Dewey batte Truman

Poll Sampling: Why Method Matters

The Dewey-Truman case is one of the best-known lessons about poll sampling. On November 3, 1948, the Chicago Daily Tribune hit newsstands with the famous headline “Dewey Defeats Truman.” In fact, it was the incumbent president, Harry Truman, who won. As a result, the mistake became famous not only as a journalistic episode, but also as an example of what can happen when research institutes build their poll sampling poorly or close it too early.

In 1948, the leading American institutes—including Gallup, Roper, and Crossley—predicted a victory for the Republican Thomas Dewey. Clearly, these were not amateur operators. On the contrary, they were leading figures in the commercial opinion research of the time. For precisely this reason, their shared error showed that the quality of a survey does not depend solely on the number of people interviewed. Above all, it depends on the design of the poll sampling.

First of all, one weakness concerned the timing of the survey. Many institutes had closed their interviews before the vote, because they were convinced that the outcome was already decided. Truman, on the other hand, ran a very intense final campaign. In particular, he gained ground in the last few weeks. In this sense, some polls did not capture public opinion badly in absolute terms. Rather, they captured a moment that had already changed by the time the election took place.

Moreover, this point is still relevant today. For instance, a poll conducted too early can miss the final movement of opinion. In poll sampling, therefore, it matters not only whom you interview. In addition, it matters when you collect the data. Ultimately, timing is part of the quality of the survey.

Secondly, another weakness concerned quota sampling. Under this approach, the institute asks the interviewer to contact a certain number of people with specific characteristics, so as to reproduce the composition of the population. Indeed, it is still a widely used method: its results, however, depend on the rules by which the interviewer selects whom to interview. In 1948, for example, that choice largely depended on the judgment of the individual operator, and consequently this margin of discretion led to some distortions.

Thirdly, a further weakness concerned undecided voters. Institutes often treated them as a group to be distributed almost mechanically among the candidates. In reality, however, a significant portion of them shifted toward Truman only at the last minute. In other words, undecided voters were not merely a statistical residual. Instead, they were an informative component of the sample.

Consequently, when you oversimplify undecided voters, the risk is that uncertainty turns into a systematic error. Here too, the lesson is clear: poll sampling concerns not only the selection of respondents, but also the way you interpret answers and hesitations.

To sum up, the Dewey-Truman case reminds us that the value of a poll lies in its method. Every technique performs at its best when you choose and manage it with expertise, and a reliable poll requires a robust design, data collected at the right time, and a careful handling of uncertainty. In the end, the quality of poll sampling is not measured by the total number of interviews. Rather, it is measured by how you build the sample, when you survey it, and how you interpret the answers.