«ABSTRACT We present evidence that ﬁnancing frictions adversely impact investment in workplace safety, with implications for worker welfare and ...»
17 One additional cautionary note in interpreting the results of the regressions in this section is that correlation among the explanatory variables combined with possible measurement error could produce biases.
In particular, Log(Employees) and Hours/Employee are subject to this concern, as average employment and hours worked are self-reported by the establishments in the BLS survey and, unlike the ﬁrm-level variables, are unaudited. As Table II shows, these variables have little correlation with most of the ﬁrm-level variables, especially those relating most directly to a ﬁrm’s ﬁnancial resources.
18 Note that the main eﬀects of Exposure and T reatment are fully absorbed by the establishment and interacted year ﬁxed eﬀects, respectively.
19 We do not run this test for the ﬁnancial crisis experiment, as exposure to treatment itself is based on a ﬁrm’s capital structure.
20 We choose a three-year window because it is long enough to reliably measure recent foreign proﬁtability while avoiding foreign proﬁts from the distant past that may no longer reside in a foreign subsidiary. Our results are robust to alternative windows for cumulating foreign proﬁts. Establishments for which P osF P = 0 include establishments of ﬁrms with foreign losses over the 2001 to 2003 period and those with no foreign subsidiaries, with approximately 95% being comprised of the latter. We obtain similar results if we exclude ﬁrms with foreign losses over the 2001 to 2003 period from our sample.
21 We obtain similar results if we extend the sample period back to 2005.
22 Approximately 80% of ﬁrms have 2007 ﬁscal year-ends between September 2007 and January 2008.
23 The conclusions of our analysis are unchanged if we use only the BLS data to classify ﬁrms in the oil business or if we exclude the ﬁnancial crisis period (see Appendix Table AI).
24 If multiple potential matches have the same propensity score, we randomly choose one.
25 We are not able to show the breakdown using more narrowly deﬁned industries because of disclosure 55 concerns. We also cannot show the industry breakdown for the oil price experiment because the small number of ﬁrms and establishments in each category raises disclosure concerns.
26 In our main analysis, we do not require matched establishments to be in the same industry as doing so would greatly limit the number of possible matches for many establishments, making it diﬃcult to match precisely on other observables. However, in Appendix Table AII, we show that the results are similar to those presented here if we do require within-industry matching.
27 One concern with the AJCA experiment is that the multinationals exposed to the AJCA shock might have invested repatriated cash disproportionately in safer activities, with more dangerous activities outsourced overseas. We would ideally like to measure the eﬀect of a cash windfall on an employee’s injury risk, holding ﬁxed the activities in which the employee is engaged. While we do not observe the mix of activities within an establishment, in Appendix B, we show that ﬁrms with foreign proﬁts do not shift employment towards establishments in safer industries after 2004. This ﬁnding provides some comfort that shifts in activities do not drive the results that follow.
28 This ﬁlter shrinks the usable sample of ﬁrms from 5,471 to 4,469. The results are similar if we do not exclude these establishments.
29 If a ﬁrm has an odd number of establishments in the data both before and after the AJCA, we discard the establishment with the median industry median injury rate.