• Choose a category
• All
• Classic-Fiction

By Root 865 0

stopped conducting polls (which are rather expensive, after all). But public opinion changes over time: in 1948, the support for Truman got stronger in the weeks before the election as people abandoned third parties in favor of candidates who had a chance of winning—and those voters tended to favor the incumbent. Thus sampling too early introduced a bias against Truman. Another subtle error arose from a faulty assumption about undecided voters. Gallup thought that the voters who hadn’t thrown their lot in with Truman or Dewey would vote in the same manner that decided voters had—that is, since the majority of decided voters preferred Dewey, the majority of undecided voters would feel the same way. This wasn’t true at all: undecided voters behave very differently from those who have expressed a strong preference. In this case, more of them finally threw their lot in with Truman than expected. The assumption that undecideds behave like other voters subtly biases a poll in favor of the person in the lead, making results look more significant than they are. Again, a systematic error killed a poll, causing it to yield the wrong result. Gallup too fell into the trap of (ignorant) disestimation. He thought his measuring instrument was much more precise than it actually was, leading him to underestimate the uncertainties that surrounded the numbers he produced.

Surprisingly, the 1948 result didn’t cause any permanent damage to the public’s confidence in polling. Gallup himself wasn’t deeply shaken by the wrong results—he was convinced that he could find the problems and root them out, gradually perfecting the art of polling. He kept faith in his methods, even though systematic errors kept cropping up to destroy his polls. For example, his failure to control for respondents’ religion might have been responsible for the wrong prediction that Kennedy would win the 1960 election by four points. (In fact, it was a squeaker where a mere 0.25 percent of the popular vote separated the candidates.)

Systematic errors are bewildering in their variety, complexity, and subtlety. Some never get fully explained. In the early 1990s, pollsters in the United Kingdom never were able to figure out why polls consistently underestimated the votes for the Conservative Party. Without coming up with a real explanation, they dubbed the error the “Shy Tory” effect. It disappeared after a few years, its departure as mysterious as its arrival. Some systematic errors are downright creepy. For example, a careful study of certain kinds of polls demonstrates, beyond a shadow of a doubt, that people regularly lie to pollsters, making their data all but meaningless. When this happens, the polls stop reflecting reality at all. They become Potemkin numbers.

Every few years, someone tries to figure out our society’s sexual habits with a poll or survey. Some of these surveys in particular are extremely sophisticated and professional—they use expert statisticians and are backed by government money. But every single one returns utter garbage. It’s because we humans are liars—we simply can’t help it.

In 2007, to much fanfare, the Centers for Disease Control, the preeminent organization for generating health statistics in the United States, released a report about Americans’ sex lives. The element that got the most attention in the press was, naturally, the most salacious: how many sexual partners the typical American has had. “Average man sleeps with 7 women,” blared the Associated Press. That is, the typical American man has sex with seven women in his lifetime. That statistic in itself was not so surprising. But it becomes shocking when you combine it with another statistic from the same study: the typical woman sleeps with four men in her lifetime. This is impossible.40

Every time a man has sex with a woman, a woman has sex with a man. Each act of heterosexual intercourse increases the average male’s number of sexual partners; however, it must increase the average female’s number of sexual partners by the same amount.41 The male average must equal the female