Yglesias posits that Gallup is being scurrilous. And my quick little linear regression indicates that indeed they are. I look at the July polling margin and the actual margin for the popular vote victor in the past fifteen elections.

The low r-square value indicates that the July poll is not the greatest explainer of variation. But the p-value on the regression is statistically significant (0.02) so the direction of the equation (positive) is in all likelihood correct, though the stats nerds will rightfully crucify the small sample size.

Bottom line: it’s better to be ahead in July, but not all that great unless you have a commanding lead.