The Great Fiasco

From the grave of the Literary Digest, whose back was broken by its 1936 straw vote,* came a sepulchral horselaugh last week. "Nothing malicious, mind you," said ex-Digest Editor Wilfred J. Funk, now a Manhattan book publisher, "but I get a very good chuckle out of this."

But George Gallup, Elmo Roper, Archibald Crossley and all the other pollsters who had been dead wrong on the election could not see the joke. They had reason to wonder last week whether their great fiasco would not put them, like the Digest, out of business.

The Peoria Journal, quoting a telegram from Dr. Gallup ("This is the kind of a close election that happens once in a generation"), retorted: ". . . The Gallup poll, had it been properly evaluated, should have told us it was going to be such an election." It canceled its contract to run the Gallup poll; so did the Nashville Tennessean, the St. Louis Globe Democrat and others.

More was at stake than election polls, which are only a small part of the business of Gallup, Roper et al. The whole $25 million-a-year industry of polling, which employs 10,000 people and serves up "scientific" answers on buying habits, audience reactions, and all manner of likes & dislikes for Hollywood, businessmen, educators, magazines, etc., was under suspicion.

Straws in the Wind. Had there been no signs that the prophets were wrong? Looking back, pollsters found that there had been.**

In Denver, a statewide poll run by Edward Whittlesey, an ex-Gallup student, and William McPhee, an alumnus of the University of Denver National Opinion Research Center, found last June that Truman would win Colorado, as he did. But they got worried when their results disagreed with Gallup's, so they jiggered them for publication in the Denver Post-- to show a Dewey victory. Said McPhee: "Whittlesey and I are thinking of going out of business."

The Crossley poll (which predicted a Dewey victory of 49.9%) had discovered an upsurge for Truman in the campaign's closing days, but underestimated it. In a statewide poll just before election, the Chicago Sun-Times found a shift to Truman (but did not trust it enough to print it) which indicated a 50.05% victory in Illinois (the actual vote was 50.68%). Said Editor Richard Finnegan: "This has taught us a poll is no good unless it follows the voter right up to the booth."

Last Straw? At week's end, the pollsters themselves were still trying to figure out how they had come such croppers. Roper, confessing that he "could not have been more wrong," asked a group of social scientists to check over all his pre-election data for clues. Gallup started to recheck his pre-election polls; his field workers were re-interviewing the same people to find out how they had actually voted.

Explained Dr. Gallup: "My percentage of error, on the basis of the four-party vote, of course, was only 2.7%." That was one way of putting it. From another standpoint his "percentage of error" was much higher. What he did was to add up his errors on all four candidates and divide by four. Actually Gallup guessed 10% high on Dewey's share of the total vote and 10% low on Truman's. He predicted that Wallace would get 4%, and Wallace got 2.4%. In Gallup's book that was a difference of 1.6%; counted another way, he had overestimated by 40%.

Crossley also tried to bluster it through, insisting that "we weren't wrong." The polls were right when taken, he argued, only the voters had changed their minds later. If that was so, the pollsters had been tripped up by their own assumptions. But none was ready to admit that his methods, as well as his interpretations, might be unsound.

This year, Roper had assumed that the voters had made up their minds last August and would not be swayed by the campaigns. He had also been joined by Gallup and Crossley in a still bigger assumption--that the 15% of undecided voters could be safely eliminated from their calculations. They had assumed that the "undecided" votes, if cast at all, would be split about the same way the "decided" votes were. Now, Gallup suspected that the huge chunk of undecided votes had gone to Truman, 4 to 1.

But newspaper editors and their readers were not nearly so much interested in the reasons for the polls' failure as in the fact that the polls had failed. They had been fooled once by the "science" of polling--and they did not intend to be fooled again. In canceling the Roper poll, the Pittsburgh Post-Gazette summed up the opinion of newsmen: "We won't pay any attention any more to 'scientific' predictions and we don't think our readers will."

* It predicted 370 electoral votes for Alf Landon, who got eight.

** In July, Kansas City's Staley Milling Co. began polling customers in eight Midwestern farm states by giving them a choice of feed sacks labeled: "A Vote for the Republican [or Democratic] Candidate." But Staley stopped its "pullet poll" in September (after 20,000 farmers had voted), because it disagreed with the national polls. As it turned out, Staley's results--54% for the Democrats v. 46% for the G.O.P.--were right on the nose.

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