Skip to main content
ABC News
Aug. 9: National Polls Shouldn’t Panic Romney

President Obama got a couple of very good-looking polls on Thursday. A CNN poll put him seven points ahead of Mitt Romney in the national race. And a Fox News poll gave Mr. Obama a nine-point lead.

Some polling firms, like Pew Research, have consistently shown strong numbers for Mr. Obama. When a polling firm like that prints another strong number for him, it isn’t necessarily news.

Mr. Romney cannot use that excuse in the case of the Fox News and CNN polls. CNN’s surveys so far this cycle have had essentially no partisan lean, whereas Fox News’s polls have shown a modest Republican one. Mr. Obama’s number was strong not just in an absolute sense, but also relative to their earlier polls.

In the Real Clear Politics average of national polls, Mr. Obama’s lead grew to 4.4 percentage points on Thursday. That’s his largest lead by its method since early April, when Mitt Romney was just starting to recover from a rough stretch in the primary campaign.

And yet, despite all that, I think the importance of these new polls could easily be overstated — and probably will be by some in the news media.

One problem is that the trend toward Mr. Obama in national polls has hardly been uniform. The Gallup national tracking poll has shown a very flat race. And Rasmussen Reports had a swing toward Mr. Romney in its release on Thursday, with his pulling ahead by four points in its survey. Weekly tracking polls from Public Policy Polling and YouGov have also not shown especially good numbers for Mr. Obama lately. If there had really been a shift in the race of the magnitude that the Real Clear Politics average implies — it has Mr. Obama gaining three points on Mr. Romney over roughly the past 10 days — we probably wouldn’t be seeing these contradictory data points.

Perhaps more importantly, there hasn’t really been a lot of news to drive something on the order of a three-point swing toward Mr. Obama. There have been far more momentous news events at earlier stages of the campaign, like the Supreme Court’s ruling on Mr. Obama’s health care bill, or the set of poor jobs reports in April through June, that didn’t seem to move the numbers much at all. So your default position should be one of skepticism toward the numbers having moved very much.

Might Mr. Obama have gained half a point, or a point, based on whatever residual factors that voters are thinking about? Sure, and Mr. Romney would rather that movement be in his direction instead. But has Mr. Obama gained three points, at a time when most ordinary Americans are watching the Olympics? Probably not.

From a more technical standpoint, my research suggests that one should not be overly eager to chase down the latest trend at this stage of the campaign. Some of our friendly competitors, like Talking Points Memo, use a technique called Loess regression to calculate a polling trendline. Our methodology is not quite the same as theirs, but we do use a version of Loess regression to calculate our trendline adjustment.

Loess regression is a nice technique, but one of its quirks is that you have to manually choose what’s called a “smoothing parameter.” You can set the smoothing parameter to be highly sensitive to new data points — a couple of new polls will swing the trendline a lot — or relatively insensitive to them, where a change might have to be sustained for a couple of weeks to make much difference.

In designing this year’s model, I went back and looked at what choice of smoothing parameters would have produced the most accurate predictions in past presidential elections dating back to 1972. The research suggested that the answer actually depends on which stage of the campaign you’re in.

In the late stages of the campaign — the final six weeks or so before the election — it’s worthwhile to be reasonably (but not excessively) aggressive. More and more voters will be paying attention to the campaign, and making up their minds once and for all about whom they’re going to vote for. Polling gets more accurate as you get closer to the election, but the trend is not linear: it gets a little bit more accurate throughout the summer months, but then a lot more accurate within these final six weeks.

Right now, however, it’s better to be quite conservative. A poll released on Aug. 9 is just not going to be all that much more informative than one released on July 9, especially if there haven’t been major intervening news events during the period. The trendlines you see at some other Web sites are probably fluctuating upward and downward much more than they should.

Loess regression is a fairly “smart” method, and even a conservative application of it could still potentially perceive a shift in the polls in the summer. Certainly, if the national tracking polls had also shown a shift toward Mr. Obama, we would be seeing some fireworks in the model right now.

Even then, however, national polls are just one of the things you should be looking at to make an election forecast. The other critical factors are state polls and the economic numbers. Our model hedges against the national polls by combining the state polls with them, and then also, hedges against the polls in general by combining them with our economic index. Of these three things, the national polls probably have the least influence on the forecast at this stage of the campaign.

