Definitions: Significance Levels: *** 1%, ** 5%, * 10%. FTE is defined as 2,000 hours for yearly regressions and 500 hours for quarterly regressions.

IRR Estimates: The table reports IRR (incidence rate ratio) coefficients. A coefficient of 1 thus indicates no change at all in predicted injuries; coefficients between 0 and 1 represent a predicted fall in injuries (e.g. a coefficient of 0.97 represents a 3% decline); and coefficients greater than 1 represent predicted increases (e.g. a coefficient of 1.03 represents a 3% rise). All estimates are rounded to the nearest thousandth and all standard errors are rounded to the nearest hundredth, so an estimate displayed as 1.000 does not imply the value was exactly 1, and a standard error displayed as 0.000 does not imply it was exactly 0.

Models: All models are negative binomial regressions with hours worked in any subunit used as the exposure term; standard errors (shown in parentheses) are clustered at the mine level. The unit of analysis for all models is the mine-quarter.

Sample: The sample for all models is limited to underground bituminous coal mines and excludes all mine-quarters with zero production and/or hours worked and/or mine-years with a disaster. In addition: the sample for the public models is limited to the time period between 1993 and 2010 and the sample for confidential models is limited to the time period between 1998 and 2010.

Independent Variables: All models include the following regressors: All models include the following regressors: union dummy, union-size interaction term, mine size measure (defined as specified in column headers), logged controller size measure (defined as specified in column headers), mine age, productivity, number of lost-work injuries (in thousands – note that the tables in the paper use units in hundreds, but here we use thousands to make small coefficients more clear) in the previous four quarters (for quarterly regressions) or calendar year (for yearly regressions), total penalty points (in tens of thousands – note that the tables in the paper use units in thousands, but here we use tens of thousands to make small coefficients more clear) in the previous four quarters (for quarterly regressions) or calendar year (for yearly regressions), a constant term, dummies indicating presence of each type of mine subunit, quarter dummies, and district dummies. In addition: the FTE Public, Employees Public, and Tonnage Public models include the NIOSH longwall dummy and a constant term; and the FTE Confid., Employees Confid., and Tonnage Confid. models include number of coal beds, mean coal bed thickness (in yards), subsidiary indicator, captive production as a percentage of total production, recoverable coal reserves, a constant term, and mining method percentages. The analysis is limited to the underground subunit only for all variables with subunit-level data, including: hours worked (FTEs), employees, and coal production (at the mine and controller levels); previous injury counts; and productivity. Note that the subunit dummies, when present, merely indicate whether each type of subunit is active at that mine during that quarter/year; as such, any subunit dummy may equal 1 even when the analysis is limited to the underground subunit.

Dependent Variables: The total number of injuries per mine-quarter. Only injuries that occurred in the underground subunit are included.