Keith Devlin did a thing on NPR explaining how Polling Margins of Error work (the accepted view) but messed up on a couple of points in the history.....as pointed out by Peggie Lewis:
"Keith Devlin conflated two events when describing the polling errors in the Dewey Truman race in 1948. The main problem with that year's polling was that it stopped several weeks before the election and missed the late swing to Truman (See Scholastic News, 2008) The previous major polling error was made in 1936 and reported by the Litery Digest on the FDR Alf Landon race. The Digest mailed out straw ballots drawing names from telephone books and DMV records. Among many other things wrong with their polling methods was the biased sample this reflected (In the depression year of 1936, a sample drawn from those who owned autos and/or who had phones was necessarily a biased sample."
Thanks to Shelli Temple from the AP Stats group for bringing this to everyones attention.