Math Mistakes in History: Predicting the President, Part II


truman.jpgAn earlier entry described the 1936 presidential election, in which George Gallup correctly predicted FDR’s victory by using more sophisticated sampling techniques than The Literary Digest. This was a blow to The Literary Digest, despite the fact that it had correctly predicted the outcomes of the previous five elections, and within two years The Literary Digest was absorbed by Time Magazine. But a decade later Gallup faced his own difficulties with presidential predictions.

The year was 1948. Democrat Harry S. Truman had become the 33rd president of the United States after Franklin D. Roosevelt’s death in 1945, but by that spring his approval rating was low. For the 1948 elections he was opposed not only by Republican Thomas Dewey, but by the States Rights’ Party candidate Strom Thurmond and Progressive Party candidate Henry A. Wallace. Polls conducted by three different agencies (Gallup, Roper, and Crossley) all predicted Dewey to win by wide margins, and the outcome seemed so certain that the agencies stopped polling more than two weeks before the election. This article by Irwin Ross quotes Elmo Roper as saying, “Thomas E. Dewey is almost as good as elected . . . That being so, I can think of nothing duller or more intellectually barren than acting like a sports announcer who feels he must pretend he is witnessing a neck and neck race.”

But of course Dewey didn’t win, and there’s that famous photo of Truman holding up the Chicago Tribune with “DEWEY DEFEATS TRUMAN” (although the Chicago Tribune wasn’t alone in presuming the result). So what went wrong with the polling? In this interview, George Gallup’s son Alec Gallup stated that the primary reason was that a significant number of people who had planned to vote for Wallace decided, shortly before the election, to switch to Truman, and the polling companies missed this last-minute change. He adds, though, that the sampling methods that had served so well in the 1936 election still had some flaws:

But certainly the probability-based kinds of sampling [e.g. random sampling], which came into being sort of in the early 1950s was definitely a superior method. And the difference is that before, the interviewer had some leeway into which households he approached. He had a given territory, but he could hit different households. Well, I don’t think there’s any question there’s a tendency to go to the households that don’t have dogs or don’t have mean-looking people sitting on the porch, or they’re going to go to places that are more accessible. And that does introduce bias. And so this new approach, probability sampling, was a definite improvement on the sampling procedures that we used.

And in this interview, George Gallup, Jr. stated another difficulty having to do with how polling companies accounted for undecided voters:

Another factor, but of less importance, was in the allocation of the undecided vote. Prior to 1948 we decided, or assumed, that the undecided would vote the way the decided did. So we split them accordingly. But, the fact of the matter is that a person who is truly undecided is truly undecided. And the safest way is to split it right down the middle. So when we started doing that we got much closer.

So while the polling techniques were not nearly as flawed as they had been decades earlier, they were still factors in this famous case.

For more information on the presidential election, visit the links above, The First Measured Century on, or The photo of President Truman above is in the public domain because it’s a work of the U.S. Army Signal Corps.


Tags: , , ,

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s

%d bloggers like this: