Skipped computer test blamed for StatsCan error

Embarrassing errors in provincial GDP numbers released by Statistics Canada in April were the result of a failure to thoroughly test a new computer system, newly released documents show.

Saskatchewan error flagged by Royal Bank; stats for most provinces revised

Embarrassing errors in provincial GDP numbers released by Statistics Canada in April were the result of a failure to thoroughly test a new computer system, newly released documents show.

The agency issued the erroneous numbers early on April 28, and that afternoon got a telephone call from a Royal Bank economist raising doubts about the accuracy of the figures.

Robert Hogue told the agency that the Saskatchewan number was seriously at odds with the Royal Bank's internal growth estimate for that province.

Days later, senior StatsCan officials confirmed the error, and on May 2 "suspended" the release on the agency website.

And two weeks after being alerted to the bogus numbers, Statistics Canada finally corrected the original figures, in a web release on May 11.

The revisions pegged Saskatchewan's economic growth in 2010 at 4.4 per cent, rather than the anemic 1.4 per cent of the April 28 release. Most of the other provinces' GDP numbers were also revised upward, including Newfoundland's and Manitoba's.

Spokesman Art Ridgeway at the time blamed a new computer, saying "two steps got reversed in order."

But internal documents obtained by The Canadian Press show the problem in fact arose from a failure to properly test the new data-crunching system.

"Testing was done using previously published data 2002 - 2008/09," says an investigation report. "However, (there was) no parallel run for 2010 provincial GDP data as such."

Funding, expertise 'challenges'

The documents do not indicate why the crucial test was omitted, but refer in general to "challenges in regards to funding and finding IT expertise for IT conversion, especially given the complexity and scope of the project."

Material related to the internal investigation was obtained by The Canadian Press under the Access to Information Act.

StatsCan's old computer system was at the end of its life cycle, scheduled to be replaced in March 2010. But because of problems and delays, the plug was pulled only in December — and the new system brought into service without full testing.

A May 11 briefing note to then-industry minister Tony Clement alerted him to the revised numbers but did not indicate why the problem arose.

Indeed, a StatsCan strategy document suggests the agency wanted to play down the error as far as possible.

Officials debated whether to issue the corrected provincial GDP numbers as a new release in a scheduled May 11 web posting of data.

"While this would more clearly flag the availability of the new results to all users, due to the sensitivity of the issue and the proximity to the Census, it was not recommended at this time," says one document.

Instead, the agency simply corrected the GDP numbers in its original April 28 web posting.

The incident tarnished the reputation of an agency that prides itself on accuracy and that has been targeted by the Harper government, notably last year when Clement ordered the demise of the mandatory long-form census in favour of a voluntary survey.

A spokesman for Statistics Canada did not respond to a series of questions about the erroneous release, except to say the agency believes it properly flagged the correction.

"All relevant systems have been reviewed and all required corrections implemented for 2010 and future releases," Peter Frayne said in an email.

The agency has been caught issuing bogus numbers previously. In 2006, StatsCan quietly corrected an inflation number — noting the published rate had understated inflation by half a percentage point since 2001.

The Canadian Press reported that agency staff had warned senior officials about the erroneous numbers for two years before any action was taken to correct them.

Statistics Canada, under pressure to cut its budget, dropped a series of surveys and statistical reports last year to save about $7 million.