The census is back with a swagger

Statistics Canada is headquartered in a tall, beige saltine-cracker-box of a building just west of downtown Ottawa, surrounded by other federal government buildings. The area is known locally as Tunney’s Pasture, but in spite of its bucolic name, you would probably dismiss it as a little too on-the-nose if you were a location scout searching for an aggressively drab government precinct in which to film. Everything about StatsCan’s outward appearance whispers of insular efficiency and exactitude. But the agency does not live in a vacuum: as last fall’s election approached, its leadership had their eyes fixed toward Parliament Hill, a few kilometres away.

Two of the three major political parties had inked into their platforms a promise to reinstate the mandatory long-form census. Preparations were already well under way for the 2016 population count. The agency couldn’t know who voters would pick, whether the Liberals or NDP would make good on their pledges to resurrect the census, or how quickly they would move, so behind the scenes, they started just-in-case preparations. “We had kind of a Plan A and a Plan B ready to go,” says Marc Hamel, director general of the census program.

Once Justin Trudeau’s party swept into power on Oct. 19, people in the know quietly warned the Liberal transition team that if they were going to restore the census, they had to act quickly. Wayne Smith, the chief statistician, even showed up for his transition briefing brandishing a hard copy of the census. Three weeks after the election, with an announcement from freshly minted Minister of Innovation, Science and Economic Development Navdeep Bains that the mandatory long-form was back, Plan A was a go.

Related: Why it’s not enough to simply restore the long-form census

This week, the furious preparations of the agency over the last several months come to fruition: May 10 is census day, when Canadians raise their hands to be counted. The voluntary National Household Survey that replaced the long-form census in 2011 ended up being neither the pointless disaster its staunchest critics had envisioned, nor the perfectly useful replacement its proponents predicted. It had serious limitations that caused 1,100 small communities to vanish off the statistical map; it produced a few weird findings that simply didn’t look right; and it made looking for change over time all but impossible. It did, however, offer a serviceable snapshot of the country. Now that StatsCan is returning to a mandatory long-form census—and in a hurry—the question is what will become of the evolving national portrait that underpins everything from people’s bus routes and commuter highways to their children’s schools and where they can grab groceries on their way home from work.

What was once the driest and most esoteric of citizen duties—the statistical backbone of the country that, frankly, most people were oblivious to—became an unlikely flashpoint in 2010. That July, then-prime minister Stephen Harper axed the mandatory long-form census, arguing it was inappropriate to compel citizens to answer questions about their education, work, ethnicity and housing, among other topics. Critics of the move—they were nearly unanimous among those who use census data, including researchers, municipal planners and community organizations—insisted that a mandatory census was the only way to get an accurate picture of who Canadians are and what they need.

Ultimately, 68 per cent of households responded to the 2011 National Household Survey (NHS)—far short of the 94 per cent that completed the long-form census in 2006, but better than the 50 per cent response rate StatsCan projected in some of its testing. The agency’s analysts did everything they could to verify and shore up the information they had. In the end, they released the data they believed was solid, but anything below a certain quality threshold—a highly technical measure that amounts to overall non-response combined with “item non-response,” or individual questions people skipped—was simply not released. “We were very transparent in saying that at the small community level, we cannot do the same level of validation,” says Hamel. That meant that 1,100 small towns and specks on the map, representing three per cent of the Canadian population, became statistical ghost towns, except for the basic information collected on the short-form census. If you wanted to know what the 1,400 residents of Shellbrook, Sask., do for work, how much education they have or their ethnic backgrounds, you’d hit a dead end.

Related: The crisis in government data

But even with all the quality control StatsCan conducted, there were a few odd glitches that spoke to the problems with using a voluntary survey to obtain a full portrait of your country. The NHS, for example, found that between 2006 and 2011, the largest proportion of Canada’s new immigrants came from the Philippines, followed by China. But a tiny numbered footnote attached to that observation warns that it doesn’t square with immigration records, which showed that in fact the largest slice of newcomers came from China. Presumably, a significant number of new Chinese arrivals either didn’t fill out the NHS or didn’t identify their recent country of origin.

However, the biggest problem with the 2011 survey is simply that it’s different. StatsCan told users flatly that the NHS results were useful for comparing different regions of the country at a single moment in time, but they shouldn’t be measured against 2006 or earlier census results, because the methodology had changed so fundamentally. And comparing data over time is “the most important single thing” for researchers, says Michael Veall, a professor of economics at McMaster University. Veall is quick to note that the NHS turned out better than he expected it would when he testified at a parliamentary committee hearing on the issue in 2010, but it still has serious limitations. “Statistical information is interesting when there’s a surprise, right?” he says. “So you find more people are doing this or more people are doing that. The trouble when we went from 2006 to 2011 [is] every time we see a surprise, we have to say, ‘Oh, is that because something really happened, or is that because there’s a problem with the data?’ ”

Even subtle changes in how a question is asked can affect comparability because of the different responses it might elicit. You can see evidence of this concern in StatsCan’s testing of its questionnaires. Minute tweaks to questions for the sake of clarity and maximizing response—should the clause “including yourself” go at the beginning or end of a question about how many people live in a household?—are balanced against the imperative to keep things as consistent as possible so results can be compared over time. The agency is perpetually in census-mode: the population count occurs every five years, but the preparation, collection, processing and releasing the data takes seven years in all, with cycles overlapping. By the time Bains announced the resurrection of the census, all the 2016 questions had been extensively tested, fine-tuned and approved.

