We show that administrative hourly wage data exhibits considerable bunching at round numbers that cannot be explained by rounding of survey respondents. We consider two explanations—worker left-digit bias and employer optimization frictions. We experimentally rule out left-bunching by...