Slate just published an article explaining some theories of what happened in NH. The one below is the one I've been talking about, basically the volatility of undecideds waiting till the last moment. (17% in NH)

Frome Slate: http://www.slate.com/id/2181118

Feiler/Skurnik Effect: What's stunning is the ferocity and speed with which Hillary's fortunes turned around in those final hours. Kf has a theory to explain that! Actually, two theories. The familiar Feiler Faster Thesis holds that voters are comfortable processing information at the vastly increased speed it can come at them. Jerry Skurnik's "Two Electorate" theory holds that voters who don't follow politics are much less informed than they used to be, which causes polls to shift rapidly when they do inform themselves. Put these two together and you've got a vast uninformed pool of voters that only begins to make up its mind until the very last minute--after the last poll is taken, maybe--and then reaches its decision by furiously ingesting information at a Feileresque pace. In fact, the percent of voters who made up their minds at the very end in N.H. was unusually large. (Add convincing statistic here!)

Two implications of the Feiler/Skurnik combo: a) Momentum from the previous primary doesn't last. When the early primary dates were set, the CW held that the Iowa loser would never be able to stop the Iowa "wave" effect in the five days between the two primaries. It was too short a time. In fact, it wasn't short enough. A three day separation and maybe Obama would have won. As it was, by the time the uninformed voters tuned in on Sunday and Monday, Iowa was ancient history.*** b) Instead, these voters saw clips of Hillary having her emotional tearing up moment. In other words, the Feiler/Skurnik Effect magnifies the significance of any events that occur in the final day or two of the campaign. After yesterday's election, expect more of these events.

Which basically has it, though it misses a few thing.

1) The kind of information processed at the last minute will tend to be superficial and largely dependent on MSM spin. A voter who remains undecided till the last few hours, isn't likely to read all their voting history, and all the debate on interpretations of that history, in the 11th hour. That underscores the unimportance of any "serious" campaign to get a message out well before the election, and to have the time to make persuasive arguments.

2) Momentum does still matter. Especially longer term averages, which predicted NH outcomes pretty accurately. Also, the perceptions of being a winner/loser are still very important, and the "last moment" voter will, by default, weight momentum and perceptions heavily.

3) As we saw in NH, the last minute voter is also very volatile, and can be effected by theatrics and last minute media sensations. The question is then, how often can that be accomplished, and is something of a gamble to try and create a last minute sensation? Hillary can only cry so many times. Even Arnold's car crushing routine got old.

So, there's definitely a battle to win the MSM, and win over perceptions. Bill for example has staked out the notion the media has been unfair to Hillary, and did so right after Iowa, which bore fruit with some pretty kind coverage to Hillary in the last moments.

Also it's still very important to establish a trend of momentum. Voters will still factor that heavilly, including last minute undecideds.

But, there is always a possibility for last minute swings as well in states with high numbers of undecideds, due to the volatility. So again, managing those last perceptions, and ensuring some media balance is vital to candidates.