STEMJazz Talklet with Teddy Mekonnen
Abstract
Teddy gave a dynamite and crisp talklet on Thursday on a very topical issue – ways to factor a desire for diversity into selection processes.
But first, let me issue a bias alert. Owing to my own discipline, I am extremely partial work that specifies metrics and algorithms that supply proof not only of convergence but achievability. That is, work that provides mathematical GUARANTEES. It also doesn’t help my bias that a key tool in Teddy’s work was majorization (also sometimes called stochastic ordering) – something I “invented” (as in reinvented decades too late :) ) for a problem in wireless communication. I almost fainted when I saw Schur (as in Schur convexity) on one of his slides. :) :) :)
Ok, my TMI bias disclaimer is done. :)
In any case, we all know about the 6/29/23 SCOTUS decision occasioned by the SFFA suit. We all know the precipitous drop in diverse admits to Ivy+ schools. And it’s reasonable to say that Brown (and perhaps especially the group of scholars that composes STEMJazz) is pretty exercised about it. The gist is that race/ethnicity can no longer be used in admissions, even though in the US it certainly codes for headwinds of varying strengths that thwart access to, well, just about everything! And these headwinds show up in systematic differentials in scores on standard “objective” metrics like the SAT/ACT, AP scores, access to enrichment activities etc. that control admission to higher education.
That schools are now forced to rely more heavily on these (known to be biased) metrics is a canonical Bad Thing since we have decades of experience that shows such metrics do not properly measure talent (i.e., the ability to rapidly acquire skills – such as those tested by the metrics – when provided access to superior education) in groups historically denied access to resources (HUGs).
So, what exactly has Teddy done? Well, he essentially says keep your “objective” measurements. Use them! But in addition, a selection process can (legally) have a criterion such as student body diversity. And you can define it any way you’d like (so long as race isn’t explicitly part of the definition). First, Teddy posits an integer lattice with as many dimensions as there are diversity categories. A class composition is a point in this lattice (a slice through the positive orthant called a simplex, for those who care about such things :) ). He then develops a “geometry” of fairness that: specifies what the optimal composition point is according to the diversity metric, identifies isomorphisms/equivalences between different class compositions (points on the lattice), identifies a region of good diverse composition – essentially a distance from that optimal point, presents a selection algorithm that “walks the lattice” of possible compositions – USING THE “OBJECTIVE MEASURE” but always moving toward the preferred optimal composition point, and finally shows that the algorithm is provably optimal (there are no fairer compositions possible that completely fill the seats in the class). Now, as you can imagine, there was VIGOROUS discussion and the alloted hour turned into 90 minutes before we we had the exit the venue. :) :) Charles was keen on the idea that showing how the solution compared to simply using the “objective” measures (likely to be the first point of attack). Everyone (but Monica especially) was a bit exercised about how diversity could be defined (i.e., in ways that ultimately disadvantage the populations that need help most). And everyone was worried about how opponents could argue that race is “embedded” in the diversity criteria and thus make the method legally assailable. And of note, Candace made sure we were all on the same mathematical page (as she is a mathematician :) ).
Nonetheless, what Teddy has done is a crisply clean mathematical start to fairer selection strategies. Now we’re hoping he can package the work for consumption by a wide audience (like admissions officers at various schools of interest) and get him a chattin with them!
