Releasing the @stefanhilts Senate Forecast Model

Well gosh, with the presidential race rigged and all, there’s not as much fun in predicting that right now, except for maybe the exciting prospect of seeing flips in Arizona or Georgia (Go Dawgs!), or of course, a McMullin win in Utah. So with that, and a little hassle from some friends, I spent the week turning my efforts to the Senate races. If I get REALLY ambitious maybe we’ll see some house predictions, but don’t hold your breath.

There are 34 races up for grabs, 10 from currently Democratic seats, and 24 from currently Republican seat, and the current balance of the Senate is 54–46 Republican, so Dems need to snag 14 races to win the balance back.

Without further ado, here’s our model’s probability distribution for total number of Democratic seats in the Senate. 51 takes a full majority; 50 is a tie broken by the VP. Since our Presidential model is predicting approximately 90% for Clinton right now, control goes to the Dems in a large majority of Senate tie cases.

Democrats have a 41% chance of taking control of the Senate

That sums out to a 41% chance of Democrats to win the Senate. Oddly, all our other major poll-watchers are putting this number at about 60%, and I can’t for the life of me figure out how. Looking at the state-level comparison between us all below, most of our numbers are similar, but somewhere in the aggregation, there’s a swing of about 20 points towards the Dems. I assume this is a result of some cross-correlation analysis (meaning that when, say, Pennsylvania goes Dem, all the higher probs do, too), but I would expect this to cut both ways, effectively loading the tails of the distributions. I couldn’t find sufficient detail to replicate the aggregation analysis on any of these sites, so if anyone’s got ideas on this, hit me up @stefanhilts.

Pretty similar, except for Missouri, and the 538’s Kentucky call.

Not many differences between @stefanhilts and the Upshot, but here’s how it aggregates up. Honestly, I feel good about my numbers despite the difference to the more major players. I’m using a very similar build to my Presidential model, but polling data is admittedly more sparse for some of these races. Still, this is a good place to put it all to the test! I’ll see you all on November 9th.