By: Mario Canseco, VP Communications & Media Relations

It’s one of the questions pollsters and market researchers receive on a constant basis: How can you be certain that 1,000 people speak for an entire country? This phrase, or variations of it, is usually followed by other comments that attempt to question the validity of a specific research finding. “None of my friends ever try rosé, so I can’t believe that 40 per cent of people drink it.” “Everybody I know wants to legalize marijuana, so 52 per cent in favour seems too low.”

There are two inherent problems with this approach. The first one suggests that a pollster or market researcher would need to speak to a much higher number of respondents in order to come up with a truly representative sample. The second one assumes that a person’s friends, relatives and co-workers amount to an exact model of the population. These two problems are to blame for the polling industry’s earliest and biggest disappointment: the Literary Digest prediction of the 1936 United States Presidential Election.

The Literary Digest survey relied on one of the largest sample sizes ever used, receiving responses from 2.3 million people. In the end, the magazine incorrectly predicted that Alf Landon would defeat Franklin Roosevelt. The prediction failed because the magazine was more widely purchased and read among Republicans than Democrats, which meant that the sample did not accurately reflect the voting population. It would not have mattered if 2.3 million or 230 people had answered the survey.

One common misconception is that the margin of error—a mathematical calculation that takes into account the size of the target population and the number of people the polling company contacted—would be considerably lower (or even inexistent) if more and more people are interviewed. However, assuming a 95% level of confidence (the often cited “19 times out of 20”), contacting 5,000 people would yield a 1 per cent margin of error. The margin of error for the Literary Digest poll was 0.06 per cent. But with a sample that was not properly constructed, the number becomes meaningless.

Good research entails ensuring that the research accurately depicts the public at large. Vision Critical’s Global Panels—the Angus Reid Forum, Springboard USA and Springboard UK—contain enough people in each major demographic group in Canada, the United States and Britain to draw random samples that represent the population as a whole.

A first-rate survey must be based upon randomly-selected, representative samples that are statistically weighted according to the most current demographic, regional, and political voting data available. What rang true for Literary Digest in 1936 is still pertinent in the 21st Century: the structure of the sample is infinitely more important than its size.