Safety at work and working conditions in Europe

Gabriele MazzoliniDEFAP - Graduate School in the Economics and Finance

of Public Administration

June 27, 2010

Abstract

This paper contributes to the analysis of the determinants of work-place accidents, using European Working Conditions Survey data. Weinvestigate the role of workplace conditions and safety at work in reduc-ing the probability of workplace accidents and the cumulative duration ofabsence following an injury. Particular attention is given to the role ofinstitutions and to di¤erences in work organization practices. The mainresults show evidence of an inverse relation between safety at work andaccidents. When disaggregate analyses are performed by personal char-acteristics, �rm attributes and working conditions, results suggest thatthe e¤ect of safety at work is statistically signi�cant only when job ischaracterized by an intrinsic risk at work.

Keywords: risk at work, workplace accidents, working conditions, physicalrisk factors, occupational health and safetyJEL classi�cations: I18, J28, J81

1 Introduction

In this paper we investigate how di¤erences in working conditions and in safetyat work a¤ect workplace accidents both in terms of probability of su¤eringoccupational injuries or cumulative duration of absence from work.For the European Union and its Member States, the promotion of employ-

ment, the improvement of working conditions and the prevention of workplaceaccidents are amongst the primary objectives to pursue, as stipulated in Ar-ticle 136 of the Treaty of Rome and con�rmed by the EU Framework Direc-tive 89/391. Although, the increasing attention of national governments inimplementing detailed legislations to promote safer workplace standards hascontribute recently to the decrease of workplace accident rates, more than 5thousands workers die every year due workplace accidents, resulting in a GDPloss of more than 4 percent of in the European Union. Workplace accidents,in fact, entail excessive costs, either on humanitarian or on economic grounds,which are related to insurance indemnity, safety intervention, security devices

1

and other expenditures. Consequently, government interventions are �nalized tomandate that the employers set up measures which guarantee health standardsand safety at work. Stringent laws and regulations, following ILO conventionsand EU Framework Directive 89/391, are implemented for obliging employersto satisfy all necessary requirements in establishing occupational health servicesand to contribute to work injury programmes with payments that can vary ac-cording with the type and the nature of risk at work and on the basis of theoccupational health services provided.The use of the economic studies on the value of mortality and injury risks

has been a key component in the evaluation of a wide range of health, safety andenvironmental policies. There is an extensive economic literature that estimatesthe value of statistical life (see Viscusi and Aldy, 2003 for a survey) and evaluatestradeo¤s between money and fatality risks for insurance companies and govern-ments Although the literature on the value of statistical life is extensive, thetheoretical and empirical literature dealing with risk at work and occupationalaccidents is incomplete and not exhaustive. The results reached so far are notunambiguous and the contributions on the analysis of the determinants of riskat work are mostly limited to single country data. Several articles contributein investigating the e¤ect of risk at work on the labour demand (DeLoire andLevy, 2001; DeLoire and Levy, 2004; Grazier and Sloane, 2006; and Scha¤nerand Kluve, 2006), gender di¤erences in studies of health and work safety (Her-sch, 1998; Leeth and Ruser, 2006; Leombruni et al., 2009; and Hoskin, 2005), thee¤ect of unionization in preventing workplace accidents (Litwin, 2000; Baugherand Roberts, 1999; and Fenn and Ashby, 2001), the role of immigrant labourin workplace accident rate (Berger and Gabriel, 1991; Hamermesh, 1998; andBauer, Million, Rotte and Zimmermann, 1998), di¤erences in injury rates bytype of contract (Guadalupe, 2003; Hernanz and Toharia, 2004; Williamson etal., 2009; and García-Serrano, Hernanz & Toharia, 2008), the e¤ect of moreworking hours on workplace accident rates (Wilkins, 2004), the consequences ofa workplace accident, either for the return to work (Galizzi and Boden, 2003a) orfor earnings (Reville and Schoeni, 2001; Crichton, Stiliman and Hyslop, 2005;and Galizzi and Boden, 2003b) and the relationship between macroeconomicconditions and mortality due to hazardous working conditions (Ruhm; 2000and Boone and van Ours; 2002, 2006).Previous studies did not investigate the relation between safety at work and

accidents, due to data limitations in providing measures of exposures to danger-ous agents. The motivation of our empirical analysis is to provide evidence thatinvesting in safety reduces the expected cost of workplace accidents, cost whichdepends on the probability that an accident occurs and on the consequences fol-lowing an injury. Following Lanoie (1991), our hypothesis is that the �rm hasto face a trade-o¤ between the precaution expenditure and the expected cost ofa workplace accident. If a workplace accident occurs and the �rm is found neg-ligent, the �rm has to pay costs, increasing with the seriousness of occupationalaccident, for replacing injured workers and the �ne for compensating victims ofaccidents. With a cost/bene�t analysis related to occupational risks, the �rmchooses safety expenditure that minimizes the expected cost of a workplace ac-

