Pre-election polls underestimate the success of women candidates

Mr. Stout and Mr. Kline find that, on average, the polls accurately estimate the vote share of white male candidates: the difference between the poll estimates and the actual election outcome are less than half a percentage point. But for female candidates, the polls underestimate their eventual vote share by over 2 percentage points, on average. For more than half of these candidates, the difference is outside of the margin of error of the polls. When the difference is greater than the margin of error, it reflects an underestimate of the female candidate’s vote share two-thirds of the time. This so-called Richards Effect cannot be explained away with other attributes, like the race of the candidates or the state of the election.

The size of the Richards Effect is larger in states with fewer women in the labor force — which suggests it stems from conservative attitudes about the place of women in politics. This leads to an interesting conclusion. Although the Bradley Effect assumes that people conceal their true opposition to the black candidate, the Richards Effect appears to work the opposite way: people conceal their true support for the female candidate, especially in areas with culturally conservative views about gender roles.