The Unlikely Truth About Polls
At this point in every campaign season, failing politicians are saying something like: "I don't believe in the polls. The only poll that counts is on Election Day."
This year you have to wonder if that might be the right attitude. Should we believe these polls?
The worrisome thing is how diverse they are. On Thursday, for example, the IBD/TIPP national poll had Obama's lead shrinking to a mere 1.1%, while the CBS/NYT poll it growing to 13%. Something like that happens almost every day. How is that even possible? If the polls are all measuring the same underlying reality, shouldn't every poll be within the margin-of-error of every other poll?
It turns out that's not what margin-of-error means. A poll's MOE is really only the margin of one kind of error, known as sampling error. The other kinds of error are nearly unmeasurable (at least until there's a real election), but that doesn't mean they're zero.
Sampling error amounts to this: If the pollster does everything right -- chooses a truly random sample of the electorate, questions them in a way that doesn't influence their answers, and so on -- and if the respondents don't lie, then the results still might be off just because of the luck of the draw. You might happen to pick a few too many McCain or Obama supporters in one day's sample, just as you might flip ten coins and get four or six heads rather than five.
But sampling error is just the beginning of what might go wrong with a poll.
No comments:
Post a Comment