I hate being a skunk at this garden party, but I once taught statistics. I also agree with the sentiments and point of view of the majority of ATSrs.
But the following is true.
2391 is adequate to give a plus or minus 3 percent at 95 pc confidence to a percentage of voters that voted for say, the Newt.
IF the exit polls were properly conducted and sampled.
check the link for the ifs that have to be answered to get an accurate and representative result.
A little thought experiment for the doubters. Imagine you have a coin, with heads and tails.
You flip it 10, 100, 1000, 2391 times, and tabulate the proportion heads (or tails). in each case.
by the time you get to 2391, the proportion is going to be 50.00xxxx. this number has a precise distribution, i forgot the formula.
I believe the exit polling results were skewed, but the sample size is not the problem.
the real test is to locate the precincts that had exit polling, and compare the exit polls with the actual votes.