Saturday 24 July 2010

StatsCan and mandatory censuses

The head of Statistics Canada has resigned over the Canadian government's switch from a mandatory long form census on a subsample of respondents to a voluntary long form census. He's certainly right that the two are not easily comparable; there's going to be a big break in StatsCan data going forward that's going to be pretty hard to adjust for in any time series work.

I would have loved to have seen, conditional on a move to a voluntary census, a transition year in which half of long form respondents are given the mandatory form while a treatment group gets the voluntary form. Then we could have at least some measure of the bias induced by the shift and, under an assumption that that bias doesn't change over time, a way of adjusting long form data after the change to make it commensurable with the earlier data series.

While I'm warmly reminded of tales of the Hong Kong governor's refusal to collect statistics that could be used back in London to direct things in Hong Kong, it's not terribly plausible that that's what's here going on.

Of course, even if the census is mandatory, there are always options for folks who don't want to be helpful. I'd link to the video, but the usual ridiculous geographic copyright restrictions prevent my doing so. Americans can see it here: Betty White's take on the Walken classic...

No comments:

Post a Comment