Don't know where you are getting that number from. Silver predicted a 55 seat gain for Republicans. Considering how hard it is to handicap House races, that ain't bad, and I challenge you to find anyone more accurate:
U.S. House of Representatives
The model for projecting the outcome of the House of Representatives was more complicated than those for the Senate and governorships. For one thing, House races are more subject to the force of national trends and events than are the other two. One way to account for this was to take into account trends in the "generic Congressional ballot." Use of such a macrolevel indicator, as well as macroeconomic indicators, is a common approach taken by political scientists to project House elections.
Furthermore, there was much less available public polling for individual House districts than there is for Senate or gubernatorial races. By the end of the 2010 election season, public polls were available for only about 25% of the districts. This is one reason why some analysts rely principally on making global or macro-level projections of the number of seats to be won by each party rather than trying to forecast the outcome in every individual district. Silver's FiveThirtyEight model, however, while weighting the generic partisan division as one factor, focused on developing estimates for each district. For this purpose he used information on past voting in the district (the Cook PVI), the quality of the candidates (in particular whether one was an incumbent), fundraising by each candidate, "expert ratings" of the races, public polls of the given race (if they were available), and, in the absence of public polls a cautious use of private polls (i.e., polls conducted by or for partisan organizations or a candidate's own campaign organization).
In response to some concerns that he was hedging his projection, Silver contended that in his model the uncertainty of the outcome was a feature, not a flaw. In comparison with previous Congressional elections, a far larger number of seats were being contested or were "in play" in 2010. While his model, which relied on simulating the election outcomes 100,000 times generated a projected "most likely" net gain of 53 seats by the Republicans (two days before the election), he emphasized that the 95% confidence interval was ± 29–30: "Tonight, our forecast shows Republicans gaining 53 seats – the same as in recent days, and exactly the same answer you get if you plug the generic ballot average into the simple formula. Our model also thinks the spread of potential outcomes is exceptionally wide: its 95 percent confidence interval runs from a 23-seat Republican gain to an 81-seat one".
On election eve, he reported his final forecast as follows:
In final vote tallys as of December 10, 2010, the Republicans had a net gain of 63 seats in the House, 8 more than the total predicted on election eve though still within the reported confidence interval.