The two people who were telling me how they lied on the census never complained to the privacy commissioner. They didn’t write their MP, they didn’t seek out the Ombudsman for Statistics Canada nor write a letter to the editor. They didn’t even, in four (or nine) years of drinking with me ever grumble about it. They may be in a minority, they may not. There is no “ability in real statistics” to reliably filter information out in a self-survey. There are techniques that, when applied retroactively in response to known data, can mitigate the inaccuracies (though on occasion cause more); but there is no “known data” in this case: just data that people are required to fill out which most do and StatsCan is left crossing their fingers that lying isn’t particularly prevalent.

As to whether or not it is or is not prevalent, however, isn’t entirely the point. The anecdote wasn’t to show “how widespread lying on the census is”, its to show that making the form mandatory doesn’t ensure that you’ll get a beautiful new set of pure data: instead of refusing to fill a question out (as Ezra Levant did in 1996) so StatsCan can say “7% of Canadians refuse to say if who does the dishes in their homes” the question is filled out incorrectly and StatsCan bravely publishes that only 13% of Canadian men don’t throw throw out the garbage. If they claimed the same percentage watched “Gilmore Girls” while Nielsen and its automatic testing box showed differently they could make adjustments. In the current system, they can’t. As social engineers ask more and more intimate questions to set big government policy, the gap between the truth and what people write down may grow in moderately unpredictable ways (I’m sure it shocks no-one that Albertans, according to a Ledger poll out today, are the most likely to lie or omit questions they don’t want Ottawa knowing the answers to).

We already see large scale distortions with social media and new sensitivity to privacy concerns. Like it or not, there is a slow (or maybe not slow! has bilingualism in non-Quebec francophones gone up or down?) shift in attitudes about adding information to databanks. To borrow your phrase, there is no statistical technique that will be able to fix this.

As for claims that many Canadians are concerned about their privacy when it comes to the questions asked on the long form, the Conservatives have been unable to provide any evidence to back up that claim. There have been three complaints to the privacy commissioner about the long form in the last decade and three out of 30+ million is a very, very small number. Similarly, I see that the blog posting you link to tells a little story about two people who lied on the long-form in two separate censuses. As Derek’s trackback on your blog posting says, that’s just “an anecdote without context to show how widespread [lying on the census] is, which is precisely the sort of potentially misleading information that real statistics are designed to counteract…”

What was said was that when the long form with its detailed list of questions is handed out, some people don’t feel comfortable giving that information and respectfully decline. They never do give out that information. In practice, a nonzero number of people never will fill out the long form. They will be threatened with legal sanctions — which for the long form StatsCan doesn’t go through with anyways — and ultimately the information that you want them to provide to the government never does get provided. If they begrudgingly do end up providing information on the form, it will likely be less valuable to StatsCan than no data whatsoever.

Census data will still be collected. Advocacy groups will still have “this critical research tool”. Companies will still pay assloads of money to Stats Can in return for access to this information. All that happens is that the missing datapoints won’t be from people looking over their shoulders to see when Statistics Canada is going to haul their ass to jail.

This blog is on issues affecting science trainees and a move like this which strips information from the public domain and many researchers is something we definitely see as having an effect.

Hope this helps explain a little why we found it important to post.

Dave