A Cognitive Framework for Modeling Coincident Software Faults: An Experimental Study

Abstract

The question of when different programmers tend to commit the same errors is a critical issue for achieving fault diversity in fault tolerance. This problem is interdisciplinary and related to theories of human error in cognitive psychology. This paper proposes a psychological framework that combines Rasmussen’s performance levels with cross-level errors, represented by post-completion error, to model situations in which different programmers are prone to making the same errors. To validate the framework, we conducted an experiment where 200 student programmers independently solved the same problem, with the same tool and language. The results indicate that programmers unlikely commit the same errors in skill-based performances, most likely make the same errors in rule-based performances. These findings suggest that natural independent development may be less effective in preventing common errors in functions involving rule-based performance and post-completion scenarios, whereas it could be effective in preventing common errors in skill-based and knowledge-based performances. The results provided new insights into the strategies for avoiding coincident faults in N-version programming, from a human factor perspective.