A School of Emergent Priorities paper surveys borrowers across all fourteen umbrella-loan bins and finds Facilities' entrance-level Recovery Score explained almost entirely by distance to the nearest door, with the self-reported honesty term contributing next to nothing.
Surveying all fourteen loan bins, a new paper finds Facilities' Recovery Score tracks distance to the door far better than borrower honesty
A formula that names honesty as an input and then barely uses it isn’t lying, exactly. It’s just telling you where the honesty term actually sits in the design.
— Associate Professor Kwame Lindqvist, School of Emergent Priorities
Facilities’ entrance-level Recovery Score decides which of the University’s fourteen umbrella-loan bins get priority restocking each term, and its own documentation credits the decision to two things: how easy a bin is to reach, and how honestly its borrowers have behaved. A new paper from the School of Emergent Priorities set out to find how much work that second factor is actually doing.
Associate Professor Kwame Lindqvist and Dr Marek Solheim surveyed 163 registered borrowers across all fourteen sites on their return habits, paced the walking distance from each bin to its nearest entrance, and logged a week of peak-hour foot traffic. Modelling the published Score against the three measures found distance responsible for the substantial majority of the pattern, foot traffic a distant second, and self-reported honesty barely registering. Dropping the honesty term from the model and recalculating which bins would be restocked first left the University’s priority order almost exactly where it started.
The School reads the result as a case for checking what a formula’s own named inputs are contributing, not assuming the naming settled the question.
“We didn’t set out to catch anyone in a technicality,” said Dr Solheim. “The Score isn’t wrong about which bins need attention. It’s just quieter than its own paperwork about why.”
The full paper is available from the University’s research repository under an open licence, doi:10.5555/slop.cocm20.