In 1948, pollsters had N.Y. Gov. Thomas Dewey up 5 points over Harry S. Truman heading into Election Day. The combination of that and bad journalism led the Chicago Tribune to print 150,000 papers with a banner headline that got the results exactly wrong.
Or maybe, as this Atlantic piece by Rebecca J. Rosen wonders, are we headed for another 1936? That year, the nation’s leading pollster, the Literary Digest, predicted a landslide
victory for Alf Landon over incumbent President Franklin Roosevelt, which also was almost exactly wrong.
The reason people wonder is that the nation is undergoing a transition in terms of how we communicate, and how we are accessible. Polling relies on the ability to assemble a truly random sample. Until fairly recently, random telephone calls accomplished this nicely, as nearly everyone had a landline phone.
Today, a sizeable and growing segment of the population, mostly young, is accessible primarily through a cell phone or on social media. This Pew survey from 2007 found 49 percent of those in the 18-29 age group in this category back then, with the percentages thinning out quickly among older people.
This Scripps Howard News Service piece by Ben Boychuk argues that we’re being set up for a big polling fail. This, he says, is because “too many pollsters use outmoded turnout models and flawed samples.”
But a lot of other people are warning, “Not so fast.”
Pollsters today are using sophisticated methods and tools such as Census data to analyze information and account for discrepancies. This piece discusses some of this and notes pollsters are even counting negative and positive tweets on a candidate.
And the Atlantic piece I mentioned earlier notes that a close analysis of the 1936 Alf Landon fail shows the convention wisdom about it — that it relied too much on telephone respondents at a time when telephones were too new, rare and mostly owned by Republicans — is false. The culprit was a survey mailed out to millions, to which a disproportionate number of Republicans responded.
The Atlantic piece ends by saying, “Statistical and polling methods have improved dramatically since 1936, making the chance of an error of this magnitude — for reasons of technological penetration or any other — highly unlikely.”
Given the speculation this year about polling and shifting communication patterns, as well as the seeming witch's brew of methods to deal with it all, the only thing I can say with certainty is that time will tell, and that we will all know for sure in about a month.