2

cident. For this purpose, analyzing the probability that an accident occurs andthe consequences of occupational accidents is crucial for estimating consistently�rm�s expected costs.This paper contributes threefold to the previous literature. First, we provide

a unique European cross-country comparison of the e¤ect of working conditionsand safety at work on workplace accidents using European Working Condi-tions Survey data. Second, we estimate the probability and the duration ofabsence accounting for the endogeneity in the provision of safety at work. En-dogenous selection arises from two di¤erent sources. Di¤erences in aversion torisk at work may in�uence workers�career preferences entailing a selection ef-fect. Firms�behaviour may also induce endogenous selection when they choosesafety expenditures accounting for national laws and regulations, which man-date to set up occupational health services to guarantee health and safety atwork standards, and for achieving quality management certi�cations (OHSAS18001 and ISO 9001) that may reduce insurance premiums due to �nance workinjury programmes. Controlling for endogenous selection, we may estimate thecausal e¤ect of reducing occupational risks on the probability of accident andon the seriousness of the consequences that an accident may entail. Finally,we highlight how the e¤ect of safety at work may vary according to di¤erencesin personal and �rm characteristics and working conditions. The aim of thesedisaggregate analyses is to de�ne more clearly the patterns of risk at work forimplications of government policy evaluation.The paper is organized in the following way. In Section 2, we present the

institutional background related to the e¤ort of creating a common and uniquelegislative framework to safeguard occupational health and safety in the Euro-pean Union; in Section 3, we review the relevant literature; Section 4 describesthe dataset; in Section 5 we introduce the main empirical issues related to inves-tigating the probability and the duration of absence due to a workplace accident;Section 6 discusses the results and concluding remarks follow.

2 Institutional framework

One possible explanation of recently decreasing rates of workplace accidentsin all European countries is the increasing attention of national governmentsto implementing detailed legislations which promote safer workplace standards.Regulations that limit risks at work are necessary to reduce expected costs of oc-cupational injuries, which are excessive both on economic and on social grounds.Government interventions are �nalized to mandate that the employers set upmeasures which guarantee health standards and safety at work. Following ILOconventions and EU Framework Directive 89/391, law and regulation, prescribethat employers satisfy all necessary requirements in establishing occupationalhealth services (OHS).Occupational health services are de�ned by Recommendation (r.112) (1959)

of the International Labour Organization (ILO) as �all services established inor near a place of employment for the purpose of protecting workers against

3

any health hazard that may occur on the work place; contributing towardsthe workers�physical and mental adjustment and contributing to the establish-ment and maintenance of the highest possible degree of physical and mentalwell-being of the workers�. Occupational Safety and Health Convention (c.155)(1981) prescribes that �each member state has to formulate, implement, en-force and periodically review a coherent set of laws and regulations concerningoccupational health and safety and working environment�. Moreover, an appro-priate system of inspection has to be implemented in order to ensure adequatepenalties for violations of the laws and regulations implemented. OccupationalHealth Services Convention (c.161) (1985) mandates that �each member stateundertakes the development of progressively adequate and appropriate occupa-tional health services for workers who su¤er from speci�c occupational risks�.Following Recommendation r.171 (1985), OHS should intervene �to supervisethe working environment; to guarantee workers�health; to provide information,education, training, advice and to ensure �rst-aid and health treatment�.The role of Framework Directive 89/391 is fundamental in three respects.

