Statistics Canada revealed that human error led it to classify working Canadians as being outside the labour force, causing the agency to issue a major correction to its July jobs numbers that had vastly under-reported the growth in hiring.
The agency said Friday that last week's error traces back to a reform in the way it processes data. Statscan is still investigating and promises a full explanation within two weeks.
After taking the unprecedented decision this week to pull last Friday's Labour Force Survey, the agency released a corrected report Friday that showed a gain of nearly 42,000 jobs, a spike that beat market expectations and was well above the mere 200 jobs the agency had originally reported.
News of the mistake earlier this week had forced Ottawa to put Employment Insurance claims on hold given that access to benefits is partly linked to regional unemployment rates.
The revised figures were a clear improvement over what had been a surprisingly grim report, but there is still plenty of concern over the fact that July's gains were entirely due to a rise in part-time employment. Full-time jobs were still down, though only by 18,000 and not 60,000 as the agency had originally reported.
The agency's chief statistician, Wayne Smith, said in a statement that this was an "isolated incident" and that Statscan will continue to publish high-quality statistical information.
But questions remain as to exactly how such a mistake could get through the agency's system of checks and balances. Many observers say the reputation of the internationally respected body has taken a hit.
"In my over two decades of experience in this business I struggle to think of a comparable foul-up anywhere in the world," said Scotiabank economist Derek Holt in a research note sent shortly after the report was released.
In an interview, Mr. Holt said his immediate reaction may have been a bit too strong. "I think they're still among the elite statistical agencies in the world," he said. "There's no doubting that this was an uncharacteristic but rather large mistake by Statscan on this particular one that unfortunately blemishes what is otherwise a pretty solid reputation."
The incident brings Statistics Canada back under the political microscope after a series of controversies in recent years. The Conservative government's decision to replace the mandatory long form census with a voluntary household survey was a major source of debate. Chief statistician Munir Sheikh resigned in 2010 after then industry minister Tony Clement claimed that Mr. Sheikh and the agency supported the move.
Since then, the agency's budget has been cut by $29.3-million over the past two years, leading to a reduction of 767 jobs, representing an 18.5-per-cent reduction in staffing.
The agency has terminated several reports and surveys to save money but insists it shielded the Labour Force Survey from any cuts.
"I can't say whether the funding cutbacks and the siege atmosphere that is evident at Statistics Canada contributed to this particular mistake, but they certainly have contributed to Statistics Canada's tarnished reputation," said Jim Stanford, a labour economist with Unifor. "You've had a government now for eight years that's often hostile to what I would call fact-based policy discussion and I do believe that has diminished Statistics Canada's standing."
Ivan Feleggi, who led Statistics Canada from 1985 until his retirement in 2008, said Friday's correction should be viewed as an isolated event and not linked to other issues.
"Accidents occur in any organization," he said. "People shouldn't make a connection if there isn't one."
Friday's revision saw the unemployment rate decline 0.1 percentage points to 7 per cent, the same change as in the original report.
Statistics Canada's monthly jobs numbers are based on a survey of approximately 56,000 households in Canada. The national employment numbers have a standard error of 28,500 jobs in either direction, meaning the agency is confident that, two-thirds of the time, the real jobs gains for July are between 13,200 and 70,200 jobs.
That margin of error is part of the reason why the month-to-month jobs numbers have the appearance of high volatility, an issue economists have flagged as a concern. As a result, many economists caution that it is more useful to rely on longer-term trends.