You’re going to be hearing about a lot of horserace polls over the next two months as the presidential and down-ballot races sprint to the Nov. 8 finish line. So avoid making a basic and common mistake in reading them.
Every poll comes with a margin of error. It is frequently in the three-to-four point range. But that’s three to four points for each candidate in the race. So in a two-person race, the actual margin of error is double the reported number.
Example: The Observer and the (Raleigh) News & Observer reported Thursday on a new poll from Emerson College Polling Society. It found Sen. Richard Burr leading Democrat Deborah Ross, 45-41. “Burr’s lead is barely outside the poll’s 3.4 percent margin of error,” the story said.
Wrong. The 3.4 percent margin of error means that Burr’s true support could be 3.4 percentage points above or below his 45 percent. Ross’s true support could be 3.4 percentage points above or below her 41 percent. So to be outside the margin of error, Burr would have to be at least seven points up.
“Margin of error” is shorthand for a margin of error at a 90 or 95 percent (for example) confidence interval. A 95 percent confidence interval means that if the pollster did this poll 100 times, 95 of those polls would come back with the candidates’ numbers within, say, 3.4 percentage points of what this poll found.
So double the reported margin of error to know whether the leading candidate is truly outside it.
Sorry to get wonky on you. But if you’re going to read, or cite, polls in the Trump-Clinton race, the McCrory-Cooper race or any other, you should at least know what the polls are truly saying – and what they’re not.