Statistics Canada uncovered serious errors in the latest batch of data from its national household survey and pulled the plug less than 48 hours before its slated release Wednesday.
The decision Monday to postpone the report until Sept. 11 is the latest controversy to hit the 2011 voluntary survey, which replaced the mandatory long-form census scrapped by the Conservative government in 2010 over privacy concerns.
Municipalities, economists, cultural and religious groups, and a former chief statistician have argued that information gathered from the voluntary survey would provide a less reliable picture of Canadian society. The data are used to plan a range of services, from roads and hospitals to low-income housing and recreation centres.
This survey is the first comprehensive look at the lives of Canadians since 2006 and those who use the results have complained that serious gaps in two earlier data releases from the survey render key information unreliable.
Statscan said it identified this latest problem with the data – including details on what Canadians earn and how they live – as officials were going through last-minute "quality" checks over the weekend.
Among other things, the widely anticipated final installment of the survey was expected to provide an updated portrait of the growing gulf between the haves and have-nots in Canada, including where they live and how much money they make.
"We were in the final stages and some of the results seemed odd, a bit," census manager Marc Hamel explained in an interview. "When we went back to the data-processing steps, we discovered that one of the steps was not applied correctly."
He denied the switch to a voluntary survey over the longer mandatory census had anything to do with the suspect data.
In a statement, Statscan benignly characterized the problem as a "data processing" issue.
Mr. Hamel said two early batches of data released from the 2011 survey in May and June are unaffected by the latest errors.
"It is unfortunate that it was in the late stages," Mr. Hamel acknowledged. "But it's lucky we found it before it was released."
Experts said the 11th-hour postponement is highly unusual and they speculated it could stem from the fact that the survey is new and less comprehensive than the census it replaced.
"This is very late in the game – as late as it can be," said Philip Cross, Statscan's former chief economic analyst, now research co-ordinator at the Macdonald-Laurier Institute in Ottawa.
Other former Statscan insiders said it isn't surprising that the agency encountered data problems because the survey marks uncharted territory. The survey involves much more extensive data manipulation than the census, they pointed out, and therefore is inherently much more prone to such errors.
The 2011 survey had a substantially lower response rate than the 2006 census – 68.6 per cent, versus 93.5 per cent. As a result, Statscan must make a greater number of adjustments to produce statistically valid data.
The postponement is also likely to cost Statscan a substantial amount of money.
Before the latest problem, Statscan said the 2011 survey would cost $652-million, or roughly 15 per cent more than the 2006 census.
The federal government sparked outrage among economists, policy makers and marketers when it announced in the summer of 2010 that it would scrap a mandatory survey sent to one in five Canadian households every five years. The government justified its decision on the grounds that the questions were too intrusive and said Canadians should not risk criminal prosecution for not answering them.
A shorter census questionnaire, sent to all Canadian homes every five years, remains mandatory. The more comprehensive survey of such things as race, income, immigration, education and housing was obtained through voluntarily responses.
Editor's note: An earlier version of this story incorrectly stated that the agency would have printed thousands of copies of the aborted report. Statscan no longer prints copies of its reports and uses only its web site to disseminate information.