Polls on Trial

Last week the poll takers of the nation were on trial and knew it. Four years ago, the practice of taking straw votes went out with a crash. The greatest of all straw polls, the Literary Digest's, took 2,376,523 straw votes by mail, and not only backed the wrong candidate but erred by 19% on the popular vote. It was a catastrophe to the Digest. It also left most of the pollsters who sprang up in the Digest's wake trembling in their boots for fear the Digest's fate might overtake them.

Some of them had reason to worry. This year Rogers Dunn, praised by Columnist Hugh Johnson, widely quoted by pro-Willkie newspapers when the Gallup Poll began to go against their man, surveyed 40 States by various formulas (not sampling of the populace), offered no estimate of the popular vote. To Wendell Willkie he gave 29 States with a total of 364 electoral votes, to Franklin Roosevelt only eleven States with 124 electoral votes.

In the Pathfinder, Publisher Emil Hurja, acclaimed as the statistician who in 1936 told Jim Farley that President Roosevelt would carry 46 States, this year forecast a Republican victory, gave Willkie a popular majority of nearly 1,000,000 votes. Harvard's Professor William Leonard Crum, writing in Barron's weekly, predicted that Willkie would "probably win with about 300 electors."

But in the same election in which the Digest poll went down to disaster a new kind of poll had its first public trial. Instead of inviting one & all to send in a postcard vote, it questioned a far smaller number of people about their opinions, carefully selecting those questioned in an effort to obtain a representative cross section of the population.

Best known of these polls was Dr. George Gallup's American Institute of Public Opinion. In 1936 the Gallup Poll successfully predicted Roosevelt's election. To be sure, it underestimated Roosevelt's strength by over 6%, but it was 13 percentage points closer than the Digest. Dr. Gallup's data last week showed a 52% majority for Roosevelt and 21 States in the President's bag. But he allowed himself a 4% margin of probable error, and day before election he wrote in the newspapers subscribing to his poll that he did not believe the data "justify a prediction of the outcome of this election."

The data were better than he thought. His over-all error was less than 3% on the popular vote. He hit the vote on the head in five States, missed by more than 4% in only three States. Because the election was close, he put eight States in the wrong column.

The American Opinion Forecasts poll, conducted by Edward J. Wall* on a scientific sampling basis, reported a 52% Roosevelt sentiment. This was the same as Gallup's prediction, but Mr. Wall allowed himself a statistical error of 2%, definitely predicted that Roosevelt would win the popular vote although Willkie might have a majority on the electoral college. But in 1940, as in 1936, the closest estimate of the popular vote was made by quiet, curly-haired Elmo Burns Roper, who has never made any great hullabaloo because he was one of the first to undertake political polls by the scientific sampling method and still makes no extravagant claims for his surveys.

He has been engaged since 1933 in making polls of popular preferences, but as a market researcher whose job is to make accurate surveys of public taste for private industry. In 1935 FORTUNE employed him to make polls of popular opinion for public consumption. For the FORTUNE Survey he first tackled political questions.

Roper still considers the quadrennial roundup of U. S. citizens at the polls chiefly as a check on the accuracy of his surveys--a means of calibrating for errors in his sampling method. He pays close attention to the phrasing of questions in his interviews, in order to get the real beliefs of inarticulate or evasive citizens.

Relying on scientific methods rather than the size of his sample, Roper, with a small staff of interviewers (81, compared to Gallup's 1,100 and Wall's 4,000), seldom samples the opinions of more than 5,000 citizens, chosen according to age, occupation, sex, economic condition, geographical distribution. (Gallup, in his State-by-State polls, may question as many as 60,000 people.) Roper's interviewers are carefully trained, rigidly supervised to prevent personal opinions from affecting their work. Always he checks back, querying interviewees by post card, to make sure that his interviewers have been accurate.

He has immense confidence in the accuracy of his surveys. What he fears more than a wrong report is "an election dominated by apathy, where the people don't take the trouble to vote, and where our poll might conceivably reflect true public opinion more accurately than the election itself." But he also insists on the limitations of his polls. He steadfastly refuses to make any forecast of electoral votes. The important and proper use of his political surveys is, he insists, not to predict elections but to obtain an over-all view of popular sentiment on public issues.

Sticking to that last, he reported in his FORTUNE Survey that popular sentiment for Roosevelt was 61.7% in 1936. The popular vote was 60.2%. Last week he reported that the popular sentiment for Roosevelt five days before election was 55.2%. The popular vote, as closely as it could be calculated from late returns this week was 54.6% for Roosevelt. He had reduced his over-all error to less than 1%.

* The Wall poll is not, as TIME mistakenly reported last month, owned or controlled by Emil Hurja.

This file is automatically generated by a robot program, so reader's discretion is required.