Chris, sorry it took me so long to respond to this question. I’ve been away and * **gasp** * not checking Fluther. Anyway, here are the basics.

Public opinion polling is, essentially, like drawing randomly from a box full of marbles. Imagine a big box full of marbles of different colors, all mixed up. If someone asked you, “what percentage of marbles in the box are red?” you could tip the whole box over and count all the marbles and see how many are red. But, if you don’t have time for that, or if the box is just too huge to do that, instead you could pull out a random sample of marbles from the box. So let’s say you reach in and pull out 10 marbles from the box and of those 10, 5 are red. You could answer the question by saying, “my best guess is that about 50% of the marbles are red.”

How accurate is your estimate? Well, in this case, not super accurate because you don’t have a very big sample (in fact, statistically, we would say that you are 95% confident that the actual percentage of red marbles in the box is between about 20% and 80%, not a particularly useful margin). But, as your sample size increases, your potential for error decreases. If you reached in again and drew out 100 marbles and 50 of them were red, then you could say, “I’m 95% confident that the actual percentage of reds in the box is between 40% and 60%.”

Public opinion polling is basically just a more sophisticated version of this. Instead of a box, we have a pool of potential voters, each of whom has a “color” (i.e. their presidential preference). Most polls reach in and draw out between 400 and 1200 voters from this pool and as a result they make an estimate from this sample. You might notice that polls with larger sample sizes have smaller margins-of-error. This is because of the principle noted above: larger sample size means less potential for error.

As elections draw nearer, many polling outfits go from drawing their samples from Registered Voters to “Likely” voters. This just means that pollsters are drawing their marbles out of a different box. “Likely” voters is somewhat subjective, so this absolutely introduces more potential for error which is not captured in the margin-of-error, though probably not very much since these polling firms test and retest their models for predicting likely voters.

Finally, there is the issue of cell-phone use. Sam is correct that a growing number of people don’t have land-lines. This could be a problem because most pollsters use telephone numbers as their “box of marbles.” If cell-phone numbers aren’t listed, then they can’t be put in the box of marbles. This only becomes a problem if people without landlines are largely different from those with landlines and if this number becomes large enough to affect the sample. I am personally of the opinion that we have not reached that point yet.