I'm no expert on the art of polling but, I know a thing or two about them. Others that know a thing or three may care to comment on this post and expand or correct as they see fit. I would not be surprised if I learn something from the comments as dKos is a hotbed of activity for folks that seem to know a lot about polls.
What does a poll predict?
Nothing. A poll is a record of a historical measure.
What is MoE?
Margin of error. Usually shown as plus or minus some percent.
Does a poll that shows 20% for Candidate X with 5% MoE predict the candiate should get 15%-25% of the vote?
Nope. A poll predicts nothing.
What does it mean then?
It means that if you conducted a highly similar poll at the same time using the same methodology you would get a response of about 20% (plus or minus the MoE of 5%).
Isn't that the same as a prediction?
Not at all. The value of a poll as a tool in predicting a vote depends on many factors.
I still do not get it. Why isn't a poll a prediction?
For example, lets say you polled Republicans on who they would vote for if they were to vote in the Democratic primary? How accurately do you think that would predict the winner? Not at all, right?
But that's silly. What if I polled Democrats on the Democrat primary?
Lets say you did. You go to a Union Hall and poll Union members that identify themselves as Democrats. Certainly you see that this may not predict what Democrats as a whole would do, right?
But that's still silly. What if I polled average Democrats?
How are you going to know that they are Democrats? How are you going to know that your method for picking them is representative of those that will vote? How do you know you method for finding Democrats does not have a systemic bias in favor or against some sub-group of Democrats? In short, you don't.
OK, But if I try really hard to get a representative sample of Democrats that are going to vote then the poll should be a pretty good basis for a prediction, right?
People change their minds. Events happen that change people's minds. People lie to pollsters. Even the people you are polling do not know if they will vote. What if the weather turns ugly on that day?
So, polls are useless at predicting an election?
No, polls can be a very valuable tool. You just want to take them with a grain of salt and understand how they select their sample and how that compares to the people that will actually vote.
What about using polls to examine trends over a period of time?
I think that comparing the results of polls using the same methodology over a period can show much insight into the direction and degree of changes in opinion. The relative numbers (up 10% over two weeks, for example) can provide insights into shifts of opinion.
Are all polls equal?
No. Some polls are scripted to try to favor a particular outcome. Some are objective in the poll script, but may survey a biased sample.
What else can you tell us about polls?
I've already told you more than I know. :-) If this entry is interesting we may get some comments from folks that actually know what they are talking about. I dunno.