«ABSTRACT We present evidence that ﬁnancing frictions adversely impact investment in workplace safety, with implications for worker welfare and ...»
Titman, Sheridan, 1984, The eﬀect of capital structure on a ﬁrmŠs liquidation decision, Journal of Financial Economics 13, 137-151.
Viscusi, W. Kip, and Joseph E. Aldy, 2003, The value of a statistical life: A critical review of market estimates throughout the world, Journal of Risk and Uncertainty 27, 5-76.
Wooldridge, Jeﬀrey, 2002, Econometric Analysis of Cross Section and Panel Data (Massachusetts Institute of Technology).
37 16,000 14,000 12,000
8,000 6,000 4,000 2,000
Figure 1. Distribution of establishment sizes.
This ﬁgure presents the distribution of establishments by number of employees or full-time equivalents for each establishment observation.
Figure 2. Distributions of injury rates and injury counts.
This ﬁgure presents histograms showing the distribution of Injuries/Employee (top portion of the ﬁgure) and number of injuries (bottom portion).
To avoid revealing information about speciﬁc establishments, the x-axis is left intentionally unlabeled.
39 0.0015 0.001
0 2001 2002 2003 2004 2005 2006 2007 2008 2009
Figure 3. Injury rates over time by foreign proﬁt status.
This ﬁgure shows the portion of injury rates not explained by other observable ﬁrm- and establishment-speciﬁc variables over time for the propensity score-matched sample of ﬁrms with (“treated”) and without (“control”) foreign proﬁts as of the AJCA in
2004. These unexplained injury rates are the residuals from an OLS regression of Injuries/Hour (times 1,000) on various ﬁrm and establishment characteristics. The solid line shows the mean unexplained injury rate for establishments belonging to ﬁrms reporting positive cumulative foreign proﬁts over the 2001 to 2003 period. The dotted line shows the mean unexplained injury rate for establishments belonging to ﬁrms reporting zero or negative cumulative foreign proﬁts over this period.
40 0.0007 0.0006
0.0004 0.0003 0.0002 0.0001 0 2005 2006 2007 2008 2009
Figure 4. Injury rates over time by debt maturity status This ﬁgure shows the portion of injury rates not explained by other observable ﬁrm- and establishment-speciﬁc variables over time for the propensity score-matched sample of ﬁrms with (“treated”) and without (“control”) a large quantity of debt maturing within one year as of ﬁscal year-end 2007.
These unexplained injury rates are the residuals from an OLS regression of Injuries/Hour (times 1,000) on various ﬁrm and establishment characteristics. A ﬁrm is deﬁned as having a large quantity of debt due within the next year if debt due within one year as of ﬁscal year-end 2007 as a percentage of assets exceeds the 75th percentile for the sample (3.064%). The solid line shows the mean unexplained injury rate for establishments belonging to ﬁrms with a large quantity of debt maturing within the next year. The dotted line shows the mean unexplained injury rate for establishments belonging to ﬁrms with little debt maturing within the next year.
42 Table II Summary Statistics This table presents summary statistics for the data used in this study. Panel A reports the number of establishment-year observations by year, where an establishment refers to a single location of a company as identiﬁed by the BLS. Panel B reports summary statistics for the 43,731 establishment-year observations that we study. HoursW orked is the number of hours worked by employees of an establishment during a year. AverageEmployment is the average number of employees working at an establishment during a year. Hours/Employee is the ratio of the two. Injuries is the number of recorded injuries for an establishment in a year. DAF W Injuries is the number of days-away-from-work injuries recorded for an establishment in a year. Each of these injury counts is also reported per hour worked and per average number of employees. The per hour rates are multiplied by 1,000 to make them easier to read. Panel C reports summary statistics for the parent-level ﬁrmyear observations in our sample. Debt/Assets is book debt divided by book assets. Cash/Assets is cash and equivalents divided by assets. CashF low/Assets is the sum of income before extraordinary items and depreciation, divided by lagged assets. Dividends/Assets is common dividends divided by lagged assets. Assets is total reported assets. AssetT urnover is sales divided by lagged assets. M arketT oBook is the ratio of the market value of equity to the book value of equity.
