Skip to main content
Menu
Internal Polls Are Usually Bunk

The political world’s perception of Tuesday’s Republican primary for U.S. Senate in West Virginia is largely being shaped by a pair of stories — one in Politico, the other in the Weekly Standard — claiming that Don Blankenship, the coal magnate who spent a year in prison for safety violations relating to the Upper Big Branch Mine disaster in 2010, is surging in the race’s final days. That’s a problem: Both stories relied on that most dangerous of sources, the biggest tease in politics: the internal poll.

Internal polls — politico-speak for polls conducted by a campaign or another entity with a stake in the race — have their uses. Indeed, in this case, we don’t have any recent public polls of the race. But you should take internal polls with several grains of salt. (FiveThirtyEight editor-in-chief Nate Silver would recommend a whole truckload of Morton.) Specifically, it’s important to remember that by definition, internal polls are shared by people with an agenda, and you need to take that agenda into account when looking at the results. In the West Virginia case, the agenda appears to be to marshal the non-Blankenship vote, but I’ll get to that in a moment.

Usually, the goal in leaking an internal poll is simple: make your candidate look good. In practice, that means that internal polls typically paint a rosier picture for the candidate conducting them than exists in reality. For example, in interparty U.S. House races from 1998 to 2014, internal and partisan polls conducted during the last three weeks of the race overstated their party’s candidate by 4 or 5 percentage points, on average.1

Instead, it’s better to form opinions from internal polls not based on their hard numbers per se, but on the context of their release. Let’s say that a candidate sends out a press release trumpeting that the race is “WITHIN THE MARGIN OF ERROR!” accompanied by a poll showing that he’s trailing 45 percent to 44 percent. That poll doesn’t tell you much about the true margin of the race, but if the best a campaign can muster is “we’re close,” you can be fairly confident that that candidate isn’t winning. To take another example, imagine a campaign with a clear front-runner and a scrappy underdog. The underdog airs a campaign ad, and then the front-runner leaks a poll showing herself ahead 55 percent to 35 percent. The poll may or may not be accurate, but the timing of its release telegraphs that the front-runner may actually be concerned about the underdog and wants to stop her momentum before it starts.

Back to West Virginia: Admittedly, there are some respectable reasons to believe these particular internal polls. The Weekly Standard article reported on two sets — likely one from each of Blankenship’s primary rivals, U.S. Rep. Evan Jenkins and Attorney General Patrick Morrisey. Both showed Blankenship with a lead.2 Not only does that consistency (at least as far as where Blankenship stands) make the polls more believable, but these also aren’t your typical internal polls released by a campaign to show how well they’re doing. Indeed, if the rule of thumb above holds true, Blankenship has an even wider lead than these internals claim.

But as Nate pointed out on Twitter, that doesn’t mean the people sharing these polls don’t have an agenda. One of the polls shows Morrisey doing much better than Jenkins; the other shows them roughly tied. It seems likely that the former is the Morrisey campaign’s internal poll, leaked to make the implicit argument that he is Blankenship’s main competition. (Morrisey’s campaign is probably hoping to consolidate the anti-Blankenship vote.) That means the second poll is probably Jenkins’s, released to counteract that narrative and argue that this is still a three-way race. So there is still ample reason to maintain a healthy skepticism of the numbers.

Overall, the political world is often too quick to trust internal polls without questioning their sourcing and thinking critically about ulterior motives. When internal polls are shared anonymously, as in this case, it’s particularly troubling, because figuring out the motivations behind the poll’s release is more difficult.

Footnotes

  1. This may be because of selection bias (i.e., campaigns only releasing the polls that are most favorable to them, essentially using them as glorified talking points), or it may be because campaigns truly believe they are doing better than they are (e.g., because a pollster wants to please their client, or the campaign is simply reading the data in a self-serving way), but regardless, the bias exists.

  2. One poll gave Blankenship 28 percent, Morrisey 27 percent and Jenkins 14 percent; the other went Blankenship 31 percent, Jenkins 28 percent, Morrisey 27 percent. We don’t know which poll was conducted by which campaign, but keep reading.

Nathaniel Rakich is FiveThirtyEight’s Elections Analyst.

Comments