The COVID-19 pandemic brought chaos and disruption to university operations, including college admissions. Testing centers closed, and 61% of applicants' SAT scores vanished. This meant that a key dimension was missing from our newly developed performance-based scholarship grid, which had led to near-record enrollment. In response, we developed an improved scholarship grid using class rank as a secondary dimension. But this approach created a new problem: 22% of applicants came from high schools that didn't rank their students. We considered using a common mean imputation process for handling missing data, but this approach disadvantaged potentially high-performing students, especially Hispanic students. Forcing non-ranked applicants to be "average" was not an appealing option when we're looking to reward overlooked and undervalued students with higher scholarship awards than what they can get at other schools. The better solution was to use a more sophisticated SAS®-powered predictive model imputation process that utilizes other known information about unranked applicants to place them on the grid more precisely. This approach dramatically increased the support for our Hispanic applicants, with 88% receiving higher rank placements and 64% receiving higher scholarship awards than would have occurred with the more common mean imputation method. This session detailed the problem, explained key model relationships and revealed SAS coding secrets.
Presentation slides are attached to this post.
