Just had a little fun with monte carlo sims on the weekend before the election, because why not. Let's assume that RCP is right about Rs winning all the Leans R seats, and Ds winning all the Leans D seats. 1/
The @UpshotNYT has polled almost all of our tossups, and the good news is that most of them look like tossups (in part because we follow their polling and they're about the only ones doing the House, but whatevs). 2/
Let's assume further that there is no error to their polling other than standard error margin stuff. If we run our sims just off the basis of their uncorrected polls, and taking into account the races they didn't poll and the leans/likely R D races, 3/
Democrats would win 27 seats on average, and take the House about 99% of the time. This isn't surprising, as there are a *lot* of races where they have Democrats leading by a point or two. 4/
Let's assume a little bit of modeling error though, and say that their polls are underestimating Republican performance by a point. Suddenly the average D pickup is 24 seats, and Ds only win the House about 77 percent of the time.
What if it is understating Rs by two points? Democrats pick up 21 seats on average, and only win the house about 22 percent of the time.
Now this is admittedly a very crude calculation. But it does illustrate how just a little bit of bias can throw things off when there are so many close races.
