On July 15, Election Lab, The Post's election model, gave Republicans an 86 percent chance of winning the six seats they needed to take over the Senate majority. Today -- 50 days later -- it gives Republicans only a 52 percent chance of winning the majority. So, how did the model go from predicting a sure-thing Republican majority to now calling the fight for the majority a statistical toss up?
I put that question to John Sides, a political science professor at George Washington University, contributor to The Post's Monkey Cage blog and one of the three-headed political science monster who built Election Lab. (Eric McGhee and Ben Highton are the two other heads.)
Here's what he told me:
[It's] not that races have narrowed, but that the model has begun weighting information differently -- mainly by (a) incorporating polling data (where possible) after the relevant primaries, and by (b) increasing the weight that polls have in the forecast. What this suggests is that in several states, Democrats are arguably 'out-performing' the fundamentals. This doesn't always translate into a high chance of the Democrat actually winning (see: Kentucky) but it does help the Democrats' overall chances of retaining a majority.
To understand Sides's point, it's important to understand how models work. At the start of an election cycle, the model is based heavily on underlying fundamentals of past elections. It's almost an entirely generic calculation that takes little account of candidates, polling etc. There's a reason for that, of course. Early in election cycles there often aren't candidates in races yet and, therefore, polling is limited. As the cycle gets into its latter stages -- like now -- the model tilts to rely more heavily on candidates and polling and less on fundamentals. (That doesn't mean the underlying fundamentals don't matter at all in the model. They just matter less and less as the election gets closer.)
And, as Sides notes, Democratic candidates are currently overperforming how past history suggests they should be doing in a number of races. In a trio of states that has caused significant movement in the odds in Democrats' favor over the past month:
* Georgia: The Election Lab forecast has moved in Democrat Michelle Nunn's favor by 18 percent.
* Iowa: The model has moved toward Democrat Bruce Braley 16 percent of late -- a move that Sides attributes to "the additional weight the model is putting on polls." He adds: "There hasn't been much movement in the polls per se, but because Braley polls a little bit better than the structural model would predict, the additional weight on the polls is enough to improve the forecast for him."
* Louisiana: The Election Lab forecast has moved in Sen. Mary Landrieu's (D) direction by 13 percent, largely due to the fact that Landrieu's polling is overperforming what the historical model would suggest AND the polling is getting more weight in the model.
Now, all of those changes do not mean Democrats will win these races. In Georgia, for example, Nunn's chances went from basically zero in the model a month ago to 18 percent now. Ditto in Kentucky where Democrat Alison Lundergan Grimes is overperforming the fundamentals in the state but Sen. Mitch McConnell (R) still has a 94 percent chance of winning, according to Election Lab.
But, the current Democratic over-performance DOES, collectively, make the odds of a Republican Senate takeover less likely. Not unlikely. But definitely less likely.