Will telephone surveys or self-filling surveys with, say, a million respondents be more accurate than public opinion polls with a sample of 1,000 respondents?

It would be logical for 1+1 to be 11 (because if two units are added then their sum must be 11?), but we all know that is not true. That is the answer to this question: logical, but incorrect.

A biased sample is biased anyway, no matter how large it is. The best example that proves this claim is the presidential election in the USA in 1936. The magazine, Literary Digest, sent out 10 million letters asking people how they would vote. They received almost 2.3 million responses and according to them, the results were that Alfred Landon leads over Franklin Roosevelt by 57:43 percent. The Digest did not collect information that would allow to evaluate the quality of the sample and to correct the groups (to value their answers more or less) that are more represented. Since the Digest sent postcards primarily to individuals with cars or telephones, their “sample” consisted of few working-class people.

The young pollster George took a much smaller sample (50,000, though a far larger sample than is taken today), but he provided a representative sample, and because of that his poll showed Roosevelt on the way to a landslide victory.