Canada's Chief Statistician Wayne Smith.. THE CANADIAN PRESS/Sean KilpatrickSean Kilpatrick/The Canadian Press
Human error – is there any other kind? – has been blamed for the embarrassingly large discrepancy between Statistics Canada's labour-force numbers published on Aug. 8 and the much-revised numbers on Friday.
Statscan calls this an isolated error, but it was not really isolated, because it was part and parcel of a sweeping ten-year system update of the Labour Force Survey system. And the very fact that the agency is expecting to take another two weeks (or more) to figure out how the error could have gone undetected shows that the mistake and its consequences were not entirely off in some safe, discrete zone.
And Statscan's explanation of the mistake is a salutary warning to the general public, and the financial world in particular, to recognize that even when Statscan gets its Labour Force Survey data right, the monthly data release comes with a big margin of error. Statscan says its estimate of 42,000 jobs created in July is accurate within 28,500 jobs in either direction, two-thirds of the time. In other words, Statscan can say with confidence that July's level of Canadian job growth was somewhere between mediocre and world-beating.
People seem to have a fairly healthy skepticism about political polling, but a constant anxiety about the economy tempts us to grasp at the latest statistics, with an almost superstitious faith. Yes, over many months, the data becomes more and more accurate. But if looked at only on a monthly basis, the statistics can gyrate more than the reality they aim to represent. They are not a perfect count, but rather a set of estimates based on samples. They're also seasonally adjusted – another estimate, based on a judgment call.
The redesign of the Labour Force Survey, which occasioned the humiliating error, is all part of an attempt to estimate reality better. Notably, Statscan is trying to adapt to a world in which they hope to get information from people through the Internet – and from more people than previously – rather than old-fashioned phone calls to a landline listed in a directory, or face-to-face interviews. Obviously, there are bumps along the way.
The bad news of Statscan's uncharacteristic bungling is happily balanced off by the more encouraging revised results. But the graver error to which we are all prone is our overestimation of the short term.