April 12th, 2008
Â
Despite my having played the role of Mr. Smartypants on the subject of polling  in the post just below this one, I must confess that I don’t really understand the science of public-opinion sampling.
All I know is that reliable polling has something to do with the laws of probability. That’s how a well-conducted survey of just hundreds of Americans can tell you what Americans in general feel about a given subject at a given point in time.
Which brings me to the fun part of what I have to say here. There’s a certain exercise in probability that can win you bar bets in most cases if you can find any takers.
Here’s how you do it:
The first thing is to note for your mark (or sucker) that there are 366 possible birthdays, one for each calendar date and an extra one for leap years. Each person on Earth is going to have a birthday on one of those 366 dates.
Now, tell your mark that he or she can choose any 57 people at random, and you’re willing to bet any amount of money that at least two of those people share the same birthday.
The names can be chosen at random from the rolls of Congress or the Illinois General Assembly or NFL rosters or whatever. The only requirement is that the names be chosen randomly.
In 99 cases out of 100, you’re going to win the bet.
How can this be? With 366 possible birthdays, how can there be almost a lead-pipe cinch that there’ll be a duplicate birthday among just 57 randomly-selected people?
I don’t know how. But that’s the way it works. Moreover, there’s a 50-50 chance of a duplicate birthday among just 23 randomly-selected people. And there’s a 90-percent chance among just 41 people.
Try it at your office, if your company has at least 57 people. Of course, if you win any bets on this thing, I’ll expect my customary 10 percent of your ill-gotten gain.
April 12th, 2008
Â
A progressive blog I regularly visit had an item the other day about how a couple of new polls “continue the trend of predicting a narrow Clinton victory in Pennsylvania.”
The polls at issue did no such thing. No poll ever “predicts” anything. A poll, if done correctly, is a snapshot of public opinion at a given point in time.Â
Even a poll conducted just a few days before an election doesn’t predict anything. It can be a strong indicator of which candidate is likely to win, but it can’t measure the impact of voters who make their final decisions in the last hours or minutes before the election. Consider the late shift toward Hillary Clinton in the New Hampshire primary, which none of the major pollsters caught.
In the case of the upcoming Pennsylvania primary, the election is still 10 days away. The polls that my blogger friend saw as predictions probably were conducted several days ago. Lots can happen between now and election day.
The news media frequently misread poll results. Reporters and editors generally are ignorant of the most basic scientific principles of public-opinion polling. The public knows even less of such matters.
Take, for example, the margin of error in a given poll. About six months ago, Talking Points Memo, one of my favorite blogs, had this to report: “New ABC News/Washington Post poll shows Obama with lead in Iowa.”
But the poll did not actually show Barack Obama with a lead. It put Obama at 30 percent and Hillary Clinton at 26 percent. That gap of 4 percentage points was within the poll’s margin of error. Hence, the poll showed a statistical dead heat.
As for polls that are conducted online or by any other means in which participants are self-selected — or polls that admit they are “unscientific,” as if that disclaimer justifies their publication as “entertainment” — the results are utterly worthless.Â
I’ll never understand why news media that otherwise brag about their dedication to accuracy permit themselves to pass along results of unscientific polls. Gannett used to do that with an “Annual Teen Survey” in its “USA Weekend” magazine. (I haven’t seen that mag in years, so I don’t know if it’s still publishing phony polls.)
