Maybe I'm an idiot and I've never run a survey but the sample size seems a bit low. If we assume half of the 38 million Canadians[0] are children and the remaining 19 million are married to each other AND they only asked 1 person per household this 31% number is based on the finances of only 0.02% of the total population.
[0] https://www150.statcan.gc.ca/t1/tbl1/en/tv.action?pid=171000...
A random sample of 500 with proper balancing reduces the margin of error for a 95% confidence level to 4.5%(<http://www.sciencebuddies.org/science-fair-projects/project_...>) regardless of total population size(<http://www.surveysystem.com/sscalc.htm>).
Now I'm curious where the survey was performed for example 2000 people from a city like Vancouver verses 2000 people from several major/minor cities across the county.
Regardless, thanks for the sample size references!
The basic idea is this. You know Canada has certain demographics. A part of the population is male and female, of a certain age, of a certain income, etc. You ask everyone in your survey those questions, as well as the things you want to understand, and then you can use some techniques to match your survey to the rest of the country. From here, then you can estimate what the entire country thinks about given just the survey.
As one can imagine, the process of matching called undersampling/oversampling matters a lot in practice.