The U.S. presidential election and the Brexit referendum are the most recent prominent examples of the shortcomings within the polling industry. They are by no means unusual or isolated illustrations. Indeed we have witnessed some provincial elections in Canada, notably in Quebec, British Columbia and Alberta, where late pre-election polls have also been substantially off the mark.
To address possible reasons for the underlying problem, it is useful to explain the assumptions underlying probability samples. The ability to extrapolate from a relatively small sample to a much larger population, rests upon the assumption that everyone in a large population, whether it be national, provincial, municipal or whatever, has an equal opportunity to be included in a random selection process. If this condition is met, the accuracy can be estimated within 3 per cent in a sample of 1,000, 19 times out of 20.
Continue reading at source...