First, it introduces general principles concerning the prevention of occupationalrisks and the protection of safety and health. Second, it de�nes the dutiesand responsibilities of employers and employees. Third, it underlines the im-portance of participation and cooperation of workers and their representativesto the establishment of tasks and requirements to avoid any occupational ac-cident (Article 1). The aim of EU Framework Directive is to coordinate anduniform national provisions on safety and health at the work place, which di¤erwidely among the European countries and which allow competition at the ex-pense of safety and health. Improving occupational health and safety standardsand taking the necessary measures to avoid occupational accidents are primaryobjectives pursued by the European countries through dialogue and balancedparticipation of employers and workers or their representatives (Article 4). Em-ployers, taking into account the nature of the activities of the enterprise, havethe legal duty to take the necessary measures to prevent occupational risks andto provide either information and training or the necessary organization andmeans. These coherent overall prevention policies are undertaken not only toevaluate and to avoid risks but also to eradicate the risks at source. This canbe achieved by replacing the dangerous tasks with non-dangerous or the lessdangerous ones, by adopting new and safer technologies and processes and bygiving to the workers the appropriate instructions and training (Article 6.2).Employers should designate one or more workers with the necessary capabilitiesand means to carry out activities related to the protection and prevention of oc-cupational risks. Alternatively, employers are bound to enlist external servicesor persons with the necessary aptitudes and personal and professional means(Article 7). The employer should take �the necessary measures for �rst aid,�re-�ghting and evacuation of workers, adopted in accordance with the natureof the activities and the size of the enterprise, and arranges any contacts withexternal services involved in providing �rst aid, emergency medical care, rescuework and �re-�ghting" (Article 8.1). In the event of the serious and imminentdanger, the employer has to ensure that all workers are able to take the ap-

4

propriate steps to avoid occupational accidents (Article 8.5). This is possibleonly when workers are informed on potential risks at work and well trained foravoiding them. Indeed, the employers and workers�representatives should be incontinuous contact to discuss potentially dangerous tasks, to designate workersinvolved in preventive activities and to plan the training on occupational risks.Employers should provide �adequate safety and health training, in form of infor-mation and speci�c instructions, on recruitment, in the event of a job change orwith respect to a change in work equipment or technology�. The training shouldbe adopted to provide information aimed at preventing new or changed risk andshould be repeated periodically if necessary. An appropriate training shouldbe provided to workers�representative with speci�c responsibility for the safetyand health of workers. EU Framework Directive 89/391 also prescribes the dutyof workers to take care of their own safety and health and of the health of otherpeople which can be a¤ected by their actions or omissions at work. Workers�obligations care are prescribed to correctly use personal protective equipment,machinery, apparatus, tools, dangerous substances, transport equipment andother means of production. They should inform immediately the employer ofany situation that may represent a serious risk for them and for other people in-volved in dangerous tasks. Moreover, they should cooperate with the employerto enable any tasks or requirements for ensuring that the working environmentand working conditions are safe and pose no risk to safety and health.Following the ILO conventions1 on occupational health and safety at work

and the EU Framework Directive 89/391, European countries enact their ownlegislations and de�ne the authorities responsible for such legislation2 which areresponsible for setting the peculiarities and requirements of occupational healthservices. However, EU countries present signi�cant di¤erences related either tolegislative structure or to speci�c requirements. The legislation may be struc-tured in very detailed and integrated with lower level regulations; alternatively,it may be characterized by only framework legislation, with plenty rooms for in-terpretation and applicability. Conversely, the competent authorities may applygeneral obligations on occupational health services and give detailed provisions.Requirements may concern the necessary quali�cations of OHS personnel, theminimum level of required resources, the de�nition of speci�c tasks for OHS andthe achievement of quality management quali�cations, as certi�cations, state-ments of the need for quality improvement or other requirements.Occupational health services are �nanced uniquely by the employer, expect

for Finland, where the National Social Insurance Institution contributes to re-imburse part of the total expenditure. OHS may be organized by a singleenterprise, by groups of enterprises or, alternatively, by public authorities, o¢ -cial services, social security institutions or any other bodies authorized by thecompetent authority or combinations of these institutions. In addition, the em-ployer, liable for occupational accidents which occurred, is obliged to contributeto work injury programmes with payments that can vary according to the type

1We discuss the main features of ILO conventions on occupational health and safety Inappendix A1.

2National legislations and competent authorities are listed in Table A1.1, in appendix.

