The Effects of Import Competition on Worker Health *

Transcription

1 The Effects of Import Competition on Worker Health * T. Clay McManus University of Tennessee Georg Schaur University of Tennessee Abstract Occupational safety and health is an important determinant of workers welfare. Existing theory predicts that trade liberalizations increase rents for the largest firms in the market. If firms share rents with workers in the way of a safer work environment, then trade liberalization leads to a safer work environment. In the transition, the smallest firms in the industry shut down. Our theoretical model predicts that firms facing greater shut down risk reallocate resources to improve productivity at the expense of safety. Therefore, at small firms, facing the greatest shutdown risk due to import competition, safety conditions worsen but firm productivity increases. We provide empirical evidence that the growth in Chinese imports in the years significantly increased injury rates in the competing US manufacturing industries over the short to medium run, particularly at smaller establishments. This is relevant for policy and welfare considerations. Even if trade improves safety in the long run, workers at marginal firms face the costs of a riskier work environment in the interim while their firm struggles to survive. Enforcement of labor standards and regulation could improve worker conditions and even facilitate the adjustment process by leading less productive firms to drop out. JEL: F16, F66, J81, J32, L60 Keywords: Import Competition, Safety, Injuries, Trade Liberalization September 2014 PRELIMINARY AND INCOMPLETE Please do not cite without permission. * We would like to thank William S. Neilson, Marc-Andreas Muendler, J. Scott Holladay, Scott Gilpatric, Matthew C. Harris, and John Bell for helpful comments, along with seminar participants at the University of Tennessee. We would also like to thank the Occupational Safety and Health Administration and the following researchers for making their data publicly available: Autor, Dorn and Hanson (2013), Pierce and Schott (2013), and Feenstra, Romalis, and Schott (2002). Department of Economics, University of Tennessee, Knoxville, TN 37919; tmcmanus@vols.utk.edu Department of Economics, University of Tennessee, Knoxville, TN 37919; ; gschaur@utk.edu

2 McManus and Schaur 1 1. Introduction Evidence from a growing subfield of international economics finds that changes in the trade environment have important consequences for labor market outcomes, most notably wages and employment, and that these are important channels through which trade affects welfare. In this paper, we consider how trade through firm and worker behavior and competitive forces in the labor market affects worker health. Occupational safety and worker health outcomes, like wages, are important determinants of worker utility and firm profits that may similarly be affected by trade exposure, positively or negatively. To the extent that they do occur, adjustments along the health margin have substantial consequences for firm profits and worker utility that are not captured by traditional welfare analysis using wages and employment effects alone: there were an estimated 9 million job-related injuries and illnesses in the US in 2007 with total costs of approximately $250B in medical expenses and productivity losses (Leigh 2011). Existing and new theories deliver ambiguous predictions for the relationship of trade and worker health. The pro-competitive effects of trade will improve safety if they encourage firms to make technology investments to improve productivity that also improve worker health. Firms may also push managers to be more attentive to worker safety as a means of reducing productivity losses from injuries. Conversely, greater competition may hurt workers when safety is at odds with productivity and firms, with their backs against the wall, prioritize productivity at the expense of health when it comes to survival in a more competitive market. We develop a theoretical model for firm behavior that links safety decisions at the firm to market conditions and the firm s future prospects. Firm managers face a tradeoff in allocating constrained resources, e.g. their time, between safety and productivity. Injuries today bear costs to firms in the future if they are operating, and a reduction in their survival rate lowers the

3 McManus and Schaur 2 expected total cost of injuries today and induces a shift away from worker health and towards improving productivity. 4 Based on the theory and in light of the robust finding that import competition reduces firm survival rates (Bernard et al., 2006a; Pierce and Schott, 2013), particularly for the least productive firms (Bernard et al. 2006b), we hypothesize that worker health will worsen in the short run following import exposure at the marginal firms which are at greater risk of being forced from the market in the shift to a new long run equilibrium. We take this prediction on injuries to data on the growth in Chinese exports to the US in the late 1990s and 2000s, modeling our identification strategy after related work by Autor et al. (2013) and Pierce and Schott (2013). Using data from the Occupational Safety and Health Administration (OSHA) on injuries and illnesses at US manufacturing plants from , we find a significant positive relationship between import exposure and injury and illness rates at competing manufacturers. We estimate that, ceteris paribus, the mean 5-year growth in Chinese imports in an industry over this period (130%) increased injury rates at the average US establishment in the industry by 10-11% over the same interval. We find evidence the effects are heterogeneous within an industry and greatest for smaller firms, theoretically the least productive (Melitz, 2003). We estimate that the mean import shock over the 5-year interval increased injury rates by as much as 14% at the smallest decile of plants, about 10% at the median, and 8% at the seventy-fifth percentile. Injury rates were significantly increasing in the level of import competition for all but the largest decile of plants in an industry. Our empirical exercise is most closely related to recent work by Autor et al. (2012) and Pierce and Schott (2013), which both study the consequences of product market competition on labor market outcomes. Autor et al. (2013) define labor markets geographically and find that 4 This predicted relationship similarly arises in the alternative case where safety improvements are supposed to be long-term investments which yield benefits in future periods in which the firm operates.

4 McManus and Schaur 3 growth in import exposure in a region increased unemployment rates, reduced labor force participation rates and reduced average weekly wages. Pierce and Schott (2013) exploit crossindustry variation in the effects of a 2001 change in US tariff policy towards China and find the policy significantly increased imports and reduced employment in US manufacturing industries. Our finding that greater import competition leads to higher injury rates at competing firms supplements their results on adverse wage and employment effects. The theoretical model explains firm behavior in the short run, congruent with the limited data window for our empirical analysis, and is therefore complementary to existing trade models which evaluate the long run effects of trade liberalization on product and labor market outcomes by comparing the new steady state equilibrium to the old. We instead consider the transition to the new steady state that follows an import shock, and particularly the consequences of this adjustment process for outcomes at the marginal firms. Those firms least likely to survive in the new long run equilibrium are implicitly the ones most affected in the short run. We find that injury and illness rates are lowest at larger firms in an industry, and further that injury rates at larger firms are better insulated against import competition. In the long run, the exit of low productivity firms and labor reallocation to high productivity firms will therefore tend improve the average level of safety provided workers in the market. This study connects to broader research on the interrelation between international trade and labor standards 5, and in particular the subset of that work that focusses on labor standards outcomes as a consequence of international trade. Considerable theoretical attention has been given to child labor, in particular, though empirical support is sparse, in part because of poor 5 See Brown, Deardorff and Stern (1996) for a comprehensive theoretical treatment of labor standards and trade, and Brown (2007) for a broad review of the related literature.

5 McManus and Schaur 4 quality data. 6 We make use of detailed longitudinal firm-level data to contribute literature novel evidence linking trade liberalizations to working conditions in the labor market in a developed country. Our results suggest that trade has implications for labor standards that reach well beyond the developing world, the focus of most related literature. To our knowledge, we are the first to empirically study the relationship between occupational safety and health and the pro-competitive effects of trade liberalizations. Our results suggest that OSH management practices should consider import shocks, and product market competition broadly, when designing and overseeing regulations. From a trade policy perspective, our findings suggest that health is an important channel through which trade liberalizations affect welfare, even within a developed economy with well-functioning legal and regulatory institutions. The remainder of the paper is organized as follows: in Section 2, we discuss related trade theory in light of its potential implications for safety; in Section 3 we develop a theoretical model relating injury rates to firm survival; we describe our empirical strategy, data, and measurement in Section 4; Section 5 presents our primary empirical results; Section 6 concludes. 2. Insight from Existing Trade Models 2.1 Welfare Effects of Bilateral and Unilateral Trade Liberalizations In standard trade models with heterogeneous firms and intraindustry trade, including Melitz (2003) bilateral trade liberalizations or productivity gains unambiguously increase real wages and improve welfare. The predictions are less clear, however, for unilateral liberalizations or asymmetric productivity gains. In Demidova (2008), the existence of a homogeneous good 6 See also Brown, Deardorff and Stern (2002) for a survey of empirical results. Edmonds and Pavcnik (2006), for example, find a positive but insignificant relationship between trade exposure and child labor.

6 McManus and Schaur 5 sector facilitates a home market effect in specialization patterns through which productivity gains abroad hurt real wages (welfare) at home. Conversely, a technology shock abroad unambiguously raises welfare at home in Demidova and Rodriguez-Clare (2013), where wage differences across countries are pinned down by a trade balance condition rather than an outside good. 7 Through the lens of a hedonic wage model, where the returns to employment are jointly defined by both the monetary wage and the level of job-related safety and each is determined competitively in the labor market, the theoretical ambiguity in the long run wage effects of a unilateral trade liberalization similarly extends to safety. 2.2 Non-monetary Dimensions of Employment in Trade Models Efficiency wage models provide a framework for considering non-wage characteristics of employment: firm output depends on the effort exerted by workers, who get disutility from exerting effort that jointly with wages defines their returns in the labor market. Trade models incorporating these mechanisms give some precedence for our considering a broader notion of compensation. They also feature richer predictions for the labor market effects of trade than simpler constructs allow. They provide an explanation for unequal worker outcomes in an equilibrium that features unemployment, and link this heterogeneity to firm characteristics such as profits and its ability to monitor worker output. In so doing, this class of models explains the relevance and determination of a distasteful non-wage component of work and the channels through which trade affects workers in this environment. Like effort, occupational injury risk is conceivably a bad for workers that, jointly with wages, defines the utility workers receive from employment. 7 Both specify a CES demand structure that implies optimal markups. Demidova (2014) similarly replicates the comparison between the two models under variable markups.

7 McManus and Schaur 6 Workers in fair wage models are presumed to base their effort decision on the perceived fairness of the wage they are paid and receive, less they shirk (Akerlof, 1982). The notion of the fair wage is generally modeled as depending partly on the employer s characteristics, like productivity or profitability, and partly on an external reference point determined by conditions in the labor market like the average wage and the unemployment rate. Workers demand higher wages from more profitable firms and higher wages when the average market wage is greater. Following the rationale above, we might consider the implications of this mechanism when applied to the firm s provision of occupational safety and health: workers at more profitable firms expect safer conditions and similarly demand greater safety when the average provision in the market is greater. Amiti and Davis (2011) incorporate a fair-wage mechanism into a trade model with heterogeneous firms with the assumption that firm wages are increasing in its profitability. 8 They predict that bilateral tariff reductions (on final goods) affect the distribution of wages through changes in the distribution of firm profits: high-productivity firms export more and earn greater profits while surviving low-productivity firms are only adversely affected in the way of greater domestic competition and earn less, and therefore import competition increases wages for some workers while lowering wages for others. They find support for these predictions using data on Indonesian manufacturers. If occupational safety is similarly increasing in the profitability of the firm, their findings would suggest that a bilateral trade liberalization has a heterogeneous effect on worker health where jobs at productive, high-profit firms become safer while job safety at low-profit firms suffers. 8 Egger and Kreickemeier (2009) similarly consider fair wages in a trade model, but with a more explicit wagedetermination mechanism that relies on a weighted average of internal and external reference points such as described above. They similarly predict heterogeneous wage effects of bilateral liberalizations such that the average wage increases but so does wage inequality and involuntary unemployment.

8 McManus and Schaur 7 In shirking models, firms offer workers a wage-effort contract but imperfectly monitor the employee s choice of effort. Each firm offers a package of wages and effort that make the job valuable enough that workers do not shirk 9, given the idiosyncratic risk of being caught and the relative cost of losing the job. The focus in these models, with rare exception, is on wages and the effort choice is consequently simplified: all firms require the same level of effort from their workers, but must pay heterogeneous wages to elicit that effort because firms differ with in their ability to monitor workers. The firm-specific wage is: (a) increasing in the returns to the currently unemployed, (b) increasing in the probability of the firm shutting down, and (c) decreasing in the firm s monitoring ability. A straightforward way to think about injuries in efficiency wage models is as a byproduct of effort, supposing a worker s exposure to injury risk at work is increasing in his effort exertion and this is one source of his distaste for it. In this case, we can think of effort in these models as a proxy for injury risk and interpret the predicted consequences of trade accordingly. Rather than consider optimal effort fixed and the firms as offering the lowest wage such that workers exert effort, we can conversely think of wages as sticky and firms demanding the highest level of effort such that workers will assent at the given wage. Reinterpreted, these models explain firm heterogeneity in the effort required of workers (and therefore injury risk), instead of wages as in the standard case. In standard shirking models, the no-shirking condition (NSC) pins down wages as a function of the exogenous level of effort demanded:. For exposition along the line of thinking above, consider instead. Rearranging the NSC 10 shows that the 9 In the worker s problem, the set of optimal decisions reduces to either zero effort ( shirking ) or the exact the amount demanded by the firm even when the choice space is continuous; effort is costly and exerting partial effort doesn t reduce the chances of being fired, and anything above the required amount is superfluous. 10 These comparative statics hold both the standard specification used in the seminal work by Shapiro and Stiglitz (1984), who use, and the specification in Davis and Harrigan (2011),. The NSCs imply,

9 McManus and Schaur 8 level of effort a firm can elicit at the given wage is: (a) decreasing in the value of being unemployed, (b) decreasing in the probability of the firm shutting down, and (c) increasing in the firm s ability to monitor worker effort. Davis and Harrigan (2011) study the effects of trade liberalizations on a labor market with efficiency wages. Their predictions for wages inform our discussion of injuries in thinking of a broader notion of returns to employment that encompasses health; conceptually, predicted wage gains (losses) are be realized as workers receiving better (worse) health in lieu of monetary wages. More explicitly, the discussion above provides a link for translating wage effects to predictions for injuries in an environment where wages are sticky and adjustments instead occur along the effort margin: the NSC pins down an inverse relationship between wages and effort such that predicted wage gains in the standard model where effort is fixed are synonymous with gains in the way of lower effort when wages are instead fixed. Per both these reasons, channels through which trade affects worker wages in their model similarly explain like changes in health outcomes here. Specifically, Davis and Harrigan (2011) augment a Melitz-model with shirking wages, which are linked to the product market in that firm marginal cost depends on its monitoring technology as well as its productivity. Relatively high wages define a good job from a worker s standpoint but also characterize a firm with high marginal costs, ceteris paribus. Trade liberalization in the model destroys some good jobs as these firms are forced from the market 11, but the change in average returns to employment is theoretically ambiguous and depends partly on the correlation between firm productivity and monitoring. Trade increases average nominal respectively, and ( ), where is the real wage, the discount rate, the value of being currently unemployed, the firm s monitoring rate, and the probability of firm shutdown. 11 This point is most clearly illustrated by considering two firms with equal productivity. The firm that pays higher wages (due to inferior monitoring) is less profitable and therefore most likely to exit the market.

10 McManus and Schaur 9 wages if the highly productive firms which gain market share are also likely poor monitors and pay higher wages as a result. 12 Their simulations suggest bilateral trade liberalization destroys as many as one-quarter of good jobs, but increases average returns to employment. 13 In the context of effort and health, good jobs are those that push workers less, expose them to fewer risks, and have better health outcomes because they cannot monitor worker output as well and so must offer rents of these sorts to prevent shirking. Adapting Davis and Harrigan s framework to this alternative environment where wages are fixed and firms adjust effort unsurprisingly yields inverse predictions for effort: pressures which otherwise push up wages instead push down effort, and vice versa. To the extent effort intensity precipitates poor health outcomes, some workers health suffers as they are displaced from safer jobs (at firms with relatively higher marginal costs) but the net change in average outcomes depends on the relationship between health and firm productivity. It is important to note that these models do not explain firm-level changes in wages or effort, for that matter due to trade liberalization insofar as the exogenous shutdown rate and discount rate are unchanged; instead, the mass of incumbent firms is replaced by new entrants with an alternative distribution that define the new equilibrium and worker outcomes change accordingly. 2.3 Trade, Technology, and Productivity Gains There is considerable evidence that factor productivity increases at surviving firms in response to trade competition. 14 Bloom et al. (2012), for example, find that the growth in Chinese import 12 This relationship also determines the extent of good job destruction, as a negative relationship implies that high wages are partly offset by high productivity and thus good jobs are more insulated. 13 This, along with lower prices, leads to aggregate welfare gains. Their simulations suggest little change in unemployment following a bilateral trade liberalization.

11 McManus and Schaur 10 competition over the same interval we study significantly increased the average total factor productivity among European firms in affected industries as a result of both low productivityfirm exit and intra-firm improvements. The sources of observed productivity gains are difficult to identify, but studies have presented various explanations that include, in support of our theoretical approach, manager adjustments to improve factor productivity and increase firm survival (Bloom and van Reenen, 2007). Below, we develop a model that explains the intensity of effort as one channel for productivity gains, which comes at the expense of worker health. Lazear et al. (2013) finds evidence that worker effort adjustments drive productivity increases: they find that the firm s realized productivity gains during a recession are the result of making do with less, getting more effort from fewer workers. 15 Theoretical models presented in Bustos (2011) and Ederington and McCalman (2013) explain that trade affects firm decisions to upgrade technology. This is supported empirically by Bustos analysis of Argentinian firm responses to MERCOSUR, and also the finding in Blundell et al. (1999) that product market competition increases innovation and patent filings among affected firms. The pro-competitive effects of trade will benefit worker health through this channel to the extent that these technology investments correlate with safer labor tasks. 3. Theoretical Model In this section, we develop a model that highlights a mechanism through which import competition affects injury rates at marginal firms facing greater risk of being forced from the 14 See Syverson (2011) for a good discussion of the effects of import competition on productivity in theory and empirical findings. Studies with similar findings cited there are: Pavcnik (2002), Eslava et al. (2004), Muendler (2004), Bernard et al. (2006), Fernandes (2007) and Verhoogen (2008). 15 Their model differs from ours in that the adjustment in their model is driven by workers who exert more effort to reduce the probability of job loss when unemployment increases as job loss is more costly. To the extent that import competition similarly deteriorates the value of being currently unemployed for a manufacturing worker who might remain unemployed for longer or be forced to switch to a lower-wage occupation (as found in Ebenstein et al., 2014) workers may increase their effort and be willing to accept worse safety outcomes following import competition.

12 McManus and Schaur 11 market following a trade liberalization. These firms with their backs against the wall are arguably the most affected by trade in the short run, but also likely absent in the new long run equilibrium. We provide a straightforward model for this short run adjustment process that yields a theoretical prediction connecting firm survival to workplace safety. 3.1 Technology The firm relies on worker effort to produce output. The plant employs homogeneous workers and elicits from each an amount of effort. We assume that the firm has constant returns to scale in total effort such that it has the production function where parameterizes the firm s productivity. It is costly for workers to exert effort, and so to elicit effort the firm must allocate resources to monitoring their output. For example, the manager at the firm might increase worker effort by spending his endowment of time monitoring worker output. Worker effort as a function of the wage paid and the amount of monitoring is given by where such that effort is increasing and concave in both the wage and the level of monitoring. The cost of monitoring is developed below. 3.2 Accidents and Injury Costs As a consequence of their job, workers at the firm face a risk of work-related injury or illness each period. These injuries are costly to the firm. In practice, firms pay substantial direct costs for injuries and illnesses that include workers compensation pay, medical expenses, and legal fees associated with processing and settling claims. For firms with experience-rated workers compensation insurance, high injury rates increase the cost of future coverage. Firms further incur equally substantial fines and other indirect costs resulting from workplace accidents that are not covered by insurance and include: disruptions in production, damage to capital, hiring

13 McManus and Schaur 12 and training costs for replacement workers, demand penalties from bad publicity, and diminished productivity resulting from employing a debilitated worker or a less-capable replacement and from lower employee morale. 16 Behaviorally, injuries bring disutility to a benign decision maker with other-regarding concerns for worker health. To simplify the discussion, we henceforth use the term injury to broadly describe stochastic, work-related events which negatively affect worker health. We model injury costs in reduced form with the assumption that consistent with many of the costs mentioned in the discussion above the firm incurs some costs in the future as a result of injuries today. As in Melitz (2003), we assume that in each period there is an exogenous probability that the firm is forced to shut down. The expected total cost of an injury today is given by: [ ] where is the cost incurred periods after an injury and the interest rate is used as the firm s discount rate. We assume that injury costs depreciate over time at a rate of such that today is given by: 17. It follows that, in expectation, the total cost over the firm s lifetime of an accident [ ] Importantly, the cost of injuries is such that the firm is in effect judgment proof : the costs are only incurred in future periods if the firm is still operating and therefore the expected 16 In the fair-wage literature, worker effort depends on how they are treated by the firm in the form of the wages they are paid. Analogously, worker effort might depend on how they are treated by the firm with regards to the workplace environment and safety. 17 This is derived in the following manner: Let be the cost of an injury as of time for all future time periods. For periods of duration,. For, we have. Taking the limit as yields by L Hopital s Rule.

14 McManus and Schaur 13 lifetime costs to the firm of worker accidents is decreasing in the probability of the firm shutting down. The specification that consequences of worker injuries are limited to the firm s survival is consistent with: financial liability for direct costs is limited for a corporation s decision makers following a shutdown; indirect costs such as diminished worker productivity, higher workers compensation insurance premiums, and demand penalties are all inconsequential if the firm shuts down; behaviorally, the manager s concern for workers well-being extends only so far as his tenure with the firm. We simplify the model by assuming that wages are unaffected by injury risk in the short run, either because workers have asymmetric information and do not observe changes in risk or because they are just paid their compensating variation in the event of an injury and are therefore indifferent. We show in Appendix B that the model s predictions are consistent under an alternative specification that allows for compensating wage differentials. 3.3 Safety Investments and Accident Rates In addition to monitoring effort, the firm chooses to expend an amount of its resources (e.g. manager s time) to improve workplace safety and lower the probability of injury. We assume that the probability a worker is injured is given by ( ) for [ ], where such that the returns to safety lower injury rates are non-increasing. For simplicity, we assume that the firm is constrained in their endowed resources and that this constraint binds: ; resources spent monitoring effort are implicitly resources not spent improving safety. The probability of worker injury as a function of effort monitoring therefore has the distribution:

15 McManus and Schaur 14 {( ) } where and. 18 We normalize, so can be thought of as the share of resources the firm allocates to improving safety. We assume that likelihood of a particular worker getting injured is independent of the number of workers at the plant. The expected total cost of injuries in a period is therefore given by: 3.4 Demand We suppose the monopolist operates in a simple economy where the representative agent has quasi-linear preferences over a homogeneous good produced in a competitive market and the monopolist s good : where parameterizes elasticity of substitution. The competitive good is chosen as the numeraire and its price is normalized to one. We assume that the consumer s endowment is such that he consumes some positive amount of both goods and therefore the monopolist faces the inverse demand curve: 18 The positive relationship between accident rates and productivity (through elicited effort) in our model arises endogenously based on the assumption that the firm is constrained in its resources and effectively decides to what degree to push output and to what degree to push effort. We prefer this to the alternative approach of modeling the number of accidents as a direct function of worker effort such that accidents are a byproduct of higher intensity effort, though such a relationship would yield predictions to those in our model.

16 McManus and Schaur 15 where is the constant elasticity of demand. 3.5 Equilibrium Effort and Safety The firm s survival probability is taken as exogenous and therefore choices today affect the future only insofar as injuries today are costly in future periods. The injury cost function captures these dynamics and therefore the firm faces the static problem: ( ) The key features of the model are that the firm faces a trade-off between improving productivity and protecting workers against injuries, and the relative incentives of each depend on the firm s strength in the market as the expected cost of an injury is decreasing in the likelihood of the firm shutting down. We focus our analysis on worker health and therefore simplify the problem by treating wages as fixed and normalizing them to one. 19 The optimal level of effort monitoring for the firm is given by the interior solution [( ) ( )] if, which we henceforth assume to be true. The equilibrium amount of effort exerted by employees at the firm and the probability of their being injured are therefore: [( ) ( )] 19 Footnote 8 highlights theoretical and empirical support for our assumption that wages are sticky in the short to medium run.

17 McManus and Schaur 16 ( ) ( ) The optimal amount of labor employed by the firm is given by: [ ( ) ( ) ( ) ] 3.6 The Injury Effects of Greater Shutdown Risk In this section, we use comparative static analysis to predict changes in the variables of interest in response to an increase in the firm s probability of shutting down. We use the functional specifications and closed-form solutions introduced above to present our results, and show that the predictions hold for a more general model in Appendix A. Proposition 1a: Firms monitor worker effort more when the probability of shutting down is greater; Proposition 1b: Firms elicit greater effort from workers when there is greater shutdown risk. Proof: ( ) [( ) ( )] ( ). ( ) [( ) ( )] ( ). The firm s level of monitoring is increasing in the shutdown rate because it lowers the marginal cost of eliciting greater effort: the higher injury rates that accompany more monitoring are less costly because the firm s expected total losses from an accident are decreasing in its probability of shutting down. A fall in the relative cost induces the firm to expend more resources monitoring worker output and therefore workers exert greater effort on the job. Proposition 2: Injury rates are increasing in the probability of the firm shutting down.

18 McManus and Schaur 17 Proof: ( ) ( ) ( ). Investments to improve workplace safety are less valuable on the margin following an increase in the shutdown rate because it reduces the expected total cost of injuries. The firm responds by substituting resources away from safety and towards improving productivity, increasing injury rates and harming worker health in the process. 20 Proposition 3: An exogenous increase in the probability of shutdown results in an increase in the productivity of labor. Proof: and therefore, which is shown to be positive in the proof for Proposition 1a. The firm becomes more productive (i.e. competitive) in a given period due to a lower survival rate as it pushes workers harder at the expense of their health. 21 In our model, labor used today unambiguously increases in response to lower survival rates. Greater competition in this short run environment where wages are sticky and demand is fixed only affects today s profits insofar as it decreases the firm s survival rate and therefore reduces the cost of effort. The resulting increase in effort demanded by the firm makes each unit of labor more productive. 4. Empirical Strategy Low-wage country import competition in an industry has been robustly found to reduce firm survival and employment (Bernard et al. 2006a; Pierce and Schott, 2013) particularly for the 20 Intuitively, the negative relationship between firm survival and accidents would similarly arise in an alternative specification where the firm s investments in reducing accidents were considered directly if those investments (e.g. replacing old machinery, updating and improving safety equipment, and safety training for workers) benefitted the firm in future periods; in that case, the expected benefits of investing in safety would be decreasing in the firm s shutdown probability, complementary to our specification that the same relationship arises from the cost side. 21 Our model s predictions explain the productivity increases found in empirical exercises as partly the result of greater worker effort, which here comes at the expense of safety. If effort is not modeled or observed as an input to production so output is instead specified as a function of labor only:. The estimate for the firm s total factor productivity in the production function depends on both the firm s randomly drawn productivity parameter and by its endogenously determined level of worker productivity.

19 McManus and Schaur 18 least productive firms (Bernard et al. 2006b) and our theoretical model predicts that this will be to the detriment of safety at the affected firms. We use trade data to test for the link between trade liberalization and worker health. Specifically, we study the effects of Chinese import competition on the US manufacturing industry using a measure of import competition adapted from Autor et al. (2013) and firm-level data on worker injuries and illnesses. The data, empirical model, and identification strategy are presented below. 4.1 Import Data and Measurement We use data on industry-level trade compiled by Autor et al. (2013) and described in more detail there. 22 Briefly, they concord bilateral product-level import data from the UN Comtrade Database to the manufacturing industries that produce them, making use of the crosswalk in Pierce and Schott (2009). The real value of imports each year is allocated across 397 manufacturing industries defined at the four-digit Standard Industrial Classification level (SIC4). 4.2 Establishment Injury and Illness Data Our data on worker health at US manufacturing plants was collected by the Occupational Safety and Health Administration (OSHA) as part of the OSHA Data Initiative (ODI). Under the ODI, OSHA and state-level regulators since 1996 have collected data on the number of reported injuries and illnesses at a sample of manufacturing establishments. Each year, approximately 50,000 manufacturing plants are drawn from the universe of 120, ,000 firms that includes all privately owned establishments with at least 40 employees and within participating states This data is publicly provided on David Dorn s website, for which we are grateful. 23 As of 2014, over 99% of manufacturing establishments in the US were privately owned. The size minimum was originally 60 employees in 1996 and 50 in 1997, and 40 thereafter. Several self-regulated states have attrited from the ODI program since beginning it in 1996: OR, WA, and WY in 1997, SC in 1999, AK and AZ in 2006, and VA in 2007

20 McManus and Schaur 19 We address the resulting concerns for identification that arise from the non-representative sample below. Each firm self-reports the total number of injury and illness cases reported in the year, the number of cases that required time away from work or restricted work or transfer to other duties, and equivalent full time employment, imputed as the total hours worked by all employees in the year divided by a full time employment standard of 2000 (50 weeks at 40 hours per week). We use this information to construct our primary measure of worker safety at the firm, the OSHAdefined Total Case Rate (TCR). The TCR is the incident rate of injuries and illnesses per 100 full-time workers in a year: As employment is defined by hours rather than employees, the TCR gives a measure of risk per unit of time worked. The distribution of TCR in the data is shown in Figure 1. Figure 1: Distribution of Injury Rates at Manufacturing Establishments in ODI Dataset

21 McManus and Schaur 20 Notes: Injury rates are measured as total cases of injury and illness per 100 full-time workers in a year. Sample is establishment-year observations with TCR, and omits 11% of observations with a incidence rate of zero, which are necessarily excluded from our primary estimation in log-differences. 4.3 Empirical Model We predict that import competition adversely impacts worker health in the affected industry, particularly at the least productive firms which are most likely to be pushed out of the market. We use firm size to proxy for productivity, so that our formal empirical hypothesis is: Hypothesis: An increase in Chinese imports leads to higher injury and illness rates at firms in the competing industries, and these effects are greatest for small firms. Our empirical model for the change in injury risk for workers at plant producing in geographic region and selling in industry is: Δ α Δ Δ S z S z ε where Δ is the change in import competition affecting the plant over the -years following time. We use the log change in TCR at the plant as a relative measure of the change in injury risk. In lieu of available sales or output measures, we instrument for firm size using the natural log of employment, and use its base-year value as employment later in the time period is likely endogenously determined by import exposure.

22 McManus and Schaur 21 We follow Autor et al. (2013) and use the tremendous growth in Chinese exports to the US to construct our measure of import competition. Consistent with the granularity of the trade data, we consider that. That is, we assume that all US firms in the industry are symmetrically affected by a change in import competition. We therefore model the relevant measure of import exposure for a plant over a particular interval as the growth in Chinese imports in the industry. 24 As with injury rates, we log-transform the data and measure industry import exposure as approximate percentage growth to accounting for differences in scale across industries. Our estimating equation therefore becomes: Δ ln( R ) α Δ ln( ) Δ ln( ) ln E ln E ε where Δ ln( ) is the log change in the value of goods imported into the US from China that compete with the output of industry. We also include time and industry fixed effects in our preferred specification. The base-year fixed effects capture any unobserved shocks that jointly affect safety across all industries in the period, such as a change in the macroeconomy or 24 An alternative approach would be to consider how import competition within a geographic region affects injury rates in the area. ADH use this industry-level import data to construct a measure of import competition at a geographical level using the employment shares across industries in a region to allocate industry-level imports to study the effects on local labor market outcomes. Advantageously, their approach allows for potential spillovers across industries in a region. We instead consider industry-level effects largely because, unlike the local labor market outcomes studied in ADH, our dependent variable is measured at the plant-level in the data, and we do not have a sufficiently large sample to estimate industry-region outcomes. We do include fixed effects at the geographic level used in ADH so that, while we do not exploit geographic variation for our identification, we do control for it.

23 McManus and Schaur 22 regulatory environment. The industry fixed effects capture common trends in injury rates across firms in an industry over the sample period. In this model, our hypothesis above predicts and. The marginal effect of import competition on injury rates at a given firm is given by ln E, so suggests trade worsens worker health at the smallest firms and implies that this effect is diminishing with firm size. A sufficiently negative implies that injury rates at larger firms are improving in the level of import exposure. To the extent that imports shocks and demand are correlated, estimates from a regression of injury rates on growth in Chinese imports to the US understate the effect of import shocks as a positive demand shock partly reduces the likelihood of firm shutdown and therefore injury rates, per our theoretical prediction while also increasing imports. Following Autor et al. (2013), we use an instrumental-variables approach to deal with the potential endogeneity in the change in US imports and isolate the effect of supply shocks. We instrument for supply-driven growth in Chinese imports in the US in a particular industry using the corresponding growth in Chinese exports to a set of other developed economies 25,, and estimate the model using two-stage least squares (2SLS). In our preferred specification, we estimate the model with 5, and so estimate the effect of a 5-year growth in import exposure on the change in injury rates over the same period. This is consistent with Amiti and Davis (2011), who estimate the wage effects of trade liberalizations using 5-year log changes. We estimate the model also using as both a robustness check and a means of describing the associated dynamics of firm responses to average import shocks over periods of varying length. 25 These are Australia, Denmark, Finland, Germany, Japan, New Zealand, Spain, and Switzerland. The trade flows to this aggregate are similarly derived from UN Comtrade data and publicly available from David Dorn s website.

