SHARE

A New York Times poll released yes­terday showed Pres­ident Obama ahead of Repub­lican rival Mitt Romney by 10 per­centage points in Ohio, 12 in Penn­syl­vania, and 9 in Florida. This and other polls have led Democrats to declare victory and Repub­licans to despair. But a closer look at the data sug­gests the results aren’t as accurate as they appear.

Most people assume that polls are accurate. It seems like any news report that includes the phrase “according to a recent poll” gets treated as the gospel truth.

Yet flawed polling is a problem, espe­cially in an election season. Polls are often unre­liable, mis­leading, and easily manip­u­lated for par­tisan gain.

So what makes polls methods of spin rather than accurate fore­shad­owings? Humans conduct the polls. Human answer the polls. They are subject to human error and mis­judgment.

Samples affect the legit­imacy of a poll. In this cycle, Democrats are often over­sampled. In Ohio, the New York Times poll skews toward Democrats: the sample was 35 percent Democrat, 26 percent Repub­lican, and 35 percent inde­pendent. But this dis­crepancy went under-reported.

Wording is also crucial. Respon­dents will say one thing in response to a question about “enti­tle­ments” and another in response to a question about “Social Security,” even if they’re asking essen­tially the same thing.

But polling data is essential in pres­i­dential elec­tions. Cam­paigns need data beyond their own internal polls. So the question emerges as to how to gain the most knowledge from a pres­i­dential poll in the middle of such a swamp. The only answer is: look closely. Every aspect of the poll matters. Perhaps the public should worry less about pre­dicting the results of elec­tions and more about the sub­stance of policy debates.

As for 2012, the race is far from over. The mis­leading polls may con­tinue to shape the media nar­rative in the coming weeks, but voters may yet sur­prise. It wouldn’t be the first time people have mis­un­der­es­ti­mated a Repub­lican pres­i­dential can­didate. Forget the polls.