How Do You Apply Best Practices to Address Statistical Discrimination?

Data Fluency Data Literacy Training
Avatar
Author

Let’s walk through a case study to illustrate how to mitigate statistical discrimination. 

Case Study Background: A mid-sized finance company, FinCo, specializes in providing loans to small businesses. They have been in the industry for over a decade and have recently faced allegations of discriminatory lending practices. The company wants to address the issue of statistical discrimination in its decision-making process to ensure fairness and equal opportunity for all applicants. 

Section 1: Identifying Biased Variables
  • Challenge: How can FinCo identify and address biased variables in its lending process? 
  • 🚫 Incorrect approach: FinCo relies solely on historical data to determine which variables correlate with loan repayment without considering potential biases. This approach might perpetuate existing inequalities, as historical data may contain biases against certain groups. 
  • ✅ Correct approach: FinCo conducts a thorough audit of its lending criteria, including an analysis of the variables used in decision-making. They consult with experts to identify potential biases and work on mitigating them by either removing the biased variables or adjusting the algorithm to account for these biases. 
Section 2: Fairness Metrics
  • Challenge: How can FinCo establish and implement fairness metrics to reduce statistical discrimination? 
  • 🚫 Incorrect approach: FinCo decides to use a single fairness metric, such as equal opportunity, without considering other dimensions of fairness. This might lead to unintended consequences and overlook other aspects of discrimination in their lending process. 
  • ✅ Correct approach: FinCo considers multiple fairness metrics, such as demographic parity, equal opportunity, and equalized odds. They work with experts to determine the most appropriate combination of metrics for their context and monitor the lending process to ensure these fairness goals are met. 
Section 3: Diversifying Data Sources
  • Challenge: How can FinCo diversify its data sources to minimize the risk of statistical discrimination in its lending process? 
  • 🚫 Incorrect approach: FinCo relies solely on its existing data sources, which may be biased and lack representation of certain groups. This approach might perpetuate existing inequalities and make it difficult for FinCo to provide fair lending opportunities to all applicants. 
  • ✅ Correct approach: FinCo actively seeks to diversify its data sources by incorporating alternative data, such as utility bill payments, rental history, or social media activity, to assess creditworthiness. This approach can provide a more comprehensive and inclusive view of applicants, reducing the risk of statistical discrimination and promoting equal opportunity. 
Section 4: Continuous Monitoring and Improvement
  • Challenge: How can FinCo ensure ongoing compliance with best practices to address statistical discrimination in its lending process? 
  • 🚫 Incorrect approach: FinCo implements changes to address statistical discrimination but does not have a system to monitor these changes’ effectiveness continuously. Without ongoing monitoring, FinCo might not identify new issues or ensure that the implemented best practices continue to be effective over time. 
  • ✅ Correct approach: FinCo establishes a continuous monitoring and improvement process, including regular audits of their lending criteria, fairness metrics, and data sources. They also encourage open lines of communication with stakeholders, such as employees, applicants, and regulators, to identify potential issues and continuously improve their practices to minimize statistical discrimination.
Related Tags: