By James McCusker
Hollywood moviemakers were the first data miners.
They saw us not only as ticket-buyers but also as plot and character sources. As a result, there are very few things in our American life that haven’t shown up in the movies in one way or another. They even made a movie about public opinion polls.
The plot of “Magic Town” begins with James Stewart trying to hold together his failing small business, which conducts opinion polls. He is desperate, of course, in a way that only James Stewart could be. And if this sounds to you like the beginning of “It’s a Wonderful Life,” you wouldn’t be wrong. Hollywood has always trusted in formulas.
This time, though, there’s no river, no bridge, and no Clarence the apprentice angel. There is a miraculous deliverance, though, which comes with the discovery of a small town, Grandview, which turns out to be a perfect statistical image of the country.
This was the El Dorado of the opinion poll industry. There would be no need for expensive, time-consuming multiple samples taken in different geographic regions; no worries about demographic differences, sampling errors or other pesky statistical issues. If you wanted to know which laundry soap was going to be popular, or which candidate would be our next president, you just asked the people of Grandview. They would be right every time.
Of course, the magic of the town depended on two things:
•Keeping Grandview unchanged;
Keeping Grandview a secret — from his competition and, as importantly, from the town’s residents themselves. If any of these conditions changed, any polls taken would be biased and inaccurate.
All of this was complicated by the Jane Wyman character, the beautiful monkey wrench in the works. She is a reformer who wants to change things in Grandview.
Stewart, Wyman and Grandview eventually work things out — it is a formula movie, after all — and everyone learns to live with opinion polls. Even in 1947, when this movie was made, it was clear that polls were here to stay. Unfortunately, so was statistical bias.
The current presidential campaign has probably set a record for news stories, analysis and commentary based on polling data. And in a polarized political environment, poll results, no matter what they show, make a lot of people unhappy.
People from the unhappy group often claim that the campaign polls are biased, and they are probably right. The polls are biased, although not necessarily in the way they supposed. The bias isn’t always political.
Some of the bias in polls originates in the “who is asked and who answers” problem. Virtually all polling today is done by statistical sampling, and it isn’t as easy as we might think to get an unbiased sample. America is an incredibly diverse country and selecting a sample that reflects this is somewhere between difficult and impossible. That’s why political polls tend to be more accurate at the state and local levels, where the dimensions of diversity are more manageable.
A significant source of bias in polls comes from the self-selection process. The people who are willing to share their opinions with a pollster are a dwindling minority. Most telephone calls made by poll takers are unanswered or result in hang-ups. Moreover, the people who actually answer do not always give truthful responses.
A recurrent source of bias in polls is the unreliable statistical model that is used to interpret the results. When a sample survey is taken, for example, pollsters use these models to estimate the likelihood that the individuals involved will actually vote, usually with inaccurate results.
Opinions of non-voters resemble what economists call “ineffective demand” — like my views on the latest Ferrari, for example — and it is often important for pollsters to set them aside in interpreting their survey results. The pollsters’ models are frequently based on historical voter turnout, though, which has turned out to be an unreliable predictor.
The math behind polling statistics is based on random, unbiased samples, and the validity of using the same math on biased samples is questionable at best.
Are the polls of any use at all, then? And the answer is: yes, kind of.
On balance, they are probably more valuable to candidates than they are to the public because political polls tend to be more accurate on issues than on choices of candidates. Candidates are constantly reshaping their campaigns in response to issue polling data and this can be very effective.
For the rest of us, though, we wouldn’t be too far wrong if we ignore the claims of scientific accuracy and view national political polling data in the same way we might read a long-range weather forecast or a horoscope: interesting but not necessarily accurate.
James McCusker is a Bothell economist, educator and consultant. He also writes a monthly column for the Herald Business Journal.