Paper Type
full
Description
Most organizations face distributed scheduling problems where private preferences of individuals matter. Course assignment is a widespread example arising in educational institutions and beyond. Often students have preferences for course schedules over the week. FirstComeFirstServed (FCFS) is the most widely used assignment rule in practice, but it is inefficient and unfair. Recent work on randomized matching suggests an alternative with attractive properties – Bundled Probabilistic Serial (BPS). A major challenge in BPS is that the mechanism requires the participants’ preferences for exponentially many schedules. We describe a way to elicit preferences reducing the number of required parameters to a manageable set. We report results from field experiments, which allow us to analyze important empirical metrics of the as signments compared to FCFS. These metrics were central for the adoption of BPS at a major university. The overall system design yields an effective approach to solve daunting distributed scheduling tasks in organizations.
Recommended Citation
Merting, Sören; Bichler, Martin; and Uzunoglu, Aykut, "Assigning Course Schedules: About Preference Elicitation, Fairness, and Truthfulness" (2019). ICIS 2019 Proceedings. 15.
https://aisel.aisnet.org/icis2019/data_science/data_science/15
Assigning Course Schedules: About Preference Elicitation, Fairness, and Truthfulness
Most organizations face distributed scheduling problems where private preferences of individuals matter. Course assignment is a widespread example arising in educational institutions and beyond. Often students have preferences for course schedules over the week. FirstComeFirstServed (FCFS) is the most widely used assignment rule in practice, but it is inefficient and unfair. Recent work on randomized matching suggests an alternative with attractive properties – Bundled Probabilistic Serial (BPS). A major challenge in BPS is that the mechanism requires the participants’ preferences for exponentially many schedules. We describe a way to elicit preferences reducing the number of required parameters to a manageable set. We report results from field experiments, which allow us to analyze important empirical metrics of the as signments compared to FCFS. These metrics were central for the adoption of BPS at a major university. The overall system design yields an effective approach to solve daunting distributed scheduling tasks in organizations.