Reducing Bias in a Personnel Assignment Process via Multiplicative Utility Solution
比较了传统稳定匹配算法与乘法效用解法在人员分配中的公平性，模拟表明乘法效用解法能更平等地对待申请者和雇主群体，且偏见随问题规模增大而增大。
In this paper we consider a type of bias which stems from the mathematical algorithm often used to determine an optimal match between two groups. We compare two different solution concepts for the matching assignment problem: the traditionally stable solution vs. a multiplicative utility approach that should avoid the bias. Simulation modeling leads us to conclude that: (1) With respect to all sizes compared in our experiment, applicants and employers groups were always treated far more equally under the multiplicative utility approach than the stable approach. (2) When using the stable algorithm, the size of the bias is affected by the size of the problem (i.e., the larger the problem size is the larger the performance discrepancy between two groups of participants). Further analyses of the sensitivity of the finding to different assumptions (e.g., correlation in the preference orderings of the two groups, or use of nonlinear conversion from ranks to utilities) also resulted in a superior performance of the multiplicative utility algorithm in terms of a more equitable outcome.