It may not have been the first profession, but one of the earliest vocations in human culture was soothsayer – the old crone or shaman who could predict the future.
Our attraction to those oracles continues to this day -- and in this technological age, we assume and expect ever-better precision and meaning in prognostications. Ironically, several drivers of the digital age have conspired to render current political polling less reliable than examining the entrails of a recently demised goat.
First, there’s the extinction of the newspaper. Remember them, in the halcyon days before craigslist decimated their profitable want-ads? Papers were primary generators and customers for polls, which require healthy budgets to reliably produce. In the internet age, newspapers have been mostly unsuccessful in confronting a problem indelicately called the hooker’s dilemma: how do you charge for it, when folks are giving it away for free? A major revenue source for polling organizations has gone the way of the typesetter.
Second, the disappearance of telephone landlines has rendered polls much more difficult to produce. A landline-only poll conducted this year would miss almost 2/3 of the adult population, and skew the sample toward the elderly and tragically unhip. In addition, cellphones are harder to locate and more resistant to robo-calling, a sourcing technique on which the authorities frown. And the decoupling of area codes from geography poses another concern for sub-national polls.
Further, as the population has been deluged with data-driven intrusions generally, we’ve also become much more resistant to participation. It almost seems quaint that folks used to donate their opinions for free. My own first question for surveyors is always where to send the invoice for my time. And if I do not recognize the calling number, or it begins with an 8, I am very likely not to pick-up at all. A remarkable 80% was the going response rate in the more agreeable 1970s; it can now take 20,000 calls to generate a thousand-person sample.
As technology has foreclosed traditional techniques, pollsters have tried to join ‘em by resorting to less-expensive internet polling. That comes with its own set of problems, mostly related to the ‘representativeness’ of any chosen sample. Almost 90% of Americans “use” the ‘net, but intensity is inversely related to both age and voting frequency. 97% of 18-29s routinely surf the web, but they constitute 13% of voters – whereas the 40% of the oldsters who do not go on-line represents 22% of all voters.
Thus, net-based polls are cheap, but not easy to get right. It has gotten so bad that the venerable Gallup organization, after compiling a dismal 2012 prediction record, actually opted-out of presidential polling this time around. They reasoned that they could not figure-out how to do it well, so they would not do it at all. If only other organizations, less troubled by garbage in-garbage out, had not leapt-in to fill the void!
But they have. Indeed, even as their reliability has fallen, polls have been put to more important uses – like winnowing down the field of GOP debaters. It’s reminiscent of the law that bans the use woefully inaccurate polygraphs for most purposes – except when it’s particularly important that they be right.
Much of the current overuse can be attributed to two phenomena: 24-hour cable news that needs filler material from anywhere they can find it, and the devolution of news divisions into profit centers within traditional broadcast networks. Just as cigarettes may be thought-of as delivery vehicles for nicotine, news reports are now mostly just carriers to generate pharmaceutical ad revenue$. Newscasts lead with weather forecasts and polls -- neither of which is ever accurate, only interesting. After all, there’s no prize for being right, and cheap, unreliable ‘outlier’ polls are differentially more likely to generate the “stunning results” required to hold viewers in the age of the clickbait headline.
So what are citizen viewers and readers to do?
The first advice is to make heavy use of the Mute button ‘til sometime between July 4th and Labor Day. A study of presidential polls from 1958-2008 confirmed what you might expect – they become more reliable the closer they are to Election Day. In fact, (I think -- my Statistics course memory is vague and fading), if I get the gist of this graph, it suggests that the reliability of polls doesn’t get much above 50% -- a coin-toss – until about 100 days out from the election. So, ‘tis better to worry about stunning weather revelations than whether Hillary is 2 points up-or-down vs. le petit Orange.
That said, Harry Enten of 538.com (Nate Silver’s organization, the gold standard in this tarnished industry) offers the following further tips, which I’ve paraphrased and editorialized a bit:
o -- During the primary election season, ignore national primary polls – they measure nothing. Only states matter.
o -- Ignore hypothetical match-ups during the same period – they also measure nothing and have a typical margin of error that swallows any distinctions that they may attempt.
o – only look at polls of likely voters vs. registered voters, conducted via live interviews. If the poll doesn’t reveal that information, expect the worst.
o – know the polling firm – their accuracy varies dramatically. 538.com offers a helpful guide to pollster ratings.
o – beware the clickbait “stunners” – they are probably wrong. Instead, look for averages of polls and trends.
o – The term “if the election were held tomorrow” is designed to heighten drama and force-reduce the Undecided fraction – but face it, many, many people are Undecided in May 2016, and some are only vaguely aware of what year it is.
o – consider source and motive of the medium that reports the poll. At best, they’re in the entertainment biz; at worst, they’re ‘fair-and-balanced.’