http://www.huffingtonpost.com/john-zogby/after-super-tuesday_b_85369.html

1) Last minute changes of heart: Not necessarily people changing their minds, just some McCain people who took the time to vote who might not have and some Romney people who might have voted but didn’t. Weather, economics, last-minute politicking can all cause that.

2) As Lem points out, sample error. Elections are non-random samples. Some sub-populations may turn out en masse and skew the results.

3) If you believed everyone who said they’d vote in the primaries, the polls would be wildly off, so they instead you go by who voted last time to determine who’s a likely voter. In unusually high turnout years like this one, this introduces error because some really do vote who haven’t recently.

4) Simple bad luck. Nearly every poll has a 95% confidence interval. That means that one poll in twenty is wrong by more than the stated margin of error. To increase the confidence interval, you can either widen the margin of error to the point that your poll doesn’t say anything, or you can poll more people (close to an order of magnitude more people are required to get from 95% to 99%). It’s usually prohibitively expensive and not always logistically possible to run tracking polls with sample sizes that high.

5) Time offset: The polls are usually taken several days before they’re released. This one, for example, took place on Saturday, Sunday and Monday. A lot can happen in the three days before Super Tuesday, and not alot of respondents have to shift before you get an effect on results.

Anyway, not to say that this wasn’t incompetence or maliciousness, just to remind you that even when everything goes well, statistics can be very tricky.

Pollsters specialize in devising questions and strategies best able to help their clients. Thus, a “Democratic Party” pollster may ask an entirely different set of questions than their “Republican” counterparts.

It is my guess that many voters changed their mind in the last few days before the election. Also, it could be very difficult to get an adequate sample due to cell phones and other new communication devices. And yes, it is even possible that Zogby is something of an idiot.

Pollster has become a dirty word for me, akin to the stock market analysts who are so very very often wrong about earnings estimates. These games are tinged with BS at best, corrupt at worst.