The Senate, seat by seat

Why this forecast

Republicans need 23 of 34 seats to keep two-thirds. Adding up every seat’s chance, the model expects them to hold about 22.5, counting 10 Senate seats not on the ballot.

• Of the 17 seats on the ballot, 3 are very likely Democratic (1 with no Republican on the ballot), 11 very likely Republican, and 3 are closer than 95 to 5.
• Seats move together: each simulated election has one statewide swing, one for each region and one for each type of district, so a good night for one party tends to help it in many seats at once (D-06, D-21).
• The Senate sits almost exactly on the line: the expected count, 22.5, is just under 23, so a small swing in either direction decides it. In the 46,046 of 100,000 simulated elections where Republicans keep two-thirds, the statewide mood came in 0.9 pts less Democratic than the central estimate; in the rest, 0.8 pts more Democratic.
• Polls of the St. Louis congressional race (MO-2) ran 2.1 pts above its fundamentals. The model treats that as weak evidence about Missouri’s mood: +0.32 pts in every legislative seat, a little more in St. Louis-area seats.
• Polls of two Senate districts (SD 8 and SD 30) ran 2.9 and 0.8 pts more Democratic than the model expected there. Read together with the MO-2 polls, they add +0.13 pts to every legislative seat and more to the two polled seats (D-25).
• Both chambers come from the same simulated elections, so they move together. If the two were unrelated, Republicans would keep both in about 4.9% of elections. Because a good Republican night helps them in both, it is 9.5%. When Republicans keep the Senate, they also keep the House 21% of the time.

What the model does not know. Two Senate-district polls (SD 8, SD 30) are used as noisy observations (D-25); there are no House-district polls. Candidates’ fundraising, local news and endorsements do not change legislative expected vote shares. Recorded defensive committee spending can widen uncertainty in selected seats without moving the expected share (D-27). Outside ratings are shown for comparison only (D-09); third-party candidates are ignored (D-08). The model also cannot know about a late event that moves many districts at once, beyond the spread it already allows for.