The pollsters were even more dismayed. On every side, they were ridiculed and abused, an inevitable development given the omniscience to which they had laid claim. In his first postelection column, Elmo Roper sounded completely deflated:

"I could not have been more wrong. The thing that bothers me most is that at this moment I don't know why I was wrong."

He promised to find out. A few newspapers canceled their subscriptions to the Gallup Poll. Many people recalled how the Literary Digest had gone out of business after the 1936 election, when its celebrated poll had predicted Alfred M. Landon's victory, only to have Franklin D. Roosevelt carry every state in the union but Maine and Vermont. The New York Times asked Wilfred J. Funk, the last editor of the Literary Digest, for a comment on the current plight of the pollsters. "I do not want to seem to be malicious," said Funk, "but I can't help but get a good chuckle out of this."

The postmortems began the day after the election and continued for years afterward. The figures told a paradoxical story: Truman had won by a decisive majority and yet the election had been close. In popular votes, the President had polled 24,179,345 to Dewey's 21,991,29 - a plurality of 2,188,054. In electoral votes, the score was Truman, 303; Dewey, 189; Thurmond, 39. The minority candidates had done poorly, with Wallace polling a mere 1,157,326 votes; Thurmond, 1,176,125; Norman Thomas, 139,572; and all the other fringe parties, including the Prohibition party, 150,167. The Wallace and Thurmond candidacies, however, had attracted enough Democratic votes to deny Truman an absolute majority; the final tally showed him with 49.6 percent of the popular vote to Dewey's 45.1 percent. The total vote had been low, with only 51.2 percent of the electorate going to the polls.

A plurality of over 2,000,000 votes was substantial, but Truman's lead in three states was so slender that the outcome might easily have been reversed in the Electoral College. Truman carried Ohio by only 7,107 votes, California by 17,865, Illinois by 33,612; his total plurality in the three states came to 58,584 votes. The arithmetic irony was clear: a swing to Dewey of less than 30,000 votes, appropriately distributed in the three states, would have given him an additional 78 electoral votes and the Presidency; the electoral vote would then have been 267 for Dewey to 225 for Truman and 39 for Thurmond. A switch of any two of these three states to Dewey (Ohio and California each had 25 electoral votes and Illinois 28) would have left Truman with the lead in the Electoral College but without the majority of 266 necessary to win, thereby throwing the election into the House of Representatives. All of which explained, of course, why Dewey waited until Wednesday morning before conceding.

There were other paradoxes. Truman had won despite the loss of four of the largest industrial states-New York (47 electoral votes), Pennsylvania (35)' Michigan (I 9)' New Jersey (I 6), all of which (with the exception of Michigan in 1940) had been carried by Roosevelt in the two preceding Presidential elections. On the other hand, Truman won three Midwest states-Iowa, Ohio, and Wisconsin-in which the farm vote was important and which Roosevelt had lost to Dewey in 1944.

Equally surprising was the impact of the Wallaceite and Dixiecrat defections. Truman was hurt, but not as badly as anticipated. His loss in the South was limited to the four states in which the Dixiecrats took over the Democratic party label-Alabama, Louisiana, Mississippi and South Carolina-whose electoral vote totaled 38. (Thurmond's 39th electoral vote came from a Democratic elector in Tennessee who could not abide Truman.) Wallace was responsible for Truman's defeat in three states-New York, Maryland and Michigan; in New York, for example, Wallace polled 509,559 votes-nearly half his national total-and Dewey won the state by 60,959. On the other hand, Wallace's vote in California - 190,381 - was far short of expectations and he was not on the ballot in Illinois. Had Wallace done just a trifle better in California and had he not been kept off the ballot in Illinois, Truman would have lost both states and the election of the President would have been decided by the House of Representatives. The Constitution provides that when the House elects a President, during January of the following year, each state delegation has but one vote; and the winning candidate must win a majority-which would have been 25 votes in 1948. One can assume that each state delegation would vote along party lines. On that assumption, Truman would have won, for the Democrats controlled 25 state delegations after the 1948 election; the Republicans had 20; three were evenly split.

