The question is motivated by reading about Gallup's success in predicting the 1936 US election with a much smaller but more representative sample than Literary Digest.

Literary Digest used phone directory and car registration to construct their (biased) sample. What did Gallup use back then? I checked Gallup's methodology today, and they actually rely on landline and cellphone directory.

This question is not necessarily about the historical details of this case, but about sampling method in times and settings where getting the sampling frame is difficult.

  • $\begingroup$ Was'nt this Gallups idea, at least, the use of sampling for opinion surveys. Around 1900 statistical agencies was focused on enumeration, and sampling "the representative method" was considered controversial? $\endgroup$ Apr 28 '15 at 13:36
  • 2
    $\begingroup$ Googling for "survey sampling history" reveals multiple sources: cbs.nl/NR/rdonlyres/BD480FBC-24CF-42FA-9A0D-BBECD4F53090/0/… or projecteuclid.org/download/pdf_1/euclid.ss/1177013352 $\endgroup$
    – Tim
    Apr 28 '15 at 13:43
  • $\begingroup$ I seem to remember a speaker telling a story about how Gallup would literally stand around at a grocery store asking people questions. I might be confusing two stories though. $\endgroup$ Apr 29 '15 at 4:59

In his book, Théorie des Sondages (Dunod, 2001, written in French), Yves Tillé tells the history of survey sampling. According to him, A.N. Kiaer introduced in 1895 at an IIS (International Institute for Statistics) congress the idea of conducting a survey to measure certain variables on the population, instead of doing a census. His talk was very controversial, many well-known statisticians deemed it nonsense.

It's not until 1925 that the IIS officially recognized the idea of survey sampling as worthy of interest. And, in 1936, Gallup applied it for the very first time to predict the results of an election.

Before the 1930s, I believe there were a lot of "surveys" conducted among newspaper readers to predict the result of elections, but as I understand, people were not aware that their samples were biased.

EDIT : I think they did not construct samples per se : they would just put an ad somewhere in the paper saying that if you wanted to participate in their prediction, you just had to send the name of the candidate you'd vote for to their postal address. I don't think they took it too seriously though (this paper says there's no specific scientific literature on how to predict elections until 1948).

For example, in Paris there's an American bar that's been "predicting" presidential elections results since 1924, but there's nothing scientific in their methodology, it's just for fun !

  • $\begingroup$ But how did these people construct their sample? $\endgroup$
    – Heisenberg
    May 9 '15 at 9:01
  • $\begingroup$ Answered your question in an edit. $\endgroup$
    – Antoine R
    May 9 '15 at 12:17
  • $\begingroup$ But then the sample will be biased, no different from that of Literary Digest. $\endgroup$
    – Heisenberg
    May 9 '15 at 14:26
  • $\begingroup$ Absolutely, but people who knew these kind of samples were biased didn't recognized survey sampling as a valid statistical method back then ; and those who didn't realize it simply ended up with wrong estimations. $\endgroup$
    – Antoine R
    May 9 '15 at 15:36

To the best of my knowledge, in 1936, George Gallup used a version of quota sampling. It worked in some cases, like in 1936, and failed in others, like just 12 years later in 1948, when Gallup, like some others, said that "Dewey defeats Truman".


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.