Thank you, Mr. Chair.
My position is well known, but I would like to reiterate it briefly. First, the census is enormously important, and the national household survey, whatever its name, is part of the census.
Why is it so important? Because many of the most significant national issues can only be assessed by it: from the progress of aboriginal people in catching up in education or living standards, to the impact on different groups of people of the 2008-09 financial crisis, the living standards of the elderly, the position of minority languages, the economic integration of recent immigrants, and so on, this is part of our regular national stock-taking. Indeed, some would say it is a critical part of the democratic accountability of governments at all levels.
Second, it is my professional assessment and that of the Statistical Society of Canada, as well as of the American Statistical Association and the French professional statisticians, that a voluntary national stock-taking could be deeply flawed. This cannot be proven scientifically. Indeed, it is conceivable, though hardly credible, that all those who choose not to respond to the survey are exactly like those who choose to respond.
However, in a practice extending over 50 years, I have never seen a study investigating the characteristics of non-respondents that would have concluded they are like the respondents on all the wide range of variables that are collected by the national household survey. I'm certain that the reason my professional colleagues in Canada, the United States, and France have felt compelled to write about our voluntary census is that they have not seen examples of such incredibly fortuitous behaviour either.
My third reason for being here is that most questions of interest are intrinsically relative, that is, they relate to the evolution of different groups over the medium and the long term. Indeed, my earlier examples are all of this character. Whether the aboriginal groups are catching up, whether recent immigrants are doing better or worse than earlier arrivals, whether the situation of minority languages has improved or deteriorated relative to 2006 or 2001, and so on.
If people behave differently as a result of the long-form census being voluntary rather than compulsory, then all these comparisons will be rendered potentially invalid. Since most variables change relatively slowly over time, the impact doesn't even have to be huge for us to be unable to differentiate between real change and illusions created by creeping biases.
Fourth, I'm here because I want to underline once more the fundamental difference between bias and sampling error. If we had a proper random sample and enumerated successfully most of those selected, our error due to having enumerated fewer than 100% of the population can be estimated. By contrast, it is exceedingly rare that bias can be estimated. The significantly lower response rate we can be pretty sure will be present, but we would normally have no idea even of its direction, let alone its magnitude.
That is why I referred to biases as “pernicious” when I testified last time. And that is why increasing the sample size from one in five to one in three solves the minor problem, but leaves the elephant in the room unaddressed.
My last point is to underline that nothing I've said is a criticism of Statistics Canada. I'm convinced they will do everything in their power to have as successful a survey as possible under the circumstances.
They will probably do better than almost anyone could have done under the circumstances, but we cannot expect miracles. They have an incredibly difficult task in front of them, not only to collect the best possible data, but subsequently to provide usable guidance to their hundreds of thousands of users about which data are likely to be relatively safe and which should carry the equivalent of what chemical companies label with crossbones.
With all my heart, I wish for them and for our users that my forebodings should all miraculously turn out to have been utterly wrong.
Thank you.