How the Media Skews Its Polling to Rewrite the 2016 Election Narrative

When NBC’s Chuck Todd recently tore apart a CNN poll that suggested Republican presidential nominee Donald Trump had moved into a 2-point lead over Hillary Clinton, it momentarily exposed the fatal flaw in many media polls.

Oversampling is a method of skewing a political poll to ensure a certain results, thereby advancing a particular narrative, rather than simply reporting on the current public sentiment. The method involves including more people from a particular group known to overwhelmingly support a particular opinion.

In Todd’s case, he was suggesting CNN had oversampled “white, non-college educated” voters to produce its result. It may have been meant to downplay concerns the network had too closely aligned itself with the Clinton camp following news that celebrity physician Dr. Drew Pinsky’s program had been canceled after he questioned the former secretary of state’s health and health care.

However, NBC and many other news outlets are just as guilty of oversampling in their own polls. As the U.K.’s Daily Mail reported back in June:

Hillary Clinton has opened up a 7-point lead over Donald Trump in an online poll that seems to reflect a ‘bounce’ for the former secretary of state after she wrapped up the Democratic nomination last week.

But the weekly tracking poll, from NBC News and the Surveymonkey company, included the opinions of 7 percent more self-identified Democrats than Republicans—the same margin as the poll’s topline result.

That raises questions about the 49-42 result.

… The most recent Gallup polling on the subject, published in January, found the Democrats had just a 3 percent advantage nationally. In that same survey, Republican-dominated states outnumbered those populated mostly by Democrats, by a 20-12 margin.

The raw information collected by Surveymonkey each week has generated a distribution of Democrats and Republicans that appears random, and it’s unclear how much it was tweaked to fall in line with Census and Labor Department numbers.

The polling data show nearly all the organization’s results this year have ‘oversampled’ Democrats. But one week’s results, published in January, included more Republicans.

Adding the sampling disparity to the poll’s margins of error, each of the last six week’s results can be seen as a dead heat if Democrats and Republicans were to come to the polls in roughly equal numbers this year.

But the oversampling could be backfiring. By creating the impression that Clinton has more widespread support than she actually has, these skewed polls could result in more Republicans and Independents rushing to the polls on Nov. 8 to prevent her from being elected.