Originally Posted by enjolras
He got every senate race correct except for three (the coin flips in Colorado and Alaska, along with the surprise in Arizona). The one in Arizona was most definitely an issue with the polling data.
He struggled a bit with the house races, but that makes sense. Polling data for individual district races tends to be sparse. It's hard to have a good statistical model without data.
Essentially his models have been shown to work well when there is a lot of data available. For a presidential election that is most definitely true.
And the data might be wrong. Polls with dems regularly +5 or more skews the polls and supposes voter enthusiasm will side with the democrats, like in 2008. Alot of assumptions to make.