Skip to main content
ABC News
Oct. 4: Too Soon to Gauge Impact of Debate on Polls

This might be bad for business — but you probably ought not to pay too much attention to the numbers you see in the right-hand column of this blog over the next day or two.

It’s just too soon answer the question of what impact Wednesday night’s debate in Denver, which instant-reaction polls judged to be a clear win for Mitt Romney, will have on the head-to-head polls.

The Gallup and Rasmussen Reports national tracking polls as they were published on Thursday, for instance, reveal nothing at all about the debate, since only a tiny fraction of their interviews were conducted after its conclusion.

To the extent there was a tempting nugget of information, it was from the Ipsos online tracking poll. That survey broke out results from interviews that it conducted after the debate, and found Mr. Romney trailing Mr. Obama, 43 percent to 48 percent, among those voters.

It’s somewhat confusing what Mr. Romney’s five-point deficit in the Ipsos poll should be compared with. However, since the Ipsos poll has been reasonably volatile, the better approach might simply be to compare the five-point lead that Mr. Obama had in the poll’s post-debate interviews against the medium-term average of the poll. Ipsos has had Mr. Obama up by six points on average since the Democratic convention concluded, slightly larger than in the average of other surveys. By that measure, only closing the deficit to five points is not a very impressive result for Mr. Romney.

However, the volatility of the poll — coupled with the fact that the post-debate interviews consisted of a fairly small sample of about 500 people — means that relatively little should be read into it. Ipsos lists the margin of error on its post-debate sample of likely voters at 5.2 percentage points. And that 5.2 percent figure only reflects the margin of error associated with any one candidate’s numbers; the margin of error on the difference between the candidates is roughly twice that, or plus-or-minus 10 points.

We don’t always emphasize the margin of error on the site because it becomes much smaller when a number of different polls are combined together. (There are some other reasons we de-emphasize the margin of error as well, but that probably requires a fuller discussion.)

But this is the textbook case where we have just one post-debate poll, the Ipsos poll, so the margin of error becomes very relevant.

For the time being, a better approach to estimate Mr. Romney’s post-debate bounce may be to compare the historical relationship between instant-reaction polls of the debate with their eventual effect on the horse-race numbers. That technique would estimate Mr. Romney gaining a net of 2.2 percentage points on Mr. Obama. But this calculation has a margin of error as well — about 3.5 percentage points in estimating the change in the margin between the candidates.

The most likely range of outcomes, however, is Mr. Romney gaining from one and four points on Mr. Obama once the effects of the debate are fully accounted for.

Of course, there is a huge difference depending on where Mr. Romney might fall within that range. If he adds four points, the race will be very nearly tied.

If he gains just one point, conversely, the situation will look rather poor for him: he’ll have had what was almost certainly the best night of his campaign, and not gotten very much out of it.

It is also worth paying attention to whether any gains for Mr. Romney come directly from Mr. Obama, or instead from the undecided voter column.

Mr. Obama appeared to lead about 49 percent to 45 percent in national polls conducted before the debate. (His lead was slightly wider in some important swing states.)

Say that Mr. Romney gets a marginally disappointing result, adding a net of two points. If Mr. Romney’s gains came from undecideds, and he was down 49-47, that would still leave Mr. Obama extremely close to clinching the race with 50 percent of the vote, assuming that the polls were right.

The healthiest sign for Mr. Romney will be if he can reduce Mr. Obama’s share of the vote in the polls to 48 percent, or 47 percent or lower.

The polling evidence we have so far, however, is just not rich enough to adjudicate between these different outcomes. And it may take the forecast model a few days to catch up to Mr. Romney’s gains, if he’s made them. Plus, there are other important events in the news cycle, like the release of Friday morning’s jobs report, that could further complicate the analysis.

So have some patience, poll-watchers. There are some points in each election cycle where the amount of news exceeds the amount of polling evidence that we have to analyze it. Because a number of major economic reports are accounted for directly by the model, we can sometimes get a head start on those — but that isn’t true for noneconomic events like the debates.

At the same time, the debate was potentially important enough that we can’t just treat is at the routine sort of uncertainty that the model naturally accounts for in its calculations. It will probably have some favorable impact for Mr. Romney — we just don’t know how much yet.

Nate Silver founded and was the editor in chief of FiveThirtyEight.


Filed under