The Empirics of Learning from Failure

The ability to learn from experience is central to an organization’s performance. A set of qualitative management studies argues that learning from failure is the exception rather than the rule. Another literature, using econometric methods, finds strongly statistically- and economically-significant effects. There are many possible explanations for this discrepancy, but we argue that one contributor is that a problem with one of the standard empirical approaches to identifying learning from failure may result in erroneously significant results. We generate simulated placebo data in which no learning takes place and show that the standard approach yields strong significant results. We provide a simple example that provides intuition for why this might be. We then propose and implement improved specifications using data on liver transplantation and find no direct evidence of learning from failure.