One particular pollster, New Hampshire-based American Research Group, has perhaps the most interesting explanation for how they got their state's Democratic result wrong. ARG put up a statement up on their site, saying the following:

While we missed the final number that Clinton would make in New Hampshire, our polling was one of only two daily polls that showed Clinton regaining support following her drop in New Hampshire the day after the Iowa Democratic caucus. Clinton was moving up in the final days and hours before the primary, and our polls and the Rasmussen polls were the only daily polls to catch Clinton's rebound.

...

Our polls missed the final Clinton number, but we did not miss the strong swing back among women reacting favorably to Clinton that started after the debate and continued with her comments in Portsmouth. We did not have a polling problem, we just ran out of time.

It is true, actually that ARG showed Hillary closing the gap. Their second-to-last poll had her down by 11 points — while the final one had her down by 9. Keep in mind that these were not composites of samples over three days, as in the case of John Zogby, but were instead conducted over 24-hour periods.

Congratulations, ARG.