Notes: In Panel A, columns labeled “Junior” use the sample of workers whose tenure is less than 11 years (median of tenure in the sample), and columns labeled “Senior” use the rest. In Panel B, columns labeled “Male” use the sample of male workers, and columns labeled “Female” use the female sample. In Panel C, the propensity function P(a h) is defined for each worker by Prob(Overtime hours ≥ a|X) for each a (= 10 and 45) and estimated using a probit model using the BSWS data in 2009 (see Online Appendix Table A.3 for the estimation result), where X is the set of workers’ attributes. Columns labeled “High P(a h)” use the sample of workers having above the median level of the estimated P(a h), and columns labeled “Low P(a h)” use the rest of the sample. All regressions control for year fixed effects, establishment fixed effects, the log of intermediate input cost, and female dummy (except for Panel B), tenure, tenure-squared, age, age-squared, and three education dummies. Standard errors are clustered at the level of the establishment and are reported in parentheses. ^*p < 0.1, ^**p < 0.05, ^***p < 0.01.