by Barry A. Liebling
The outcome of the presidential contest of 2016 was a stunning surprise to both savvy and casual observers. Most election experts examined the voter surveys and concluded that Hillary Clinton was nearly certain to beat Donald Trump. When Trump – against the expectations of some of his most enthusiastic supporters – captured the presidency the prestige of survey companies fell dramatically. How could they have been mistaken? And more important, what can be done to improve the accuracy of future polls?
It is not a matter of lack of effort. Enormous amounts of money was spent on both partisan and independent pollsters – organizations that did their best to estimate what voters would do. With all their resources and the latest technology these companies saw that an accurate reading of the electorate is elusive.
In fact, there is inherent uncertainty in any large-scale survey, and surveys that focus on close presidential elections have built-in reliability problems. There are compelling reasons why the power of election polling is not likely to get better.
Consider first the need for accuracy. The more robust differences are among individual preferences, the less important it is to be precise. Conversely, if people are close to evenly-split in their attitudes towards issues – especially presidential candidates – the more vital it is to get the numbers right. Take a mundane fictional example. Suppose we asked a large sample of people whether they would like to eat a dish of ice cream or a dish of raw chopped onions. The magnitude of the ice cream victory would be tremendous. If you had to make a decision about what food to serve, you would not be very concerned about exactly what proportion of people choose ice cream (Is it 80% or 90%? Who cares?). Circling back to political elections, you can be sure that pollsters did not spend a lot of time and money surveying the citizens of Cambridge Massachusetts – where Hillary Clinton obtained more than 70% of the vote. Better to focus survey efforts on districts where the outcome is less certain.
So how do you increase the accuracy of polling in locations where the electorate is closely matched? It is not easy. Two difficult problems are getting the sampling right and encouraging survey participants to give candid answers.
Start with sampling. The classic, textbook way to select participants is to obtain a probability sample. This means the surveyor uses a method that (nearly) assures that each type of person of interest has an equal probability of being interviewed. In the old days researchers would use maps of residential neighborhoods to randomly select houses to target. The interviewer would knock on doors and request the cooperation of the residents. Of course, sometimes people would not be home, but instead of going on to another house the interviewer would deliberately do a “call back” – return to the house and try again another day. A research design that attempts to obtain a probability sample might have three or four call backs. Of course this is expensive and time consuming. And even if the method is followed strictly, some people will never be home or will refuse to be interviewed.
Today house-to-house interviewing is unusual. Instead, surveys are typically conducted by telephone (both conventional land lines as well as mobile phones – often with multiple call backs) and via the internet. But while technology has advanced tremendously (that is the good part), the difficulty of obtaining a projectable sample remains. No matter how you attempt to contact people some will never be available and some will refuse to speak with you. Even worse, it is more difficult to audit the veracity of electronic polls than the old-fashioned house-to-house method. How are people who were never interviewed different from those that are included in a survey? That’s a tough question to answer. You may not know until after the election.
OK, suppose you solve, or at least minimize, the sampling problem. There is still the difficulty of getting candid, genuine answers from the survey participants. And here we run into another obstacle to obtaining reliable results. Some questions are emotionally neutral, and people will not hesitate to tell you what they really think. There is no stigma, embarrassment, or reason to conceal your strong preference for wanting to eat a dish of ice cream instead of a dish of raw chopped onions. No authority figure is likely to make you feel uncomfortable if you reveal that you have a yearning for a frozen treat.
But political contests are not emotionally neutral. The mainstream, leftist culture which dominates public schools, universities, the entertainment industry, most print publications, and most television has been a partisan supporter of the Democratic candidate Hillary Clinton. The message from the “ruling class” is not simply to vote for the former Secretary of State. Failure to comply with the “progressive left” agenda is a serious transgression. Potential voters for Donald Trump were threatened repeatedly that voting “incorrectly” will tar them as racists, sexists, xenophobes, homophobes, and deplorables. Surely these terrible accusations discouraged many voters from expressing their true opinion in surveys. For some people it is safer to say they are undecided or that they may not vote at all than to risk the condemnation of the left’s cultural elite.
Now look to the future of political polling. Getting participants that are close to a probability sample is consequential to the extent that the candidates are closely matched. If one is way ahead of the other, fastidious sampling is not required. But the most interesting contests are those where the outcome is uncertain. Also consider how social pressure influences the results of surveys. The mainstream left-leaning culture will remain in power at least in the near term, and their unrelenting haranguing will always inspire dissenting people to keep their authentic attitudes to themselves. Unless the world changes dramatically, election polls won’t improve.
*** See other entries at AlertMindPublishing.com in “Monthly Columns.” ***