T angibleAssetRatio is net plant, property, and equipment divided by total assets. Capex/Assets is capital expenditures divided by lagged assets. Panel D reports correlations among the explanatory variables used in the paper.
44 Table IV Panel Variance Statistics This table presents a summary of the relative variation between and within establishment, ﬁrm, and industry groups. The ﬁrst two rows report the mean and standard deviation of the variable for the full sample. The second set of rows reports the standard deviation across diﬀerent establishments controlling for the time-series mean and within each establishment controlling for the establishment mean. The third set of rows reports the standard deviation between and within ﬁrms. The fourth set of rows reports the standard deviation between and within each of the 48 Fama-French industry categories. See Table II for deﬁnitions of the injury rate variables.
48 Table VII Comparison of Treated and Untreated Establishments This table details various ﬁrm and establishment characteristics of treated and untreated establishments in each of the three quasi-natural experiments. An establishment is in the treated group in the AJCA experiment if it reported positive cumulative foreign proﬁts over the three years prior to 2004 (the year of the AJCA), and in the untreated group otherwise. An establishment is in the treated group in the ﬁnancial crisis experiment if its parent ﬁrm had debt due in the next year as of ﬁscal year end 2007 in the top quartile in the sample, and zero otherwise An establishment is in the treated group in the oil experiment if its parent ﬁrm is in the oil production business and zero otherwise. Oil-producing establishments themselves are omitted from the sample. Panel A reports means for all treated and untreated establishments. Panel B reports means for treated establishments and matched control establishments. See Table II for variable deﬁnitions. ***, **, and * indicate that a characteristic diﬀers between treated and untreated establishments at the 1%, 5%, and 10% level, respectively, based on a t-test. Panel C reports the breakdown of treated and control establishments by broad industry category. The industry breakdown for the oil price experiment is excluded because of disclosure concerns.
49 Table VIII Workplace injuries and Cash Flow Shocks This table presents estimates of the eﬀects of cash ﬂow shocks arising from three quasi-natural experiments on injury rates. Panel A reports estimates from OLS models where the dependent variable is Injuries/Hour. Panel B reports estimates from Poisson models where the dependent variable is Injuries and the exposure variable is HoursW orked. Panels C and D report estimates from analogous regressions, where DAF W Injuries/Hour and DAF W Injuries are the dependent variables. All models include establishment, industry-year, and state-year ﬁxed eﬀects. In the AJCA experiment, the sample is restricted to the years 2002, 2003, 2005, and 2006. T reatment is one post-2004 and zero pre-2004, and Exposure is one if the parent ﬁrm’s cumulative reported foreign proﬁts in 2001 to 2003 were positive and zero otherwise. In the ﬁnancial crisis experiment, the sample is restricted to the 2006 to 2008 period. T reatment equals one in 2008 and zero in 2006 and 2007, and Exposure equals one if the parent ﬁrm’s debt maturing within one year as a percentage of assets of ﬁscal year-end 2007 exceeds 0.03064 (the 75th percentile for the sample). The oil price experiment uses all years in the sample (2002 to 2009). The sample consists of only non-oil producing establishments. T reatment is equal to the natural log of the average oil price for the year, and Exposure equals one if an establishment’s parent ﬁrm is in the oil business (either has an oil-producing establishment in the BLS data in any year in the sample or is identiﬁed by Capital IQ as being in the oil, gas, and consumable fuels business, excluding coal mining) and zero otherwise.
In each experiment, treated establishments (Exposure = 1) are matched with untreated establishments (Exposure = 0) using propensity score matching. See Table VII for information about the characteristics of treated and untreated establishments in each matched sample. Control variables Debt/Assets, Cash/Assets, Log(Assets), M arketT oBook, and T angibleAssetRatio, CashF low/Assets, Dividends/Assets, AssetT urnover, Capex/Assets, Log(Employees), and Hours/Employee are included in the regressions but omitted from the table for brevity. See Table II for variable deﬁnitions. Standard errors clustered at the ﬁrm level are reported in parentheses below each point estimate. ***, **, and * indicate statistical signiﬁcance at the 1%, 5%, and 10% level, respectively, based on a two-tailed t-test.