5

and the nature of risk at work and on the basis of safety at work standardsprovided with occupational health services. Consequently, employers may re-ceive a discount on the insurance fee if they introduce preventive measurementswhich help to reduce risks of an occupational accident and to improve occupa-tional health services. The lack of minimum safety and hygiene conditions orof preventive measurements required by national legislation is associated witha punitive action in the form or a surcharge. Di¤erences in premium dependon the industrial sector, size of company and degree of safety measures appliedto reduce occupational risks. Reduction of the level of occupational accidentsand risks and improvement of working conditions may result in a signi�cantdecrease in insurance premium of work injury programmes that employers hasto pay to insurances. In addition, national governments may introduce incen-tives for inducing employers to implements measures related to workplace healthpromotion. Workplace health promotion includes all joint measure of employ-ers, employees and society to improve health and well-being at the workplace3 .The goal of workplace health promotion is to increase the level of quality man-agement aimed at reducing occupational risks thanks to measures that mayintroduce innovations in behavioral prevention at the workplace, in work orga-nization and design and/or in�uence health determinants. For all these reasons,the employer may choose to implement further measures to guarantee higher-than-mandated health and safety standards at work. The above-mentionedinnovations may also be introduced by the employer to satisfy the quality re-quirements of management systems veri�ed periodically by external monitoringauthorities.

3 Background literature

The use of the economic studies on the value of mortality and injury risks hasbeen a key component in the evaluation of a wide range of health, safety andenvironmental policies. There is an extensive economic literature that estimatesthe value of statistical life (see Viscusi and Aldy, 2003 for a survey) and evaluatestradeo¤s between money and fatality risks for insurance companies and govern-ments. Although the literature on the value of statistical life is extensive, thetheoretical and empirical literature dealing with risk at work and occupationalaccidents is incomplete and not exhaustive. International empirical evidenceshows that on-the-job injury rates are quite di¤erent for di¤erent groups ofworkers: injury rates are higher for men, they vary with workers nationalityand type of contract. The socio-economic literature has tried to empiricallytest whether these di¤erences are explained by di¤erent occupational choices,di¤erent working conditions within sectors or occupations or other factors cor-related with gender and nationality, such as, for example, education, languageskills and degree of risk aversion. The results reached so far by the literatureare however ambiguous. Moreover, the empirical contributions to the analysis of

3The Luxemburg declaration on Workplace Health Promotion

6

the determinants of risk at work are several but mostly limited to single countrydata.In this context, DeLoire and Levy (2001), DeLoire and Levy (2004), Grazier

and Sloane (2006) and Scha¤ner and Kluve (2006) investigate the e¤ect of risk atwork on the labour demand function. DeLeire and Levy (2001) �nd that, otherthings equal, US women have a higher level of concentration than men in saferjobs. Using family structure as a proxy for aversion to the risk of death, DeLoireand Levy (2004) �nd that, in the US labour market, individuals with strongaversion to risk of accidents, such as single parents raising children, choose saferjobs. However, family structure only partially explains this correlation sinceoccupational risk choices are in�uenced signi�cantly also by gender. Grazierand Sloane (2006) present similar �ndings investigating the UK labour market.Scha¤ner and Kluve (2006) con�rm the theory of wage compensation for anoccupational risk showing that, in Germany, observed gender wage di¤erentialsare partly due to di¤erences in the occupational injury risk of the jobs occupiedby men and women.Focusing on gender di¤erences in studies of health and work safety, Hersch

(1998) �nd that in the US women are 71 percent as likely as men to experiencean injury or illness after controlling for di¤erences in employment. Leeth andRuser (2006) con�rm this result showing that, after controlling for occupation,US male workers experience twice as many workplace fatalities as women butfewer nonfatal injuries. Investigating occupational injuries in Italy, Leombruniet al. (2009) �nd that average injury risk has decreased from 4.7 percent in 1994to 3.8 percent in 2003. This change is almost fully explained by the fact that theinjury rates for male decreased from 6.1 to 4.8 percent, while female rate wereroughly constant around 2 percent, with a temporary increase in the years 1999-2001. Among women, injuries are more frequent when performing tasks thatimply high speed performances. Moreover, these numbers may underestimatewomen injury rates because women have been found to su¤er injuries (such ascarpal tunnel syndrome or back injuries) that are more di¢ cult to diagnose andtherefore undercounted (Hoskin, 2005).Several articles focus on the determinants of a workplace accident. Litwin

(2000), Baugher and Roberts (1999) and Fenn and Ashby (2001) investigatethe e¤ect of unionization in preventing occupational accidents. Litwin (2000)argues that trade unions improve working conditions and that there is no sig-ni�cant monotonic inverse relationship between union density and injury rates.Union presence seems to also in�uence risk perception, as found by Baugher andRoberts (1999). Fenn and Ashby (2001) �nd that establishments with a higherproportion of unionized employees are associated with higher risk of reportedinjuries or illnesses; in addition, employees in larger establishments su¤er lowerprobability of being injured.Focusing on the role of immigrant labour, few contributions studying the

US (Berger and Gabriel, 1991; Hamermesh, 1998) provide evidence showingthat immigrants are not employed in higher risk jobs. Recent studies point outhowever that occupational di¤erences by workers nationality can be explainedby di¤erent average characteristics of the workers, such as language pro�ciency

7

and educational levels. Bauer, Million, Rotte and Zimmermann (1998) arguethat there are no di¤erences in workplace risks due to work organization betweennative and foreign workers. However, foreigners are employed in industries whichface higher risks due to less safe technologies and, consequently, higher rates ofworkplace accidents. Focusing on the interaction between job risks of nativeand foreign workers, guestworkers have a positive e¤ect on the workplace safetyof natives.Ambiguous results emerge when considering di¤erences in injury rates by

type of contract. Economic theory predicts the existence of a pure �contractuale¤ect� leading to a higher accident rate which depends on the short durationof the temporary contract. Temporary contracts reduce the incentives to in-vest in speci�c human capital or result in higher workers e¤orts to increaserehiring probabilities. On the one hand, Guadalupe (2003), assuming that in-jury rates depend on di¤erences in incentives provided by contracts, identi�esa systematic di¤erences between accident rates of workers employed on �xedterm contracts and those of workers employed on permanent contracts in theSpanish labour market. On the other hand, although temporary workers ex-hibit higher work injury and illness rates than permanent workers, they exhibita lower likelihood of work injury and illness than permanent workers once theanalysis controls for a given set of working conditions (Amuedo-Dorantes 2002).Hernanz & Toharia (2004) compare Italy and Spain and �nd that di¤erences inthe probability of workplace accident for workers with temporary and perma-nent contracts vanish when job characteristics are controlled for. Similar resultsare found by Williamson et al. (2009) who focus on short-haul truckers in Aus-tralia. García-Serrano, Hernanz & Toharia (2008) argue that workers employedthrough Temporary Help Agencies (THAs) exhibit lower probability of su¤er-ing a serious/fatal accident and lower duration of absence after a work-relatedaccident as compared to "direct" temporary contracts and permanent workers.Focusing on the consequences of working hours, Wilkins (2004) does not

�nd robust evidence that more hours worked per week a¤ect workplace acci-dents rates because positive relationship between working hours and workplaceaccidents rates, found using cross-sectional data, is not con�rmed by results ofa longitudinal analysis.Focusing on the consequences of workplace accidents, there is no broad con-

sensus on the empirical �ndings on the relationship between risk at work andlabour market outcomes. The comparison and generalisation of results is madeeven more di¢ cult by fact that most studies are based on single country cross-sectional data or on a case studies. Investigating the consequences of workplaceaccidents on the return to work, Galizzi and Boden (2003a) do not �nd robustevidence that the use of after-tax earnings, that may change the magnitude ofincome replacement, is a valid incentive to return to work. The authors �nd thattime o¤work is a¤ected by workers�pre-employment history and that it is lowerif workers return to their pre-injury jobs. If workers experience a long period o¤the job after the accident, they carry with them the burden of their injuries evenafter returning to work. Reville and Schoeni (2001) and Crichton, Stiliman andHyslop (2005) investigate the consequences of workplace accidents in terms of

