Written on 22 October from the forecast and the coverage as they stood at the end of 21 October.

The Alaska Senate race between Mary Peltola and Dan Sullivan became a dead heat on 21 October. Sullivan's lead in the forecast shrank from 1.6 points to nothing, and each candidate's chance settled at 50%.

A new poll from Data for Progress had Peltola ahead 51 to 49 in a two-way matchup. Alaska elects its senators by ranked-choice voting, which complicates polling there. A voter's first choice may not be the vote that counts, so pollsters often publish both the first round and a simulated final round, and the two can differ by several points.

The governor's race in Alaska is also close, with Tom Begich about a point and a half ahead of Bernadette Wilson. That race has only three polls, so the forecast still leans heavily on the state's voting history.

Overall, the chance of Democrats winning the Senate rose from 60% to 64%, helped by a new CNN poll among others that lifted the national generic-ballot average by about half a point. Part of the rise also reversed the previous day's artificial dip, after a fix to how the model estimates the national environment, described in our note on Ohio.

Nebraska, and the sponsor of a poll

Nebraska's 1st district shows something that surprises people. A new Public Policy Polling survey had the Democrat, Christopher Backemeyer, one point behind Representative Mike Flood. Yet the forecast moved Flood's lead from about 1 point to almost 6.

The reason is the survey's sponsor. It was conducted for a Democratic-aligned client, and polls released by partisan sponsors are, on average, too favourable to their own side. Measured across about 20,000 historical polls, the average overstatement is close to four points for Democratic sponsors and about three for Republican ones. The model adjusts every sponsored poll by that amount and gives it less weight than an independent one.

So a sponsored poll showing its own candidate one point behind reads, after adjustment, as that candidate trailing by about five. The same adjustment applies with the parties reversed.

Kansas debates

Roger Marshall and Adam Hamilton met for a Kansas Senate debate, and The Hill published its takeaways. The forecast, which does not read debates, had Marshall ahead by about 2.5 points, a little closer than the previous week. With only four polls, Kansas remains one of the races where a few new surveys could move the forecast the most before election day.

The list of coin flips grows

By the end of the day, the Senate races in Alaska, Iowa and Texas were all within a point in the forecast, and Kansas within three. History suggests leads this small, two weeks out, are close to meaningless: in races since 1998, a candidate leading by under two points at this stage won about half the time. It is fair to say the Senate will be decided by the races nobody can call.