The Problem With Surveys
Telephone polling before the vote indicated a large win for the Republican candidate, Thomas Dewey. So obvious did the result seem to be that the Democratic candidate, Then President Harry Truman went to bed resigned to defeat, and newspapers went to press with the headline, 'Dewey Wins!'
When the votes were counted, Harry Truman had unexpectedly won. What went wrong? What went wrong was the way the samples were conducted. People were phoned randomly by picking their numbers in the telephone directory. The numbers were randomly selected, but this method did not result in a random selection of voters being sampled, since ricjer people were over represented in the telephone directory, and they tended to vote Republican.
That lesson was learnt. For decades care was taken to conduct representative samples using quota sampling, according to which the population is segmented, and quotas from each segment taken according to their representation in the population.
For decades this seemed to work well. Until 2015, when the British voted unexpectedly to leave the European Union, and the American electorate elected Donald Trump as President, both in defiance of the polling data.
What went wrong now? Both of these were essentially anti establishment protest votes. People who had never voted registered in large numbers. Because they had previously been under represented in the electorate, they had been under represented in the voting surveys.