This is a fascinating piece that came out in the WashPost two days ago where four professors predicted the electoral returns of 404 of 406 House races in 2012 using mentions on Twitter for their predictor. They threw out any races where there was no opponent (hence there are not 435) and used the normal socio economic factors for the independent variables (dependent is the margin of winning percentage). Here is the draft of the as yet unpublished paper. While it goes to say that traditional electoral forecasting is over, it does mention that you will still need it to gauge how races are going. Whether you buy this, with microtargeting it is certainly an interesting time to be teaching electoral returns (wish AP and state exams reflected these items).