http://www.electoral-vote.com/evp2010/House/house.htmlPrediction Methodology
To make predictions about how CDs will go without having polling data, we have to assemble some other data. In particular, for each CD we know the PVI, the 2008 election results, whether the incumbent is running again, and if so, how firmly established the representative is (i.e., how many elections has he or she won).
With this data available, here is the algorithm used in the table below. These predictions (given in the last column) will be used as our default value until a poll is published. For every race, each of the following steps is evaluated in order until one of them applies.
1. If a candidate has no major party opponent, the candidate wins.
2. An incumbent who got >= 58% of the vote in a contested 2008 election wins.
3. Any CD where the 2008 race was decided by less than 3% is a tossup.
4. Any incumbent first elected before 2000 wins.
5. Any Republican incumbents in a district that is R+3 or more wins.
6. Any Democratic incumbent in districts that is D+8 or more wins.
7. An open seat in an R+5 or more district goes to the Republican.
8. An open seat in a D+5 or more district goes to the Democrat.
9. All other races are Tossups.
If they don't have polling data from a relatively unbiased polling firm, they make assumptions based on the 2008 election. And the 2008 election was heavily Democratic. Nate Silver and the folks at FiveThirtyEight.com use a completely different formula, with much less of a bias from the 2008 election.
I follow both sites, but at the moment, I think that Nate, unfortunately, has the more accurate forecast.