To be clear, the model has shown a favorable trend for Mr. Obama lately. His chances of winning the Electoral College rose to 73.3 percent on Wednesday, a new high. His projected margin of victory over Mr. Romney in the national popular vote, 2.8 percentage points, is also a new high.

But the shift is more because the consensus of evidence has been slightly favorable to Mr. Obama than any one piece of evidence pointing toward a major inflection point in the race.

The economic numbers are still very mixed, but the jobs report was better in July, personal income growth has been accelerating and the stock market has been rallying. I’m certainly not going to render a prediction about the long-term future of the euro zone, but it now seems less likely that it will blow up soon enough to substantially affect the election in November. (Perhaps whichever candidate wins will have a huge mess on his hands in 2013 instea

Mr. Obama has also had a fairly strong set of polls from swing states. It’s just not a good sign for Mr. Romney that he’s down in almost all polls of Ohio, and more often than not, in polls of Florida and Virginia as well — although he has posted some better numbers in Colorado recently.

Finally, and despite my earlier caution about how time in a campaign should be measured on a relative rather than absolute scale, we are seeing some time tick off the clock. Mr. Obama is almost certainly ahead right now, so any day that the status quo is preserved is basically a good one for him.

Despite that, it is doubtful that Mr. Obama leads by as many as four or six points now (as some other polling aggregation Web sites suggest), and even more doubtful that he is seven or nine points ahead.

In addition to the issue I mentioned earlier about potentially placing too much weight on new data points, most of the other polling sites do not adjust polls on the basis of whether they were conducted among registered voters or likely voters. Our model does, and it assumes — based on historical trends as well as what data we have from this year — that Mr. Romney will perform about two points better in polls of likely voters than polls of registered voters.

This will not be a big issue in the state polls, the vast majority of which are already being conducted on a likely voter basis. But most of the recent national polls, including the Fox News and CNN polls, reported results among registered voters instead. If you take the Real Clear Politics national polling average and apply our likely voter adjustment to it, it would place Mr. Obama ahead by somewhere in the neighborhood of two-and-a-half or three points right now, which is almost exactly how our model sees the race.

This is not to suggest that Mr. Romney has nothing to worry about. Polls like the Fox News survey may be outliers, but as I noted on Wednesday, even the outliers tell you something about the central tendency. The fact that Mr. Obama’s outliers show him as many as nine or 10 points ahead, while Mr. Romney at best gets polls that show him three or four points up, is not a good sign for him.

Another concerning fact for Mr. Romney is that the spread between the horse-race polls and Mr. Obama’s approval ratings has been widening of late. Real Clear Politics — drawing from almost exactly the same set of polls that show a 4.4-point lead for Mr. Obama in the horse race — have Mr. Obama’s approval ratings almost exactly break-even instead.

The split may be caused by voters who account for the personal qualities of the candidates, rather than seeing the race as a pure referendum on the incumbent. Can a president get re-elected with an approval rating of 47 or 48 percent? It’s tenuous, obviously. But if that president’s favorability scores are at or above 50 percent, and his opponent has net-negative favorables instead, it’s not that hard to see how he could be a few points ahead in the horse race.

But concern is one thing for Mr. Romney; panic would be another, and it is not yet warranted. If Mr. Obama is really ahead by somewhere in the mid-single digits, we should be seeing more confirmation of that in the coming days. More likely, this remains a two or three point race instead.

The recent national polls aside, the numbers in this race have been exceptionally stable. That’s actually a bad sign for Mr. Romney: the harder the numbers are to move, the more robust a small lead is. But it’s still a small lead, and the conventions, the debates, and Mr. Romney’s vice presidential selection all lie ahead.

When should Mr. Romney really get worried? There are only a few days between the Republican and Democratic conventions this year, but candidates typically get a bounce after they hold their party conventions, and Republicans will hold theirs first. It’s usually a high-water mark for the opposition-party candidate, and if Mr. Romney does not at least pull into a tie with Mr. Obama during polls conducted during that interim period — and preferably, poll ahead of Mr. Obama by a couple of points — it might be time for him to hit the panic button.

For now, Mr. Romney should be thinking about making the best vice presidential pick that he can, designing a good Electoral College strategy, and figuring out how he can shift the race back toward a focus on Mr. Obama’s economic performance — and not a pair of national polls. Mr. Romney was fairly cool-headed during points in the primary campaign when things seemed to be going poorly for him, and that attitude could pay real dividends now.

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


Filed under