Optimization and college enrollments

This is Joseph.

Dean Dad had a great post a while back on trying to improve classroom scheduling.  In it, he had a great example of the perils of overfitting.  The context is that the Gates Foundation had suggested that a more efficient algorithm for classroom management could improve costs.  The problem is that they aren't testing this in prospective environments -- leading to some issues:

Finally, and I can’t stress this enough, enrollment is a moving target.  Patterns change unevenly, so in any given year you’ll get a fresh new crop of anomalies.  When enrollment continues until just before classes start, as is typical at community colleges, you’re left making predictions based on partial information.  And you have to make those predictions early enough for alternate plans to be realistic.  That’s never a perfect science -- every year, someone complains that his course would have filled if we had just given it more time, which is unprovable either way -- but it’s inherent in being “responsive.”  If we locked down enrollments months in advance, we would have time to squeeze out some efficiencies.  With numbers changing until the last minute, it’s harder.  You’ll never capture that if you only look at “snapshots.”

It's also worth noting that the staffing side is also an issue.  Paying for "Just in time" staffing for a community college would almost certainly reduce the qualifications of the instructors, do strange things to the workload of permanent staff, or increase prices.  After all, if you do not know if you are going to be taking a class until the first day of classes (when enrollments are likely close enough to know) then you will likely charge more or look for another job.  Landlords are notoriously unsympathetic to "most of the time this works and I can pay rent" pleas.

So the report might be interesting as an estimate of the maximum possible benefit.  But in the real world you never get 100% efficiency in any process, especially one that has ebbs and flows.  Even McDonalds, good as they are (and they are really good at this) cannot completely eliminate wait times due to the lunch rush.