Introducing the Path Method: A Solution to Nonlinear Challenges in Variance Analysis and Explainable Artificial Intelligence

Introducing the Path Method: A Solution to Nonlinear Challenges in Variance Analysis and Explainable Artificial Intelligence

Accountants’ variance analysis shows which variations of drivers are responsible for variations in the performance variable of interest. However, the existing variance analysis suffers from critical shortcomings if the relationship between the performance variable and its drivers is nonlinear, which is usually the case. The authors suggest a novel method in variance analysis, called the path method, that avoids these shortcomings and applies to all continuous (including nonlinear) relations between drivers and a performance variable. They also outline how to improve the existing methods in variance analysis so that their shortcomings vanish and all methods in variance analysis yield the same results. The authors also show how DZ Bank uses the Path Method to address challenges in Explainable Artificial Intelligence due to regulatory requirements (e.g., EU AI Act).