In 2011, 54 per cent of households completed the short-form census online, and 45 per cent filled out the NHS that way. This year, StatsCan is pushing for an overall Internet response rate of 65 per cent. To that end, no one got an actual census form in their mailbox this week. StatsCan does mailouts to about 80 per cent of Canadian households, sending a letter inviting them to complete the census online with a secure code, or to call a toll-free number if they prefer a hard copy. By last November, those forms had already been printed, so people who opt for the old-fashioned method will fill out something entitled “National Household Survey,” which is now technically a census (the questions are identical; it’s just the voluntary or mandatory nature that shifts). Those who don’t receive a letter in the mail—primarily people in rural areas who use post office boxes tied to more than one dwelling—will get a paper form hand-delivered, to be mailed back when complete. Census mail is not individually addressed, so unless a census worker records which numbered form was given to which household, there would be no way to track who lives where. Lastly, about two per cent of Canadian households—some First Nations and those in very remote locations—will respond by in-person interview with enumerators.

Online responses are quicker for the agency to process because the data is already captured, and it’s also more accurate. The digital questionnaire forms a sort of Choose-Your-Own-Adventure path that steers people through the relevant questions based on their previous responses, and nudges them if they miss something.

Given the rapid back-and-forth on the status of the census, StatsCan has upped the “sampling fraction,” or ratio of households who receive the long-form, from the one in five who previously did to one in four for 2016 (the 2011 NHS went to one in three, in a bid to mitigate expected low response rates). That meant that where forms are hand-delivered, the agency had to reprint them to adjust the geographical codes, but Hamel is quick to point out that was a “cost-neutral” change. The 2016 census is estimated to cost $715.2 million in total; the 2011 NHS cost $22 million more than a mandatory census would have because the long-form went to more households and more work was needed to process the data.

This time around, Hamel is less concerned about confusion on the part of citizens than he is about outright rebellion. “I think the information is going to be fairly clear,” he says. “Is everybody in support of the return of the long form? Those are the parts that we’re not sure [about].” The 2016 census advertising campaign revolves around a sort of “What’s in it for me?” theme, with whimsical animations and earnest actors explaining how census answers shape the parks and doctors’ offices in people’s neighbourhoods.

The first results from the 2016 census will come out in February of next year, and the agency plans to have all the data published 18 months after census day—one year faster than the 2011 release, which was slowed by all the extra work required.

There will always be an asterisk next to Canada’s 2011 national portrait, but the reinstatement of the mandatory census for 2016 should render 2011 data more useful in the long run. The data collected this week cannot be used retroactively to “correct” the results from five years ago, but it will provide helpful context. Think about a given phenomenon—say, the employment rate among Millennials, everyone’s favourite demographic curiosity—like a line-graph, with dots corresponding to the figures for 2016, 2011, 2006 and so on back through the census cycles. Once the 2016 results are in—and presumably more robust and comparable to earlier iterations—where 2011 falls in this game of statistical connect-the-dots will suggest how much confidence or scrutiny those results deserve.

Veall wonders if the very debate over the census might inspire more people to resist completing it this time around (the penalties laid out in the Statistics Act are a $500 fine, a three-month jail term, or both, and the Liberals have refused to discuss whether they will move to change that). Veall would rather see a return to a “cultural norm” where people simply fill out the census because it’s one of the things that makes society tick. He likens it to jury duty: everyone knows someone who gets out of it, but most don’t kick and scream much when their number comes up.

Ian McKinnon, chair of the National Statistics Council that advises StatsCan, has come to believe that the temporary demise of the long-form census was, accidentally, the best possible advertisement for it. He’s lost track of the number of people with no connection to statistics who have told him with amazement that they had no idea, until the long-form controversy, how widely census data is used. “In the midst of some of the frustration about the diminishment of the robust data series we had, the optimistic side was, boy, a lot of people who never thought about it are now much more aware,” McKinnon says.

So far, it looks like he may be right. On May 2, the day the invitation letters appeared in Canadian mailboxes, the census website crashed under the rush of people filling out their forms. “We thank Canadians for their enthusiasm,” StatsCan said. You could practically see that sensible beige rectangle at Tunney’s Pasture flushing a pleased shade of pink.