Tuesday, August 17, 2010

How much damage will be done by gutting the census?

Stephen Gordon at Worthwhile Canadian Initiative thinks a great deal:

Before the decision went public, Statistics Canada estimated that the response rates for the long form would fall from about 95% to 50%, but it could be brought back up to 70% if they spent a lot of money pestering non-respondents. This result was of course deemed to be unacceptably low a few months ago, but even this level of failure seems like an unattainable dream now. Now that the census has been made a playground for partisan politics, many CPC partisans will decide that their party's cause will be best served by not filling out the voluntary forms, and many opposition supporters will boycott the long form in protest. The long form data will be a dog's breakfast.

And there will no doubt be many people who will have missed the distinction between the short and long forms and will wrongly assume that they can pitch their short forms into the recycling bin without consequence. After making such a big deal about not forcing people to fill out the census form, the government will be hard-pressed to justify using the powers of the State to force citizens to fill out the short form. There's a real risk that the short form data won't be usable, either.

These are problems that will explode in the government's face over the next few months. But it won't end there. As the months and years wear on, every single census release will be accompanied with a lengthy discussion of to what extent changes since 2006 reflect reality or the failed 2011 census. This is going to go on for years; a cursory search kicks up at least 11 discussion papers based on the 2006 census published by Statistics Canada in the past 12 months. (See here for a summary of output from the 2006 census.) And that's just Statistics Canada. It doesn't include studies done at other federal departments such as Finance or HRDC, or by other agencies at other levels of government.

Of course, non-government researchers will have to devote any number of person-years dealing with the wonky 2011 numbers. Many future studies will no doubt be obliged to use a binary indicator variable to capture the 2011 outlier; this indicator will be known as the Clement Dummy. There will be snarky variations on this notation during seminars.

And to make matters worse, Gordon points out that even if the 2016 census is conducted properly, this will seriously mess up a lot of longitudinal studies. It will be much harder to determine empirically if government programs are working properly, which of course is what the Cons probably want. After all, if they want to cancel a program for ideological reasons, it will be a lot easier to justify if there's insufficient information to determine how well it works than if there's a whole bunch of data showing that it works well.

No comments: