«ABSTRACT We present evidence that ﬁnancing frictions adversely impact investment in workplace safety, with implications for worker welfare and ...»
These include maintaining existing equipment, replacing old and worn parts and machines, buying equipment with better safety features, and automating dangerous tasks. The physical assets involved can include both sophisticated machinery as well as simpler equipment. As an example of the latter, replacing steel cable used for hoisting objects with (more expensive) 6 synthetic ﬁber cable can reduce injury risk by decreasing recoil and the incidence of sharp edges upon breakage.
Firms also expend considerable resources on less tangible activities that impact safety, including work ﬂow organization, policies and procedures, training, and supervision. For example, lockout-tagout procedures prevent faulty machinery from being used until properly repaired.4 Alcoa introduced a forklift speed limit of four miles per hour on a production ﬂoor to reduce collisions (Clark and Margolis (2000)). While such a policy may seem mundane, the leading source of workplace injuries in 2012 was ﬂoors, walkways, and ground surfaces.5 Many plants establish safety committees to devise safety improvements. Perhaps the biggest innovation in safety management in the last few decades is the real-time, automated collection of data on a ﬁrm’s production processes, which expedites the mitigation of potential hazards.
Like investment in physical assets, these organizational and policy activities consume ﬁnancial resources. Allocating employee time to work on safety committees requires hiring more employees or paying overtime to maintain a given level of production. The same holds for training employees. Moreover, policies are only eﬀective if they are actively enforced.
Firms must therefore devote time to monitoring and auditing to ensure that employees follow prescribed practices. In addition, practices such as lockout-tagout for broken equipment may lengthen the time that productive equipment is out of operation.
Systematic estimates of the amount that companies spend on workplace safety are not available, as companies do not generally track such spending. However, anecdotes suggest that this spending can be substantial. Patterson-UTI Drilling Co., an oil and gas drilling company, estimates that it spent $150 million on training and safety improvements between 2001 and 2010, which amounts to 7% of its total income and 32% of its SG&A expense over the period.6 Following three fatal accidents at its Mission Valley Plant between 1971 and 1988, Alcoa spent $4 million making safety improvements in 1988 at that plant alone (Clark and Margolis (2000)). While these examples may not be representative, safety experts with 7 whom we spoke indicated that safety-related expenditures at their companies are substantial.
While safety-related activities are implemented at the establishment level, they are driven by ﬁrm-level decisions through budgetary and policy initiatives. An establishment may cut spending on safety in order to meet short-run budgeted cost targets. Safety practitioners with whom we spoke repeatedly mentioned that budget constraints were an important impediment to implementing workplace safety measures. Anecdotally, the Chemical Safety Board (CSB) blamed a catastrophic explosion at BP’s Texas City Reﬁnery in 2005 that killed 15 employees at least in part on an explicit decision not to replace a worn valve due to cost-cutting pressures.7 Firm-level policy initiatives include hiring safety consultants to help improve safety practices, setting safety targets and holding managers accountable for achieving them, and implementing a safety culture.8 A lack of ﬁnancial resources at the ﬁrm level can impact both tangible and intangible investments in safety at the establishment level. Improved safety generates returns to a ﬁrm over time in the form of reduced downtime, increased productivity, fewer lawsuits, and a lower compensating wage diﬀerential. However, a ﬁnancially constrained ﬁrm may turn down even positive NPV projects in order to conserve resources in the short run. The longrun nature of returns to investment in safety may make it especially vulnerable to cuts in the face of ﬁnancing constraints.9 In this sense, a high level of workplace safety may be a luxury that a resource-constrained ﬁrm cannot aﬀord. Moreover, serious workplace accidents are infrequent events, making the beneﬁts of spending to improve safety diﬃcult to quantify and hence justify to investors.10
B. Financing and Workplace Safety: the Identiﬁcation Challenge
The ideal experiment for studying the eﬀect of ﬁnancing constraints on workplace safety would involve taking two identical ﬁrms, randomly shocking one with additional ﬁnancial resources (e.g., cash or borrowing capacity), and then observing subsequent changes in injury 8 rates in both. If ﬁnancing constraints impede investment in safety, the injury rate should fall in the ﬁrm receiving the shock relative to the ﬁrm not receiving the shock. The main challenge in approximating this ideal experiment using actual data is that observed variation in ﬁnancial resources is not exogenous, and ﬁrms with diﬀering levels of resources are likely to also diﬀer along other dimensions, some of which are unobserved. This raises the concern that any observed relation between measures of ﬁnancial resources and injury rates could be driven by an omitted variable or reverse causality.
We consider four speciﬁc alternative mechanisms that could induce a negative relation between injury rates and measures of ﬁnancial resources, similar to the eﬀect that ﬁnancing constraints should produce. First, employees working with heavy manufacturing equipment may face an especially high risk of injury. This equipment also tends to make good collateral because of its redeployability, and thus ﬁrms using such equipment may borrow heavily. A large existing debt load could at least create the appearance that a ﬁrm’s additional ﬁnancing capacity is limited. Second, poor operational management can increase injury risk while also depleting ﬁnancial resources. Third, a fast-growing ﬁrm may experience temporarily high injury rates due to employee inexperience or excess workloads, and growth tends to consume ﬁnancial resources in the short run. The fourth mechanism is reverse causality. Costs associated with actual injuries deplete ﬁnancial resources.
Still other mechanisms could produce a positive relation between injury rates and ﬁnancial resources. For example, the existential threat created by a persistent lack of ﬁnancial resources may force a ﬁrm to operate with a high degree of eﬃciency, which could in turn lead to lower injury risk. This view is consistent with Jensen’s (1986) argument that debt promotes operational eﬃciency by disciplining management. In addition, expenditures on safety consume resources in the short run, which could induce a quasi-mechanical positive relation between measures of ﬁnancial resources and injury rates. These mechanisms would make the eﬀects of ﬁnancing constraints on injury rates more diﬃcult to detect.
Our data on workplace injuries come from the SOII of the BLS. Through a joint eﬀort with OSHA, the BLS gathers data for hundreds of thousands of establishments each year in a stratiﬁed sampling process to produce aggregate statistics on the state of occupational risk in various industries in the U.S. Employers covered under the Occupational Safety and Health Act and employers selected to be part of the BLS survey are required to maintain a log recording any injuries “that result in death, loss of consciousness, days away from work, restricted work activity or job transfer, or medical treatment beyond ﬁrst aid.” These employers must make their injury logs available to OSHA inspectors and supply the data contained in the log to the BLS.
Each establishment in the data has a unique identiﬁer. Each establishment-year record contains establishment name, location, SIC code, number of injuries (Injuries), number of injuries resulting in days away from work (DAF W Injuries), average number of employees (Employees), and total number of hours worked (Hours). We use these data to construct annual measures of the injury rate at each establishment. Our primary injury rate measure is Injuries/Hour, which is equal to Injuries divided by Hours. We also construct DAF W Injuries/Hour, which is equal to DAF W Injuries divided by Hours.11 We multiply both of these injury rate measures by 1,000 to make the numbers easier to write.
The BLS data also include, for the period 2002 to 2009, the employer identiﬁcation number (EIN) of the establishment’s parent company. We use the EIN to match the establishment-level data to ﬁrm-level data in Compustat. Thus, our sample period is 2002 10 to 2009. Each ﬁrm in Compustat can have multiple establishments.
We calculate several ﬁrm-level ﬁnancial variables using the Compustat data. The variable Debt/Assets is book debt (the sum of Compustat items dlc and dltt) divided by total assets (at). The variable Cash/Assets is total cash and equivalents (ceq) divided by total assets.
