Commentary on Eskreis-Winkler and Fishbach (2019): A Tendency to Answer Consistently Can Generate Apparent Failures to Learn From Failure

Abstract

Recent research suggests that failure undermines learning: People learn less from failure (vs. success) because failure is ego-threatening and causes people to tune out. I argue that the core paradigm (the Script Task) provides a confounded test of that claim. When people do not learn from test feedback, they may give internally consistent answers on a subsequent test. The Script Task’s scoring guidelines mark consistent answers as correct following success but incorrect following failure. As a result, differences in performance between conditions may result from equivalent learning combined with consistent responding when people do not learn. A descriptive mathematical model shows that lower performance alone is insufficient to conclude that people learn less. An experiment with U.S. Amazon Mechanical Turk workers demonstrates that a retroactive manipulation without feedback replicates the effect. Because the effect of failure on performance is confounded with consistency, the Script Task is not diagnostic regarding whether people learn less from failure unless consistency is ruled out.