Polls of individuals on a phone can't take into account peer pressure in a room.
Having observed an Iowa Caucus and talked to lots of other observers of the Iowa Caucuses, I think it would have been impossible to predict the outcome unless you had interviewed the campaign's Field Directors and then assessed their strategies.
As has been noted plenty on the thread to Kos' post yesterday about the accuracy of the pollsters, the polls were all pretty accurate for how people first entered the room on Caucus night. In the Caucus I observed (and it was pretty similar all across the state), it was essentially 20% for each of the four major candidates. When it came time for recruiting undecideds, the Kerry and Edwards precinct captains were extremely well trained and very persuasive. The Dean precinct captains didn't get out of their seats to even try to persuade precinct captains.
I watched the Kerry and Edwards precinct captains (and sometimes 2 or 3 others from their group) recruit undecideds. They would offer to do the gardening for their undecided neighbor. They would smile and cajole and actually pull them by their arms. If an undecided joined their group, the entire group would cheer. They appeared to be the "popular" kids and made you want to be with them.
Why didn't the Dean precinct captains make an effort? They didn't really know they were supposed to. I just watched the Dean campaign's Iowa Caucus training video (at http://www.deanforamerica.com/deanvideo ) and it confirmed this. The video is entirely about "It's okay to go to caucus. It's kinda fun to get to know your neighbors. It's important. And, no one will hurt you." It seems to be targeting 20-somethings who might be afraid of a caucus, but it provides no strategic advice on what to do at the caucus. The Dean campaign spent their final days organizing large numbers of out-of-state volunteers, getting them maps and helping them get into the streets.
In contrast, the Edwards campaign spent the final four days of the campaign calling their strongest supporters in each precinct and encouraging them to be precinct captains, and training them (they had a 50-page book for each captain complete with lines to use while recruiting undecideds). Whenever a new person agreed to be a captain, the person on the phone in headquarters would ring a bell and everyone in the office cheered.
The Kerry and Edwards groups both arrived early to the Caucuses, got the best seats, put tables out that made their group look like the sign-in area, and the whole group seemed to know to look cheerful and fun. The Dean and Gephardt folks came later and didn't do anything cohesive. They sorta stared into space or chatted quietly among themselves. I think some undecideds went to the Kerry and Edwards groups as soon as they entered the room, just to join friends or neighbors they knew, because they looked like better places to sit and enjoy yourself.
Where would the undecideds have gone without peer pressure? Although it's hard to know, it's certainly POSSIBLE Dean would have been ahead (as in the SUSA poll) had it been a primary not a Caucus. But, how can you excuse the Dean Field Director for not knowing how Caucuses work, and the campaign's senior staff for not checking that the Field Director had a plan that made sense?