What causes innovation failures? Exploring the impact of different sources of vicarious learning from failure

Product recalls represent one of the most severe forms of innovation failure, with substantial economic, reputational, and safety implications. Despite their importance, prior research has only partially examined the determinants of recalls, often overlooking productlevel heterogeneity and the interaction between organizational learning mechanisms. This study addresses these gaps by integrating innovation management and organizational learning literatures to examine how direct and vicarious learning from failure affect recall likelihood. Using a multilevel framework and a comprehensive dataset of 49,011 medical devices filed by 10,115 firms with the U.S. Food and Drug Administration between 2000 and 2022, we estimate multilevel probit models that account for both product- and firm-level determinants. The results reveal an inverted U-shaped relationship between direct learning from failure and recall probability. Moreover, vicarious learning from failures experienced by prominent competitors significantly flattens this relationship, indicating a substitutive effect between direct and vicarious learning, whereas learning from severe failures plays a moderating role only within this reference group. This study advances research on innovation failures by clarifying the multilevel determinants of product recalls and showing that learning from failure is non-linear, bounded, and socially structured.