24 McManus and Schaur Results In our preferred specification, we use two-stage least squares to estimate the model for five-year changes. The results are shown in Table 1. Standard errors are robust to within-firm clustering. In columns (1) and (2), the injury effects of trade are modeled as symmetric across all firms in an industry. The model in columns (3) and (4) includes an interaction term allowing for import competition within an industry to have heterogeneous effects across firm size. Table 1: Injury Effects of Import Competition: Second Stage IV Results Dependent Variable: 5-Year Log Growth in Injuries Rate (1) (2) (3) (4) Δln(Imports China to US ) ** (0.0339) ln(employment) *** Δln(Imports China to US) ln(employment) ( ) ** (0.0345) *** ( ) *** (0.0638) (0.0125) ** ( ) *** (0.0652) (0.0126) ** ( ) SIC4 Fixed Effects Y Y Y Y CZ Fixed Effects N Y N Y N Notes: All regressions include a constant and year fixed effects. Smaller sample size in (2) and (4) is due to missing geographic information for a small share of establishments. Robust standard errors are clustered by plant. * p<0.10, ** p<0.05, *** p<0.01 The results in columns (3) and (4) indicate that size indeed matters, in support our empirical hypothesis that import competition deteriorates worker health and safety at the smaller firms in the affected industry. Figure 2 presents our estimates for the distribution marginal effects across firm size, along with the distribution of firm size in the sample. We find significant evidence

25 McManus and Schaur 24 that the supply-shocks associated with greater import competition have an adverse effect on worker health at competing domestic plants with less than 250 workers. Figure 2: Heterogeneous Health Effects of Import Competition by Firm Size The mean 5-year growth rate for Chinese imports in an industry was 130% for our sample. We use this increase to scale our coefficients for interpretation. This mean change in industry-level imports was associated with a 14% increase in injury rates for the smallest decile of firms, a 10% increase at the median, and an 8% increase at the seventy-fifth percentile. In specifying our variables in five year changes, we are consistent with specifications in related work (e.g. Amiti and Davis, 2011), but necessarily sacrifice observations within our unbalanced panel by extending the interval and therefore increasing the likelihood that a plant closed, changes ownership, or switched industries. We estimate our preferred model for

26 McManus and Schaur 25 different interval durations and present the results in Table 2. The results demonstrate the relative persistence of the observed effects over the short to medium run. Table 2: Injury Effects of Import Competition at Different Intervals: Second Stage IV Results Dependent Variable: Log Growth in Injuries Rate Over Different Time Intervals 1-Year Changes 2-Year Changes 3-Year Changes 4-Year Changes 5-Year Changes 6-Year Changes Δln(Imports China to US ) (0.339) ln(employment) *** Δln(Imports China to US) ln(employment) ( ) (0.0252) ** (0.0867) ** ( ) ** (0.0147) * (0.0618) ( ) ** (0.0101) *** (0.0630) * (0.0105) *** ( ) *** (0.0652) (0.0126) ** ( ) ** (0.0623) (0.0160) ** ( ) N Notes: All regressions include a constant and year, industry (SIC4), and CZ fixed effects. Robust standard errors are clustered by plant. * p<0.10, ** p<0.05, *** p< Conclusion We find empirical evidence that the growth in Chinese imports in the US in the late 1990s and early 2000s significantly increased worker injury and illness rates in the competing industries in the short to medium run. The heterogeneous within-industry effects were greatest for small firms and diminishing in firm size. We estimate that the mean 5-year growth in Chinese imports (130%) increased injury and illness rates by 10% on average at plants in the competing industry over the same interval, and about 14% at the smallest decile of plants. Our theoretical model predicts that firms respond to greater shutdown risk by allocating resources towards productivity at the expense of safety. Based on the theory, we hypothesize that import competition will deteriorate worker health outcomes in the short run at marginal firms at risk of being pushed from the market during the transition to the new open economy

27 McManus and Schaur 26 equilibrium. This mechanism substituting resources away from safety and into productivity allows firms to prolong their life in the market and slow down the adjustment process following a trade liberalization, but does so at the cost of worker welfare. Our empirical results imply that worker health is an important channel for welfare effects of trade, even for a developed economy with strong institutions, and therefore should inform trade policy. Further, they suggest that occupational safety and health supervision and regulatory policies targeted towards industries affected by import competition, and the marginal firms within them in particular, could help ameliorate the adverse effects that arise there and may facilitate the adjustment process by leading less productive firms to exit the market sooner in the transition.