The Literary Digest poll of 1936 is widely cited as a major turning point for intentions surveys. This magazine had a perfect record in predicting the outcomes of elections since 1920 and its forecast was off by only 1% in 1932. But in 1936 the Literary Digest predicted a landslide victory for Landon over Roosevelt (55% to 41%). The actual result was Roosevelt with 61% and Landon with 37%. Thus, there was a 20% error in predicting Roosevelts vote. This failure is considered to be a major factor in the demise of the Literary Digest in 1938.
Interestingly, Squire provides the first empirical study of the failure. He did this by drawing upon a May 1937 Gallop survey on the causes of the Literary Digests failure. Was the failure due to sampling bias (the most popular hypothesis, judging from prior discussion of this case), or upon non-response bias (a less popular hypothesis)? Squires analysis attributes part of the error to sampling. Nevertheless, this problem could not have been sufficient to cause the incorrect prediction. The most significant error was non-response bias.