As things turned out, Truman's winning coalition of states bore a remarkable resemblance to the design sketched by Clark Clifford in his memorandum of November 1947. The President took every state in the West but Oregon, won seven of the eleven states of the old Confederacy, the four border states of Kentucky, Oklahoma, Missouri, and West Virginia, and five states in the Midwest-Illinois, Iowa, Minnesota, Ohio, and Wisconsin; his only victories in the Northeast were in Massachusetts and Rhode Island.

Not only had the experts failed to anticipate the specific design of Truman's victory, but neither the pollsters nor the journalists foresaw that Truman would poll over 2,000,000 more votes than Dewey. The New York Star had the rare distinction of being able to boast that two of its writers, columnists Jennings Perry and Gerald Johnson, had predicted that the election would be close. In September, Robert Bendiner had also published an article in The Nation entitled "Don't Count Truman Out." was largely a review of a book by Louis Bean, How to Predict Elections, which described the rise and fall of the Democratic and Republican electoral "tides," over long spans of years, and argued that the Republican tide had begun to ebb by 1947. Bean foresaw the possibility of Truman's victory, but did not predict it.

How had almost all the experts been so wrong? As regards the journalists, the New York Times provided some instructive answers after querying the forty-eight correspondents who had provided the newspaper's state-by-state survey on the Sunday preceding the election. The correspondents reported that in taking their soundings, they had talked with local political leaders, checked with their colleagues, interviewed some rank-and-file voters and, of course, studied the polls, both local and national. An inordinate amount of reliance had been placed on the polls, far more than on "grass roots" reporting. One correspondent confessed:

"I was too gutless to put stock in my own personal hunch, based on nothing better than tours of the area, chats with businessmen, union men, miners, ranchers, farmers, political leaders."

He felt that he could not "pin down" his hunch, so he followed the trend indicated in the polls. Local political leaders also relied heavily on the polls in assessing the situation in their areas; the consequence was that when a reporter interviewed a politician, he often only picked up an echo of a poll which he had already read.

Newsmen traveling on the campaign trains were at an inevitable disadvantage in reporting the sentiments of the voters, for they lingered so briefly in a town that they had little opportunity to talk either to the voters or the local politicians. They were meticulous, as we have seen, in reporting crowd turnout and in describing the demeanor and attitude of the crowds, but they were usually unwilling to draw conclusions which hindsight showed to be obvious. Thus the great throngs which Truman attracted in the last two weeks of the campaign led almost no one to hazard the guess that he might just conceivably win; the tendency remained to contribute the crowds to idle curiosity or to Truman's "entertainment value" rather than to partisan zeal. The journalists were unwilling to credit the reality of what they saw and heard, so firmly had they accepted the assumption of Dewey's victory. In large part, this was a result of the polls; it was also a consequence of the political diagram of the campaign, with the Dixiecrat and Progressive party breakaways apparently making it impossible for Truman to win a majority in the Electoral College. If the electoral vote were the only error in prediction, it would have been understandable (the outcome in the Electoral College, after all, was narrowly decided). The more grievous error was the complete misreading of the popular vote.

The press had clearly been led astray by the myth of Dewey's invincibility. In a letter to his own paper, the New York Times' James Reston wrote: "There were certain factors in this election that were known (and discounted) by almost every political reporter. We knew about the tradition that a defeated candidate had never been nominated and elected after his defeat. We knew that the national income was running at a rate of $210 billions a year, that over 61,000,000 persons were employed at unprecedentedly high wages, and that the people had seldom if ever turned against tile administration in power at such a time. . . . in a way our failure was not unlike Mr. Dewey's: we overestimated the tangibles and underestimated the intangibles . . . just as he was too isolated with other politicians, so we were too isolated with other reporters; and we, too, were far too impressed by the tidy statistics of the polls."

Time's judgment on its colleagues was even harsher: "The press was morally guilty on several counts. It was guilty of pride: it had assumed that it knew all the important facts-without sufficiently checking them. It was guilty of laziness and wishful thinking: it had failed to do its own doorbell-ringing and bush-beating; it had delegated its journalist's job to the pollsters."

Everybody's mistakes, in the end, could be attributed to the polls. Eight days after the election, the Social Science Research Council set up a committee of academic experts to analyze what had gone wrong. The Gallup, Roper and Crossley organizations all cooperated in the investigation, making available both their files and their personnel. On December 27, 1948, the committee issued a detailed report and subsequently published the staff studies on which it was based.

