Skip to main content
Beware of (Temporarily) Falling Polls

I am generally not blogging about national polls at this site, but today’s Rasmussen tracking poll shows a marked decline in Barack Obama’s support after a bad day in Friday interviews. It would seem likely that this is related in some measure to the controversy over Reverend Jeremiah Wright.

This downtrend could be a statistical fluke — but more likely it is real. However, the important question is not if the downtrend is real (i.e. statistically significant), but how long it will last. At the extremes, the goalposts are that it will be over and done with within 24 hours — now that Obama is starting to press his defense — and that it will last throughout the balance of the election cycle.

More Politics

At fivethirtyeight, we tend to take the long view. We don’t look at polls just from the last week. We don’t throw a previous poll from an agency out just because it has published a newer one. We recognize that there is a lot of uncertainty related to all polling data taken this far from an election.

When you see a poll that is ’50’ today, and was ’40’ yesterday a lot of people will be inclined to extrapolate the trendline, and assume that the poll will be ’60’ tomorrow. Over the long run, however, the poll is more likely to go back down to ’40’ than to continue to move upward to ’60’. The very definition of the term ‘bounce’ (and there are negative bounces as well as positive ones) implies that it is something temporary — you go up, and then you go back down.

We see these bounces almost every year around the conventions. We have seen them at various times in the primary campaign. Barack Obama and Hillary Clinton have been pretty much locked in deadstep since Super Tuesday in primary tracking polls. Whenever one has started to pull away, based on the latest events in the news cycle, they have shortly thereafter come back down.

When I studied 2004 polls to come up with the time-based weightings that I use at this site, I found that it was helpful to account for older polls, even when you have newer ones. There were several points in time, in fact, at which the newest polls were less accurate predictors of election outcomes than somewhat older ones.

I recognize my discussion here is a little obtuse — at some point, we will get more quantitative about how to assess polls from different points in time. But the point is, don’t assume that any “bounce” you see in the numbers is a “trend”. More often than not, bounces are fleeting and temporary things. Trends are important too, but they operate at a much slower, glacial kind of level, particularly in general election campaigns.

UPDATE: Gallup shows a more modest 3-point in its primary election tracker, and no change in its general election tracker.

BTW, one thing to keep in mind is that you always have multiple events cycling into and out of the news cycle at any given time. One benefit of the Wright story to Clinton, for instance, is that it took the Ferraro story out of the news — which itself may have been producing a temporary downward bounce in her numbers.

Nate Silver is the founder and editor in chief of FiveThirtyEight.

Comments Add Comment