Originally Posted by ColoradoDarin
Except when he wasn't. Like 2010.
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.