The pollsters, the committee concluded, had made many errors. All three national polls had not been alive to the possibility of a decisive swing to Truman in the last two weeks of the campaign. Gallup and Crossley stopped polling too early. Gallup's final prediction, published the day before the election, was actually based on two national samples gathered during mid-October. Crossley's final forecast was derived from a combination of state surveys taken around mid-August, mid-September and mid-October. Roper's final estimate used data he collected in August, which provided the basis for his September 9 column in which he predicted Dewey's victory and announced he was no longer going to publish periodic surveys. (Roper, however, did take a poll in the final week of the campaign, which he did not publish. It showed a slight upswing for Truman, but Dewey was still far in the lead.)

The Gallup and Crossley organizations obviously assumed that the final stages of the campaign would have no significant impact on the preference of voters. As he indicated in his September column, Roper believed that the entire campaign was likely to be irrelevant, given Dewey's enormous lead (44.2 percent to Truman's 31.4 percent). The absurdity of these assumptions was proven when Gallup and Roper conducted post-election surveys. Respondents were asked if they had voted, who they had voted for, and when they had made their decisions. One out of every seven voters claimed to have made up his mind within the last fortnight of the campaign; three-quarters of these delayed decisions favored Truman. Other data corroborated these findings. A post-election poll conducted by the Survey Research Center of the University of Michigan reported that 14 percent of the Truman voters recalled making up their minds in the two weeks prior to the election; 3 percent decided on election day. The figures for Dewey voters were 3 and 2 percent, respectively. "Even if one makes allowance for errors in such reports," said the Social Science Research Council analysis, "one must conclude that failure to detect and measure changes of mind about voting during the closing days of the campaign account for a considerable part of the total error of the prediction."

Some of the voters who made up their minds toward the end of the campaign had previously been undecided; others switched candidates. The Michigan survey showed that, on balance, the last minute vote changes favored Truman. Thus, of the voters who said they were for Dewey in October, 14 percent switched to Truman and 13 percent did not vote. Of the Truman supporters in October, more failed to vote-25 percent-but only s percent switched to Dewey. Thus Truman benefited; as a report on the Michigan survey pointed out, "To lose a vote because of non-voting is only half as penalizing as to lose a vote through its going to one's opponent."

Gallup, Crossley, and Roper made equally grievous errors in handling the undecided vote. Voters who would express no choice, but said they planned to vote, were nearly twice as numerous in 1948 as in 1944; in the Gallup figures, they accounted for 15 percent of the total. The pollsters either disregarded the undecided, on the grounds that they were unlikely to vote, or divided them among the candidates in the same proportions as the decided voters. In past elections going back to 1936, such simple solutions had created no embarrassing distortions, because of Roosevelt's substantial lead, but in 1948 not only did the undecided bulk large, but over half of them voted-and they voted for Truman in the ratio of about two to one, according to the post-election surveys.

The pollsters also failed accurately to gauge intent to vote. "The error in predicting the actual vote from expressed intention to vote was undoubtedly an important, although not precisely measurable, part of the over-all error of the forecast," the SSRC report stated. "The prediction of human behavior from an expression of intent is, in the present state of knowledge, and particularly with the actual methods used, a hazardous venture. This is a central problem for research, which has been largely ignored in pre-election poll predictions."

The polls' difficulties were further compounded by errors in sampling and interviewing techniques, which resulted in too few interviews with people of grade school education. The Democratic vote thus tended to be underestimated. Finally, the polls added to their own problems by an exuberant desire to make flat predictions, not distracting the reader with qualifications or "technicalities." The reader was not taken behind the scenes and shown how the forecast was put together. Instead of merely trying to indicate the state of public opinion at some particular point of time and within stated margins of error, the polls insisted on picking the winner. The polls had made many of the same errors in the three previous Presidential elections, but they did not result in the wrong choice of the winner, given Roosevelt's large lead over his opponents. And, in the past, luck had enabled some of the pollsters' errors to cancel each other out. For example, underestimation of the northern Democratic vote had been compensated for by overrepresentation of the South, which had been solidly Democratic until 1948. After 1948 the poll-takers became a more modest and chastened lot, and their techniques improved considerably.