Back in 2006, university Presidents got so mad at Maclean’s that they stopped providing data to the publication.  Recognizing that this might create the impression that they had something to hide, they developed something called “Common University Dataset Ontario” (CUDO) to provide the public with a number of important quantitative descriptors of each university.  In theory, this data is of better quality and more reliable than the stuff they used to give Maclean’s.

One of the data elements in CUDO has to do with teaching and class size.  There’s a table for each university, which shows the distribution of class sizes in each “year” (1st, 2nd, 3rd, 4th): below 30, 31-60, 61-90, 91-150, 151-250, and over 250.  The table is done twice, once including just “classes”, and another with slightly different cut-points that include “subsections”, as well (things like laboratories and course sections).  I was picking through this data when I realised it could be used to take a crude look at teaching loads because the same CUDO data also provides a handy number of full-time professors at each institution.  Basically, instead of looking at the distribution of classes, all you have to do is add up the actual number of undergraduate classes offered, divide it by the number of professors, and you get the number of courses per professor.  That’s not a teaching load per se, because many courses are taught by sessionals, and hell will freeze over before institutions release data on that subject. Thus, any “courses per professor” data that can be derived from this exercise is going to overstate the amount of undergradaute teaching being done by full-time profs.

Below is a list of Ontario universities, arranged in ascending order of the number of undergraduate courses per full-time professor.  It also shows the number of courses per professor if all subsections are also included.  Of course, in most cases, at most institutions, subsections are not handled by full-time professors but some are; and so assuming the underlying numbers are real, a “true” measure of courses per professors would be somewhere in between the two.  And remember, these are classes per year, not per term.

Classes Per Professor, Ontario, 2013

Yes, you’re reading that right.  According to universities’ own data, on average, professors are teaching just under two and a half classes per year, or a little over one course per semester.  At Toronto, McMaster, and Windsor, the average is less than one course per semester.  If you include subsections, the figure rises to three courses per semester, but of course as we know subsections aren’t usually led by professors.   And, let me just say this again, because we are not accounting for classes taught by sessionals, these are all overstatements of course loads.

Now these would be pretty scandalous numbers if they were measuring something real.  But I think it’s pretty clear that they are not.  Teaching loads at Nipissing are not five times higher than they are at Windsor; they are not three and a half times higher at Guelph than at Toronto.  They’re just not.  And nor is the use of sessional faculty quite so different from one institution to another as to produce these anomalies.  The only other explanation is that there is something wrong with the data.

The problem is: this is a pretty simple ratio; it’s just professors and classes.  The numbers of professors reported by each institution look about right to me, so there must be something odd about the way that most institutions – Trent, Lakehead, Guelph, and Nipissing perhaps excepted – are counting classes.  To put that another way, although it’s labelled “common data”, it probably isn’t.  Certainly, I know of at least one university where the class-size data used within the institution explicitly rejects the CUDO definitions (that is, they produce one set of figures for CUDO and another for internal use because senior management thinks the CUDO definitions are nonsense).

Basically, you have to pick an interpretation here: either teaching loads are much, much lower than we thought, or there is something seriously wrong with the CUDO data used to show class sizes.  For what it’s worth, my money is on it being more column B than column A.  But that’s scarcely better: if there is a problem with this data, what other CUDO data might be similarly problematic?  What’s the point of CUDO if the data is not in fact common?

It would be good if someone associated with the CUDO project could clear this up.  If anyone wants to try, I can give them this space for a day to offer a response.  But it had better be good, because this data is deeply, deeply weird.