If you're a political junkie, you probably read Nate Silver already, but be sure to look at his post in which he knocks down one piece of conventional wisdom: that incumbents who are under 50% in polls are in grave danger even if they lead their opponents. If you're not a political junkie, you may not have known about this one, but many have believed that most undecided voters generally break for the challenger in those races, so that an incumbent winning, say, 45-35 was in grave danger. Silver runs the numbers, and finds that it's just not true.
One caveat -- Silver compares final vote to polls one month out. That can answer the question of how undecided voters break at the end of a campaign, which has been the emphasis of previous similar studies (see this old Mark Blumenthal post Silver cites, and the links contained in it, if you want to know everything there is to know about studies of this effect). A somewhat different question, and one I'd be interested in knowing the answer to as well, is whether polling under 50% six months or a year out is a good indication of trouble for an incumbent. Of course, there are a lot fewer polls taken well in advance of the election, but I suspect that Silver (or Blumenthal) has the answer to this one in his database, and perhaps he'll take a look at that one, too.