Managing Recruitment Processes with Multiple Position Types

Abstract

Problem Definition: Universities around the world recruit students for their undergraduate programs every year. In Hong Kong, for each university or unit within a university that offers multiple programs, the recruiter needs to determine a cutoff score for applicants applying to each program. Those with scores above the cutoff are placed on a list of qualified applicants, which is submitted to the central clearinghouse. Admission outcomes depend on applicants' choices of programs, their preferences, and the cutoffs of all the programs. While the applicants' choices are known to the recruiter, their preferences, as well as the cutoffs of the programs not managed by the recruiter, are not. The objective of the recruiter is to maximize the average quality of the students admitted, while ensuring that the number of applicants admitted to each program does not deviate from a preset recruitment target substantially. Methodology/Results: We provide a model framework to predict admission outcomes and optimize cutoffs. The model resembles assortment problems in retail and is computationally challenging. Despite useful properties that can reduce the computational complexity, direct computation is feasible only when the number of programs is small. We present an approximation that is easy to compute and asymptotically optimal. We test this approximation in simulation studies as well as in a case study. Managerial Implication: We compare our model framework to two heuristics. One is the linear inflation heuristic, which is commonly used in practices, and the other is the decoupled heuristic, under which programs are managed independently. We show that our model significantly outperforms both. Our model framework is applicable to any recruitment process where multiple position types are competing for the same pool of applicants, and hence joint optimization is needed.

Keywords: Recruitment, Multiple Position Types, College Admissions, Mixed-Integer Programming, Asymptomatic Analysis