This study explores the dynamics of organizational learning from accidents in the U.S. coal mining industry, with a focus on both direct and vicarious experiences. Using 24 years of panel data (2000–2023), we investigate how organizations’ attention to different types of failure information—frequency and severity (magnitude)—affects their subsequent performance. Findings reveal a U-shaped relationship between direct accidents and future performance, where moderate levels of accident experience enhance learning, but low or excessive accident experiences undermine it. Moreover, vicarious learning from the frequency of others’ accidents reduces an organization’s future accident frequency, while vicarious learning from the severity of others’ accidents increases an organization’s subsequent accident severity due to reputational damage spillovers. These results emphasize that the effectiveness of learning from failures hinges on the type of information attended. Theoretical contributions include a nuanced understanding of organizational learning mechanisms and boundary conditions, while practical implications suggest strategies for mitigating failure and optimizing learning in high-risk industries.
