Wednesday, March 16, 2016

A Query About Queries

I have amazing readers.  I’m hoping some of you can help me with this one.

If you were trying to help a community college improve its students’ success rates, what statistics would you find the most helpful?

Let’s assume that you have some of the usual suspects.  For example, let’s assume that you already have access to the average pass rate, the average retention rates (fall to spring and fall to fall), and the usual variations on graduation and transfer rates.  (The latter refers to transfer before graduation, which is relatively common at many community colleges.)  Let’s also assume that you’re able to disaggregate each of those by race/ethnicity, gender, part-time/full-time status, and age.  

What else would you want to look at?

The goal is to identify areas in which changes or interventions are likeliest to make the most difference.  

Because I’m working in a community college setting, where “open admissions” is part of the mission, the goal is not to exclude risky students.  (“Drown the bunnies” is not an acceptable answer.)  Selectivity is out of bounds.  The goal is to help as many real students as possible to succeed.

Resources are limited, which implies a few things.  First, it suggests that concierge-level service for every student -- which would probably do wonders for completion rates -- isn’t gonna happen.  Wildly expensive solutions may be theoretically interesting, but in practical terms, they’re irrelevant.

Second, limited resources apply not only to the interventions, but to the data gathering itself.  Our ERP system isn’t perfect, and we can’t afford to hire dozens of new institutional researchers to look under every rock.  

Finally, there are both ethical and political considerations to weigh.  For example, I take the ethical position that system fixes are separate from individual performance fixes.  Assume, for the sake of argument, that the personnel won’t change.  That means looking at structure and process, rather than individual people.  And assume that any key statistic you use to make decisions may become public, and may be used in very different contexts for different agendas.  Some level of that may be unavoidable, but it’s best to steer away from measures that are likeliest to backfire.  

What would you look at?