It matches everything I’ve been reading on http://greaterfool.ca for the last 7 years.

Like this from 3 months ago: https://www.greaterfool.ca/2019/02/02/tapped-out/

Well.MNP are licensed insolvency consultants so regardless of the sample size the entire survey smacks of bias.

There are way better indicators of insolvency than this opinion poll. We'd expect a related tick up in actual bankruptcies, or their precursors like auto loan and mortgage defaults.

Even so, it confirms my first hand experience. Canada looks fairly affluent looking from the outside in but most people live paycheck-to-paycheck and in rural areas there is poverty at levels that you would not expect in what is nominally a first world country.

Why? Is 2,070 not a significant sample size?

That would depend entirely on the sampling methodology.

Since “online surveys” are as a general rule very poorly constructed (frequently self-sampling) I believe it’s that, rather than the sample pool size, that the parent is responding to.

Additionally Ipsos Reed was hired to run this poll on behalf of MNP, whose press release this information is coming from.

MNP, among other things, offers debt consolidation services. There’s incentive here for Ipsos Reed to construct this sample pool to get a result that gets people worried about their debt load.

So how many of these 2000sampled families are actually on-line or have proper internet connections? How is this distributed across the country? How many have proper jobs? Are unemployed ? How does it match to the age distribution among all inhabitants? And why would 200 $ be a right cut-off for income?

Ah I gotcha, that's a fair point.

So, how is Sudbury doing ;) ?

No clue. Wish I had that kind of data.

The problem isn't that it's a bad sample size per se. Given Canada's population (~37M), you get something between 2-3% margin of error at 95% confidence level with that sample size. The problem comes down to how you get your samples — online polling especially is skewed by how you pick your audience. For example, if you only poll your friends, your friend's friends and so forth until you hit the number, you will likely see a pretty uniform result, which is obviously far from giving any reasonable insight.

Yeah, most presidential election polls around here survey roughly that amount of people - and they consistently get it 'right', most likely because they also 'sample it right'

This might very well be enough. For 35 million people 2% error is 2400 people. So I think we can say it is 46-50% of the population does not have $200.

However if the samples were already biased into certain groups or not even Canadian, then it gets thrown off.