With all the interest in this year's election there are a good number of people who are driven into a frenzy with all of these polls, trumpeting the latest poll results and assigning a good deal of meaning to them. There are other, more cynical, people who refuse to believe that anything meaningful could be gained from polling such a small number of people, especially when they are never called themselves. Over the next few days I'll try to explain how these polls work with an eye to answering the following questions:
1. What does the margin of error mean?
2. How is it calculated?
3. Are polls always correct?
4. How many people are really needed to accuately poll?
5. Why do polls by different organizations always seem to differ in similar ways?
Today I'll try to briefly answer questions one and three.
What does the margin of error mean?
The margin of error (or MOE) in the poll is the range that you can expect reality to differ from the results of the poll. For instance, if a poll between Bush and Kerry results in Kerry 48% and Bush 44% with a MOE of 3% then the real split in the entire population should be somewhere between 45% and 51% for Kerry and somewhere between 41% and 47% for Bush.
Are polls always correct?
No. There is a magic number built into every poll called the confidence interval. For political polls the confidence interval is 95%, which means that the poll will be within the margin of error 95% of the time. (For medical or engineering tests, where the results matter, the CI will often be 99% or higher. For politics 95% is good enough.) What this means, and nobody ever mentions this, is that one out of evey 20 polls should be way off base with reality differing from the poll by more than the margin of error.
Next: How is the margin of error computed?