Tuesday, August 3, 2010

Conservatives and Statistics

A lot of people think that the Conservatives are getting rid of the mandatory census long form to deliberately produce unreliable data that cannot be used to attack their policies. I don't think the Tories are that devious - they are probably just incompetent with data, and therefore don't appreciate the value of accuracy.

For example, in the Stockwell Day crime flap today, we got several indications that the Tory cabinet sucks with data - they fail to use it correctly even when it supports their case:

- Day did not bother producing any data to back up his assertion that crime is actually increasing in Canada (despite a drop in reported crimes) because more crimes are going unreported to police.

- Later, his colleague Rob Nicholson came to the rescue with a complete non sequitur, stating that 34% of crime in Canada go unreported, without mentioning whether the proportion has gone up.

- Upon examining the actual 2004 Statistics Canada report, one realizes that 34% is actually the proportion of crime that is reported, and in fact 64% of crime is unreported (with 2% unknown). [By the way, the AP seems to have done the verification, while the CBC just reported Nicholson's statement. (Hopefully this will have been fixed by the time you click on the link.) Journalistic rigour, CBC?]

- That 34% reported crime figure is down from 37% in the previous report for 1999, which is itself down from 42% in 1993.

So the data suggests that indeed, an increasing proportion of crime is going unreported. But the Tory ministers either didn't bother with it, or stated the level of crime reporting both erroneously AND without referring to the time trend, which is required to make their case. You really wonder how these people made it through university.

By the way, the next report, covering 2009, is due next month, so we're debating these figures exactly at the time where we have the most outdated data. Are the Tories afraid that the rate of crime reporting has stopped decreasing? Again, my guess is that they were instead simply guided by anecdotal evidence, electoral interest and ideology.

Besides, in the 1999 report, StatCan suggests that part of the decrease in crime reporting may be driven by lower damage in property crimes (smaller proportion of $1,000+ cases and bigger proportion below $100) and higher insurance deductibles: many people won't bother reporting a theft if they wouldn't get reimbursed anyway. In fact, reporting of violent crime has increased from 31% to 33% between 1999 and 2004.

So even if one believes that crime incidence has increased, it's unclear that we should be worried about it: serious crime appears to be decreasing, so it may well be that the overall social harm from crime is also decreasing. And even if one agrees that crime is harming Canadians more than before, it's another logical leap to conclude that we should build more prisons. After all, Québec, with its relatively light approach to punishment, has much less violent crime than the rest of the country. (Of course, you could argue that the causation runs the other way, but the point is that it's not at all clear that more punishment leads to less crime. What is clear is that prisons cost money.)

But, even given supportive data, Conservative ministers are unable to craft a coherent argument for the simple proposition that crime incidence has increased. Is it any surprise that these people don't think much of weakening the census?

No comments: