Mr. Romney to still be experiencing some of his convention bounce at the same time that Mr. Obama begins his.

If you take the superposition of the two convention bounces — meaning, essentially, that you add them together — you wind up with a funky looking chart like this:

We’d expect the polls to initially move by about four points toward Mr. Romney this weekend, reflecting the effects of the Republican convention. However, the bounce would be short-lived, since the Democrats will begin their convention on Tuesday.

Then Mr. Obama gets his convention bounce. However, it would be smaller than Mr. Romney’s since not all that much time will have passed since the Republican convention, canceling it out in part.

The benchmark is that we might expect Mr. Obama’s polling just after Charlotte to be one or two points better than it was before either party’s convention, meaning that he’d be ahead in national polls by three or four points.

This set of assumptions squares pretty well with the historical data, in which the challenging party gets a fairly clear bounce, but the incumbent party’s is muted. Still, it would be a poor sign for Mr. Obama if he failed to lead in the polls after his convention in Charlotte.

Adjusting for the Convention Bounce

As I mentioned, the FiveThirtyEight forecast will adjust for the convention bounces.

Mathematically, this adjustment is not complicated: we’ll just subtract out the projected effects of the convention from the polls.

So, for example, if Mr. Romney holds a five-point lead in a poll conducted this weekend, the model will instead treat that as a one-point lead. This is because this should be right in the midst of his convention bounce and because our benchmark for his convention bounce is four points this year.

This also implies that if Mr. Romney does not gain any ground in the polls after this convention, the model will treat that as a significantly negative factor for him. Then Mr. Romney will have to hope that Mr. Obama does not get any convention bounce either.

I know that some readers will not like the convention bounce adjustment. But it’s critical to remember that a convention bounce is almost always temporary. Sometimes, the bounce will be large enough that a candidate will go from a losing position to a winning one even after it fades, as Mr. Clinton did in 1992. But the polls almost always overrate a candidate’s standing in the interim.

We do hedge against the convention bounce adjustment in two ways, however. First, in addition to adjusting the polls during the convention period, we’ll also assign them less weight. (Specifically, polls conducted when the absolute value of the convention bounce is at its highest will receive a 50 percent penalty, with everything else scaled accordingly.)

More important, we have an alternative for those of you who prefer to look at the polls on an “as-is” basis. That is what we call our “now-cast,” which is our estimate of what would happen if the election were held today. We give the now-cast less emphasis than our forecast, but you can find it by clicking the “now-cast” tab in the right-hand column of this page.

The now-cast will not apply the convention bounce adjustment. It’s possible that one of the candidates could pull ahead in the now-cast while losing ground in the forecast, if he gets some sort of a convention bounce, but it’s a below-average one.

The now-cast also differs from the forecast in other ways. It does not include any economic variables, as the forecast does. It weights the most recent polls somewhat more heavily. And it has a smaller margin of error associated with it, because there is less uncertainty about what would happen in an election held today than the one that will take place in November. But it should give you more options for how you evaluate our data.

Still, the forecast is our signature product. We expect the convention bounces to be small this year. But if Mr. Romney gets no convention bounce at all, or a bounce of only one or two percentage points, it will be appropriate to take a more pessimistic view of his chances of winning in November.