For a better browsing experience, please upgrade your browser.

Skip to main content
Menu
Flying Blind Toward Hogan’s Upset Win In Maryland

Of all the upsets Tuesday night, one stands above the rest: Maryland governor. FiveThirtyEight’s gubernatorial model projected that Democrat Anthony Brown would defeat Republican Larry Hogan by 9.7 percentage points — Brown was a 94 percent favorite.

In fact, Hogan beat Brown by 5 percentage points. Our forecast wasn’t even close.

More Politics

What happened? First, 94 percent favorites are supposed to lose sometimes (6 percent of the time, to be exact). Hogan’s chance of winning was roughly equivalent to the chance No. 14 seed Mercer had of beating No. 3 seed Duke in the NCAA men’s basketball tournament this past March. Underdogs can win.

But there was also a problem more specific to Maryland: Our gubernatorial model relies on polls, and polls alone, and the most reliable public pollsters stopped surveying the race a month before Election Day. The final surveys from The Baltimore Sun and Washington Post, taken between Oct. 2 and Oct. 8, had Brown ahead by an average of 8 percentage points.

The only late polls came from the Maryland Republican Party or from Hogan’s campaign. FiveThirtyEight has a policy not to use partisan-sponsored polls because they historically have shown a bias toward the party or campaign that commissions them. In many cases, partisan sponsors quash surveys that look unfavorable to their candidate. We never even hear about those.

In the case of Maryland, it’s clear the polls sponsored by Republicans caught Hogan’s rise. The final poll taken a week before the election by Wilson Perkins Allen Opinion Research (WPA) for Hogan’s campaign gave the candidate a 5 percentage-point lead. This was up from the 1 percentage-point Hogan deficit that WPA found in mid-October, and a 5 percentage-point deficit in September.

It’s unlikely, however, that including WPA surveys in our model would have made a major difference. The Huffington Post Pollster model included these surveys, and it still had Brown as a 92 percent favorite. There simply weren’t enough late surveys for a model to trust the late surge WPA was showing.

What could have made a difference was if the public pollsters had kept polling. (Good polls are expensive to take, so it’s difficult to blame media organizations from wanting to reallocate their resources elsewhere.) If the public pollsters shifted by the same amount as WPA did in the final few weeks, then they would have shown a closer race. There still would have been a polling error given Hogan’s large victory, but it would been cut by a substantial margin.

The model, recognizing the lack of up-to-date polling data, did have a wider band of uncertainty around its projection, but not enough to include a 5 percentage-point Hogan win.

Harry Enten is a senior political writer and analyst for FiveThirtyEight.

Filed under , , , ,

Comments Add Comment