Poll Position
There are quite a few polls to choose from for the Presidential election. Some show a tight race, others a blowout. I have no idea which is correct--it all depends on proper sampling, wording of questions, etc.
But there's a more basic problem. I've heard friends express trouble with the idea that a small sampling of people can predict how millions will act. I used to have this problem until a friend, Jim--a computer major--taught me to look at it a different way. And now, dear reader, I will pass on his wisdom. All math experts can stop reading here.
It doesn't matter how large the voter group is--let's say it's a million. If you pick a smaller number--say, a thousand--there will be only so many potential combinations of that thousand within the million. That number of combinations may be mindbogglingly huge, but it is finite.
Now imagine if you counted every single combination. You might find, say, that Candidate A gets 54% and Candidate B gets 46% of the vote more often than any other result. Then you might also note that you get within 3% of the 54/46 result 95% of the time.
Voila! It's no longer a matter of "prediction." As long as you are randomly sampling your thousand voters, you can be confident that 95% percent of the time, the results will be close to 54/46. The more votes you sample, the more confident you can be, though it soon becomes a case of diminishing returns. (If you poll everyone, you can be 100% confident with 0% variability--this technique was used in the movie The Rise And Rise Of Michael Rimmer). So pollsters poll enough people to be confident without bankrupting themselves.
I hope that was clear. Understanding something isn't exactly the same as explaining it.
posted by LAGuy @ 1:28 AM
4 Comments:
A computer major? From college? Told you personally? And to think the MSM says you can't trust what you read on these blogs.
Though your explanation was clear, then why the large difference in the results of the 2 polls that you linked to at the top? I'm sure USA Today and the ARG both will contend they have 95% accuracy. Also, if math experts were not supposed to read your explaination, why is that?
There are a few reasons that explain many of the differences in the polls. First, note that most are within the margin of error. Second, note even if the confidence level is at 95%, that still means one out of twenty polls is gonna be outside the margin. Most important, each poll uses questions with different wording and different weighted samples of the population. (No one is sure how many of each group will end up voting, so there is some educated guesswork going on.)
I asked math experts to stop reading because this stuff is old hat to them. All they'll do is point out my mistakes.
LAGuy.
Here's what I want to know: Why does the media report the margin of error? I fully understand your explanation (I even know what validity and reliability mean to pollsters, not to mention push polls and FRUGs and SUGs). What bothers me is that margin of error is so technical a concept that it's misleading. A computer will crank it out easily every time, and it will be right.
Of course, whether the poll tells you what you think it tells you all depends on your having asked the right questions and having run a fair poll, which I suspect is far more difficult than pollsters will admit, even assuming they want to run fair polls. In fact, I take it back; I've interviewed several of them, and I KNOW they won't admit how hard it is. (Understandable, since they don't want to be quoted saying their business consists of pulling fantasy numbers out their collective ass.)
But given all these difficulties and given the paucity of mathematical understanding at large, all margin of error can do is mislead most readers into thinking that the poll is, say, 95 percent "accurate," whatever that means.
Post a Comment
<< Home