Scientific Polling? Scientific? Jude Wanniski November 2, 2000

Memo To: Political writers and commentators
From: Jude Wanniski
Re: Back to the drawing board

When I became a Washington political reporter and columnist for the Dow Jones National Observer in 1965, I soon realized I was the only fellow in the political press corps who had a background in science. I’d spent four years at Brooklyn Technical High School and two and a half years at Brooklyn College and UCLA, planning a career in science, specifically geophysics, with the intent of becoming a petroleum geologist. As poor a student as I was in that realm, I was forced to think in terms of the scientific method. The discipline is what led me to reject demand-side economics, because it does not work, and to understand the supply-side analytical framework, because it does work. It also has helped me understand why political polling does not work even though it is presented as being “scientific.”

A long time friend, Dr. Gordon Prather, is a nuclear physicist who has more than twenty years experience applying statistical methods to nuclear data analysis. He also has had years of experience statistically sampling electronic signals. In a number of Worldnetdaily columns, Prather has argued that the reason the projections of pollsters have been both contradictory and wildly fluctuating is that they have been misapplying statistical analysis and sampling techniques. And whereas a scientist soon realizes when he is misapplying sampling and analysis techniques, the pollster never does, especially when he is able to get away with claiming he was right at the time, but that the mood of the electorate has changed.

The idea of a “margin of error,” he argues, is ludicrous when you have only one data point. How can one pollster say his margin of error is 2.5% plus or minus and that Gore will win by six percentage points and another say his margin of error is 2.5% and Bush will win by six points? There is no doubt that one pollster used some formula to determine “margin of error” some years ago, and so now all pollsters use the same arithmetic. It is all a silly business and it is being exposed this year as never before. Of course, when the returns are in on November 8 next Wednesday, there will be one of the dozen or so who will be credited as being the most accurate. Prather would say the most accurate will be the one luckiest to have hit upon the right combination of people in the closing days. He assumes pollsters are constantly finding in their random samples enormous swings from one day to the next, and throwing their findings out, as being too silly to report. It is theoretically possible to run into 500 people who will tell you they plan to vote for Pat Buchanan. It is much safer to throw out those numbers and poll until you find numbers that are close to the consensus -- if you want to continue getting paid. Exit polls are more “scientific” in that the pollster knows the voting behavior of the people at that place and can relate their present voting behavior against the past.

Here is a marvelous account of the 1960 election by William Manchester in his narrative history of the United States, The Glory and the Dream, v.II. This is the night of Election Day:

An IBM-CBS computer enlivened the early evening by predicting, on the basis of data available at 7:15 p.m., that Nixon would win -- its incredible odds were 100-to-1 -- with 459 electoral votes to Kennedy’s 68. Then, as hard numbers poured in, the country appeared to be going Democratic in a landslide. Kennedy took Connecticut, always the first state with complete returns, by 90,000. He was winning New York City by a huge margin and carrying Philadelphia by 331,000, 68.1 percent of the vote. Cook County, under the watchful eye of Dick Daley, was giving the Democratic ticket a lead that seemed to place it beyond the reach of downstate Republican Illinois. At 10:30 Kennedy’s popular vote plurality was 1,500,000. He was then being projected the winner by 4,000,000 or 5,000,000. The IBM-CBS machine was giving him 311 electoral votes; NBC’s RCA-501 computer was putting it at 401. Viewers in the eastern United States were switching off their sets and going to bed, believing that it was all over. Jacqueline Kennedy whispered to her husband, “Oh, Bunny, you’re President now!” He replied quickly, “No... no... it’s too early yet.”

It was indeed. Kennedy’s high-water mark came shortly after midnight. His margin then exceeded 2,000,000, and the first returns from Los Angeles County indicated that he might carry California by 8,000,000. It was precisely at that point that the ticket began to run into trouble. Something unexpected was happening on the far side of the Appalachians. In the swing county of Lexington, Kentucky, for example, Kennedy was running behind Stevenson in 1952 and far behind Truman in 1948. Early Kansas returns put Nixon ahead or abreast of Eisenhower in 1956. Over the next two hours the picture cleared. It was not reassuring to the watchers in Hyannisport. The GOP ticket was sweeping: Kansas by 62.1 percent, South Dakota by 58.3, North Dakota by 55.4, Nebraska by 62.1, Wisconsin, conceded to Kennedy in all the polls, was going Republican by over 60,000 votes and the Democratic lead in California was disappearing as returns came in from the Los Angeles suburbs. Nationally, Kennedy’s popular vote margin dwindled to 1,700,000 to 1,600,000 to 1,100,000. Plainly it was going to be less than a million. It might vanish altogether... Official returns in December gave him 34,226,925 to Nixon’s 34,108,662 -- a margin of 112,881, less than two-thirds of one percent of the popular vote.

* * * * *

Here is Dr. Prather’s latest column on the subject. There are others which you can easily locate in his WND archive:

http://www.worldnetdaily.com/bluesky_prather/20001031_xcpra_sampling_v.shtml