# The Price of Diversity

**Authors:** Hari Bandi, Dimitris Bertsimas  
**arXiv:** [2107.03900](https://arxiv.org/abs/2107.03900)  
**Date:** 2021-07-03  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2107.03900v1/

## Key points

- Authors Hari Bandi and Dimitris Bertsimas address systemic bias in datasets related to individual choices, focusing on gender, race, and ethnicity.
- Society faces a challenge in combating bias and achieving diversity while maintaining meritocracy.
- The authors propose an optimization approach involving flipping outcome labels and training classification models simultaneously to promote diversity without significantly compromising meritocracy.
- They introduce an implementation tool using optimal classification trees to identify necessary changes in selection processes for promoting diversity.
- Case studies on parole, admissions to the bar, and lending decisions using real-world datasets demonstrate the effectiveness of their approach.
- Achieving diversity can come at a low or negative cost, indicating that modifications can enhance diversity without adversely impacting meritocracy—in some cases even improving it.

## Summary

In their paper titled "The Price of Diversity," authors Hari Bandi and Dimitris Bertsimas address the issue of systemic bias in datasets related to individual choices, particularly concerning gender, race, and ethnicity. They highlight the challenge society faces in combating this bias and achieving diversity while maintaining meritocracy. To tackle this problem, the authors propose a novel optimization approach that involves flipping outcome labels and training classification models simultaneously. This method aims to identify necessary changes in selection processes to promote diversity without compromising meritocracy significantly. Additionally, Bandi and Bertsimas introduce a unique implementation tool utilizing optimal classification trees. This tool offers insights into which attributes of individuals lead to label flipping, aiding decision-makers in making informed changes to current selection processes in a comprehensible manner. The authors present case studies on parole, admissions to the bar, and lending decisions using real-world datasets to demonstrate the effectiveness of their approach. Through their research, Bandi and Bertsimas show that achieving diversity comes at a low or even negative cost, indicating that modifications can be made to enhance diversity without adversely impacting meritocracy—in some cases, even improving it. Overall, their work provides valuable insights into addressing systemic bias and promoting diversity in decision-making processes while upholding principles of fairness and equality.