Looking for signs.

Photographer: Brendan Smialowski/AFP/Getty Images

Don't Trust the Polls. Trust Their Average.

Jonathan Bernstein is a Bloomberg View columnist. He taught political science at the University of Texas at San Antonio and DePauw University and wrote A Plain Blog About Politics.
Read More.
a | A

Yes, it’s time for a back-to-school refresher class on how to read the presidential polls, which at this point are good indications of what will happen in November.

First assignment: What to make of the CNN poll released this morning showing Donald Trump coming out on top, by two percentage points, over Hillary Clinton among likely voters.

You’ve heard it a million times from me and others (such as political scientist Matthew Dickinson and number-cruncher Harry Enten): Look at the polling averages, not individual surveys.

As Greg Sargent at the Plum Line explained it this morning:

Surprising poll results can either be outliers, or can reflect statistical noise or short term fluctuations. Fortunately, we have a remedy for this: The polling averages, which have massive samples that cover longer periods of time and help screen out the noise. Depending on who is doing the averaging, Clinton is up by three (The Upshot), four (Real Clear Politics), or five (Huffpollster).

Those are the main numbers needed to make sense of what’s happening. And they show what they’ve shown throughout the campaign: Clinton remains likely -- not certain, but likely -- to win.  

Looking at individual states and the Electoral College implications right now is more likely to confuse than enlighten. Yes, first-rate projection sites such as Drew Linzer’s at Daily Kos and Nate Silver’s FiveThirtyEight use predictions for each state, but be careful when trying to interpret any individual state's poll. Again, polling averages for states tell us more than any specific survey, and even then these averages can be misleading because there are fewer polls (and sometimes very few quality ones) in many states. 

It’s still the case that the national estimates tell us what we need to know, because the winner of the national vote will almost always win the electoral vote. If there is a popular vote vs. electoral vote split, it’s still too early to see whether one party has an electoral-vote advantage.  

Remember: Man-bites-dog stories always get more coverage, so “surprising” results such as today’s CNN numbers will always get the attention. And probability theory tells us that we absolutely should expect goofy-looking statistical fluctuations from time to time. 

Yes, surveys can sometimes spot something new that is happening with the electorate. But there’s just no way to know from a single poll whether such a thing is really happening.

Individual polls, even from quality pollsters, could be poorly done, introducing more error. Some Democrats on Twitter this morning are going after the CNN poll, just as Republicans have gone after surveys that are good for Clinton. Polling is hard, especially since the response rates are so small these days. It has always been as much art as science, and it involves all kinds of debatable choices.

But questioning the methodology is the wrong way to respond to a poll with unwelcome results. Going down that path – carefully scrutinizing every result one doesn’t like – is certain to lead to biased results.

The right way to deal with a (seemingly) weird poll is the same way one should deal with every poll: Toss it into pool with all the others, and look at the averages.

  1. FiveThirtyEight currently estimates a slight advantage for Trump in how the math turns votes into Electoral College votes. He more likely to win the Electoral College while losing the popular vote than Clinton is likely to lose the popular vote while winning the Electoral College. But it's by no means clear that advantage would hold up if the race does get very close; it would depend on which votes, in which states, moved to Trump.

This column does not necessarily reflect the opinion of the editorial board or Bloomberg LP and its owners.

To contact the author of this story:
Jonathan Bernstein at jbernstein62@bloomberg.net

To contact the editor responsible for this story:
Katy Roberts at kroberts29@bloomberg.net