51 Table IX
Workplace Injuries and Debt Maturity During the Financial Crisis:
Characteristic-by-Characteristic Matching This table presents estimates of the eﬀect of having a large quantity of debt maturing at the onset of the ﬁnancial crisis on injury rates using a series of samples matched on individual ﬁrm or establishment characteristics. The ﬁrst column identiﬁes the matching characteristic. The second shows the mean value of the characteristic for ﬁrms in the treated group. The third shows the mean value of the characteristic for ﬁrms in the untreated group. The third column shows the t-statistic for the diﬀerence in means. The fourth and ﬁfth columns show the coeﬃcient on T reatment ∗ Expsoure and its standard error clustered at the ﬁrm level from the OLS ﬁxed eﬀects regression in Table VIII, Panel A, column (2). Injury rate trend is the annualized pre-2008 change in the portion of an establishment’s injury rate not explained by other ﬁrm and establishment characteristics.
See Table II for deﬁnitions of the other variables. ***, **, and * indicate statistical signiﬁcance at the 1%, 5%, and 10% level, respectively, based on a two-tailed z-test.
53 Notes 1 Dharmapala, Foley, and Forbes (2011) and Faulkender and Petersen (2012) examine the eﬀect of the repatriation tax holiday, Almeida et al. (2012) study the eﬀect of the onset of the ﬁnancial crisis, and Lamont (1997) studies the eﬀect of an oil price shock in 1985.
2 Danna and Griﬃn (1999) argue that these costs are likely to be greater than those due to compensating wage diﬀerentials.
3 In other related papers, Bae, Kang, and Wang (2011) ﬁnd that ﬁrms with more debt score lower on a third party rating of employee friendliness, and Brown and Matsa (2013) show that ﬁrms in ﬁnancial distress have fewer and lower quality job applicants.
4 Lockout procedures involve isolating and disabling power sources in dangerous machinery in a systematic, step-by-step way. Tagout procedures ensure that only speciﬁc employees can unlock and untag a machine, ensuring that malfunctioning equipment is not accidentally brought back online before it is repaired.
5 Source: http://stats.bls.gov/news.release/archives/osh2_11262013.pdf.
6 Source: http://www.mysanantonio.com/news/energy/article/Eagle-Ford-pay-is-high-but-work-can-be-fatalphp.
7 Source: http://www.csb.gov/assets/1/19/CSBFinalReportBP.pdf.
8 See DuPont case study on Norfolk Southern: http://www2.dupont.com/Sustainable_Solutions/en_US/ assets/downloads/case_studies/NorfolkSouthern_CaseStudy.pdf.
9 While regulatory safety inspections and penalties could force ﬁrms to bear more of the cost of workplace hazards in the short run, OSHA and its state aﬃliates inspected less than 1.2% of worksites in the U.S. in 2012, according to OSHA’s website. Firms with high safety standards may also cut spending on safety in ways that do not trigger formal violations of safety rules when ﬁnancially constrained.
10 See http://blogs.hbr.org/2010/06/the-safety-calculus-after-bp/ for a discussion of this last issue.
11 We obtain similar results throughout if we compute injury rates per employee rather than per hour worked, but overall exposure to injury risk is ultimately a function of the number of hours that employees spend working.
12 We obtain similar results throughout if we measure ﬁrm size using total sales or total employees.
13 As in an OLS model, the ﬁxed eﬀects allow each unit of observation to have a diﬀerent baseline-level injury rate.
14 The economic magnitude of a coeﬃcient from a Poisson model can be assessed by examining the incident rate ratio associated with the coeﬃcient, eβ − 1. This represents the expected percentage point change in 54 injury count per unit change in an explanatory variable, and is 0.422 for the Debt/Assets coeﬃcient. A one-standard-deviation increase in Debt/Assets (0.219) then is associated with an 9.2 percentage point increase in expected injuries in the following year, somewhat larger than the association implied by the OLS coeﬃcients in Table V.
15 The negative binomial model produces an α parameter estimate of 0.729, which is statistically diﬀerent than zero at the 1% level, suggesting that the Poisson model’s assumption of equal mean and variance is likely violated. Violation of this assumption does not bias model Poisson estimates, but does reduce their eﬃciency (Wooldridge (2002), ch. 19).
16 Note that the additional data requirement lowers the number of usable observations.