The variable CashF low/Assets is the sum of income before extraordinary items (ib) and deprecation and amortization (dp), divided by lagged total assets. Dividends/Assets is common dividends (dvc) divided by lagged total assets. The variable Log(Assets) is the natural log of total book assets.12 The variable AssetT urnover is total sales (sale) divided by lagged total book assets. The variable M arketT oBook is the market value of assets divided by total book assets, where market value is the sum of the market value of common equity (the product of shares outstanding, cshpri, and the ﬁrm’s stock price, prcc_f ), preferred stock (pstkl), and book debt, minus the book value of deferred taxes (txdb). We set the value of preferred stock or deferred taxes to zero if the relevant item is missing in Compustat. The variable T angibleAssetRatio is net property, plant, and equipment (ppent) divided by total book assets. Finally, the variable Capex/Assets is capital expenditures (capx) divided by lagged total book assets. We winsorize all of these variables at the 1st and 99th percentiles to reduce the possible inﬂuence of outliers, and lag the balance sheet variables by one year.
We exclude from our sample any observations for which any of the ﬁrm-level Compustat variables described above is missing. We also exclude all establishments belonging to ﬁnancial ﬁrms (SIC code 6000-6999) or regulated utilities (4900-4999) from our sample. This leaves us with a primary sample consisting of 43,721 establishment-year observations for 25,380 unique establishments that belong to 2,251 unique ﬁrms. The median number of times an establishment appears in the sample is one, reﬂecting the fact that most establishments are sampled by the BLS only once during the sample period. However, 7,918 establishments appear in the sample multiple times and are together associated with 25,053 establishment-year observations (3.16 per establishment). For this subsample, we can account for establishment ﬁxed eﬀects in our regression analysis.
Table II presents summary statistics for the sample. Panel A reports the number of establishment-level observations in the sample by year. The number of observations is fairly stable over time.
— Insert Table II here — Panel B presents establishment-level summary statistics calculated from the BLS data.
Consistent with the BLS’ conﬁdentiality policy, we report only means and standard deviations and do not report statistics such as medians and individual percentiles that would present data for individual establishments. The average establishment in our sample has 353 employees, though this number varies widely across the sample. The average employee works 1,718 hours a year, or approximately 43 40-hour work-weeks. The average injury rate is 4.13%, in line with an average annual injury rate of 4.55% over our sample period as reported by the BLS in its aggregate statistics. Slightly less than one in three injuries results in days away from work. Panel C presents ﬁrm-level summary statistics for our sample. The mean values of the variables are in line with those for Compustat ﬁrms as a whole.
Panel D presents correlations among the variables. Overall, these correlations are modest.
Only the correlation between Capex/Assets and T angibleAssetRatio exceeds 0.3 in absolute value. That this correlation is relatively high is unsurprising, as capital expenditures increment tangible assets by deﬁnition. The lack of strong correlations overall suggests that when we use these measures as explanatory variables in regressions, our estimates should not be excessively sensitive to small changes in the values of the variables.
An interesting and useful feature of the data is establishment-level rather than ﬁrmlevel identiﬁcation of industry. This identiﬁcation allows us to assign each establishment a 12 unique industry rather than pooling them over a coarse and potentially inapplicable ﬁrmlevel industry classiﬁcation. Table III reports the number of observations and injury rates (per hour worked and per average number of employees) for our sample across establishments in diﬀerent industries. We deﬁne industries using the 48-industry classiﬁcation of Fama and French (1997) and assign each establishment to one of these industries based on its SIC code as reported in the BLS data. Five industries with fewer than 25 observations in the sample each are omitted from Table III for the sake of brevity and because the relatively small number risks revealing the identity of individual establishments. Injury rates are highest in the Candy & Soda, Fabricated Products, and Transportation industries. Not surprisingly, they are lowest in primarily white collar industries such as Entertainment, Computers, and Trading.
Figure 1 plots the relative distribution of establishment size. As can be seen, establishment size varies widely from a minimum size of 10 full-time employees to well over 1,000 employees. Half of the establishments in the sample have fewer than 100 employees.
Figure 2 plots the distributions of injury counts and rates for the sample. Almost 30% of establishment-years in the sample have zero injuries. This is not surprising given the relatively low overall injury rate of 4.13% and the small size of many establishments.