Many will have seen news yesterday about large cutbacks in the works at Statistics Canada. On the basis of the news that lots of PSAC members had received notices that their jobs may be “affected,” a number of pro-Statscan commentators rushed to say that the agency needed to be saved because it provided such fantastic, non-partisan analysis.

Well, yes. But yesterday’s notices appear not to have gone to any analysts, since they are not PSAC members.  The employees who got notices would appear to be the ones involved in data collection, field interviews, data processing, etc.  Not to put too fine a point on it, these units are the ones that make Statistics Canada a target, because they are unbelievably expensive.

I have had occasion in the past to compare prices between a private sector data collection agency and Statistics Canada. Using exactly the same methodology, Statistics Canada costs were between two and three times that of the private sector agency. On any given survey, that’s going to be a seven-figure difference.

Now, obviously, quality costs money. And Statistics Canada, just by dint of being itself, has some advantages over a private sector provider in terms of getting higher response rates (people are a lot more willing to stay on the phone with “Statistics Canada” rather than some outfit running out of RackNine). But it’s not that big a gap; in the project in question, the difference between our project and a comparable one Statistics Canada had done was about twelve percentage points (65% vs. 77% if memory serves). Every percentage point matters, of course, but that’s a lot of money per point.

Besides, it’s not as though private sector firms are incapable of dispassionate and highly professional data collection. South of the border, data collection and analysis of the entire suite of post-secondary education surveys (the National Post-Secondary Student Aid Survey, the Beginning Post-Secondary Survey and the Baccalaureate and Beyond Survey) conducted by the U.S. Department of Education has always been awarded via competitive tender. As anyone who has used these studies knows, the work is world-class.

And, who knows? If budget cuts for surveys are significant enough, then Statscan might have to get serious about pursuing more solutions based on administrative rather than survey data, a development which – as I’ve argued before – could actually improve the quality and timeliness of the information.

So while the Tories’ past behavior has given all Canadians cause to worry about the honorability of their intentions, let’s not jump to the conclusion that a budget cut is necessarily a disaster. It might actually open the door to a better statistics agency.

People often berate Statistics Canada when it comes to producing data on education in Canada. And not entirely without reason: there are some statistics that Canadians seem to be especially bad at producing. But it’s also worth noting that there are other kinds of data that Statscan is extraordinarily good at capturing – data that researchers in other countries would kill to have.

When it comes to research, there are broadly two types of data. The first is factual, aggregate data, which usually comes from administrative sources and is usually analyzed by means of a time series. We use this approach to measure things like aggregate enrolments, tuition fees, number of professors, university and college finances, etc. Call this Type I data.

Statistics Canada’s record in obtaining Type I data is disappointing. We seem to have a ludicrously difficult time in this country collecting data on part-time academics, college students, or the number of new students entering into higher education in a given year. Consistent ancillary fee data is also a challenge. There are a number of reasons why this is so, not all of which are Statistics Canada’s fault. But at the end of the day, the Type I data products Statistics Canada puts out are pretty disappointing.

But Type II data is a totally different story. Type II is based on surveys, not administrative data, and at its best is longitudinal. This is the stuff that helps researchers draw empirical conclusions about the relationship between inputs and outputs. When it comes to Type II data, Statscan is the envy of the world. The data in the Access and Support to Education and Training Survey does a good job of covering key topics in education across the life cycle, even if the public use file is a bit on the irritatingly useless side. And the Youth in Transition Survey, a magnificent 12-year longitudinal effort linked to the PISA test, is quite simply the best data source for the transition from high school to adulthood anywhere in the world.

So is Statscan doing a good job or a bad job with educational statistics? Well, if what you’re worried about is the ability to compare Canada to other OECD countries every year in Education at a Glance, then it’s doing a bad job, because that’s all Type I data. But if you’re interested in understanding the dynamics of access and outcomes in education, then it’s doing an exceptionally good job.

One small problem though: Statscan’s core funding is devoted entirely to Type I data; the funding for Type II comes entirely from HRSDC, whose commitment to supporting these surveys is best described as “soft.”

Cause for concern.

With most of the world undergoing a serious bout of youth unemployment, there’s been a lot of focus on graduate earnings and whether or not we are “overproducing” graduates. As I’ve noted before, some of this talk is nonsense, but given the times, the focus on outcomes isn’t surprising.

Don’t tell Margaret Wente, but in China the government is actively cutting majors that don’t produce high levels of post-graduation employment. In the U.S., there’s an increasing number of stories (like this one from the Wall Street Journal) trying to point graduates to the “right” disciplines in a tight labour market. As others have pointed out in fact, a lot of the highest disciplinary rates – the ones that really attract attention – actually have really small enrolments. What the data really shows is that graduates of nearly all fields of study in America have unemployment rates lower than those of non-graduates (something that isn’t true in China).

Turning to Canada, we’re clearly doing better than most in terms of unemployment. What we’re not doing so well is producing timely data on graduate outcomes (doesn’t that WSJ data make you drool?). Our best data comes from the National Graduates Survey – which looks at people who graduated in 2005. Not much use in today’s environment.

Admittedly, this kind of data is expensive to collect via surveys, and that’s why a budget-challenged Statistics Canada isn’t rushing out to do more. More data may be available when the National Household Survey results arrive in 2013, but it’s still not clear how useful that survey will be.

But there’s another way to get this data. Statscan has information on nearly all Canadian students in its Post-Secondary Student Information System (PSIS). It is possible, using probabilistic matching, to link this data to the Longitudinal Administrative Database (LAD), which is a database containing the complete tax records of one out of every five taxfilers. With that kind of link, it is possible to get continuous, year-by-year updates on how well students are doing in the labour market, and to report it however we want, even by field of study.

A PSIS-LAD link would give us high-quality, timely, policy-relevant labour-market data – all with no new data collection costs. Can someone explain why we aren’t doing this already?