8

their e¤ects on earnings. Analysing the long-term labour market consequencesof permanent partial disabilities, Reville and Schoeni (2001) estimate that earn-ing losses are in the range of 25 per cent of the income of uninjured workers,mainly due to a decline in employment. Crichton, Stiliman and Hyslop (2005)argue that injuries have negative e¤ects on future labour market outcomes ifthe workers receive more than 3 months of earnings compensation for injury.The e¤ects are substantial for longer-duration injuries and increase with injuryduration, without any evidence of decline over the �rst 18 months after leavingAccident Compensation Corporation. Analyzing the impact of earnings com-pensations on di¤erent workers categories, the authors argue that earning lossesare more sever for women, older workers and workers with lower pre-injury earn-ings or with less stable employment histories. Regarding the di¤erences in thee¤ects of work-related injuries on the work career between genders, Galizzi andBoden (2003b) show that three and a half years after the post-injury quarter,women lose an average of 9.2 percent of earnings, while men lose 6.5 percent inthe US labour market. Di¤erences in observed personal, employer, and injurycharacteristics do not explain the gender disparity in lost earnings. Womenare employed less after injury, and this �employment�e¤ect accounts for abouthalf the gender gap. Discrimination remains a possible explanation for the un-explained half of the gap. The authors discuss why an injury might triggeremployer discrimination. Even if employers do not discriminate deliberately,they usually have limited information about workers� attributes. Therefore,they may end up preferring workers for whom they can obtain more accuratepredictions of skills or mobility behaviour (Aigner and Cain 1977).Finally, one strand of literature investigates if variations in macroeconomic

conditions in�uence the rates of workplace accidents. Ruhm (2000) �nds aninverse relationship between macroeconomic conditions and mortality due tohazardous working conditions. Boone and van Ours (2002, 2006) argue thatvariations in workplace accidents are due to the procyclicality in workers�ab-senteeism. Indeed, workplace accidents are a¤ected negatively by both the levelof unemployment and the change in employment status.

4 Data and descriptive statistics

We use a dataset which provides information on workers experience and work-ing conditions across European countries based on the 2000 and 2005 waves ofEuropean Working Conditions Survey (EWCS). The questionnaires collect in-formation on respectively 21,703 and 29,680 workers4 . Relevant for our analysis,the EWCS includes questions on the number of days o¤ work due to workplaceaccidents which the worker su¤ered in the 12 months previous to the interview,

4The EWCS samples followed a multi-stage strati�ed and clustered design with "a randomwalk" procedure for the selection of the respondents at the last stage. All interviews wereconducted face-to-face in the respondent�s own household. The target number of respondentswas around 1,400 in EU15 countries in 2000 wave and around 1,000 in EU27, Turkey, Croatia,Switzerland and Norway (except in Cyprus, Estonia, Luxembourg, Malta and Slovenia, wherethe target number was around 600) in the 2005 wave.

9

the relevance of the exposure to dangerous agents, working time and level of in-volvement on the job5 . The �nal database selected for the analysis is composedof 37,147 observations, of which 7,642 relative to 2000 wave and 8,401 to 2005wave of the EWCS6 .For the analysis, we construct two dependent variables, accident and ab-

sence, using information reported by the worker on absences from work due toaccident(s) at the workplace over the previous 12 months. If the worker su¤ereda workplace accident which entailed at least one day of absence from work, thevariable accident is equal to one; otherwise, it is equal to zero. Only 3.95 percentof the workers considered su¤ered a workplace accident. The workplace accidentrate in our sample is found to be rather in line with aggregate statistics7 . Thediscrete variable absence is constructed measuring the total number of days o¤work due to workplace accident(s) in the past 12 months. The outcome of thevariable absence is equal to zero if the worker does not su¤er any workplaceaccident; otherwise, the variable may assume any value in the interval 1 to 365.Among injured workers, the mean number of days of absence from work due toworkplace accident is 26.08.Table A2.1 in the Appendix A2 shows a complete list of the variables used

in the analysis and their means. Demographic characteristics of respondentsinclude gender, age, if native or not and family composition. Job attributes arerelated to working quali�cations (based on the International Standard Classi�-cation of Occupations (ISCO)). Firm characteristics include information on theeconomic sector in which a �rm operates (classi�ed using International StandardIndustrial Classi�cation of All Economic Activities (ISIC)) and the size of theenterprise. We also include in the empirical analysis information on whether theemployee works part-time and if he/she has a long-term contract. To controlfor di¤erences among EU countries, we include country �xed-e¤ect dummiesand a time �xed-e¤ect dummy in the model. These variables help us to capturerespectively the heterogeneity in probability and duration of accidents amongEU countries and whether the observation belongs to the second or third waveof EWCS.Table A2.2 reports descriptive statistics of workplace accident rates, in the

�rst column, and average durations of absence o¤work for workers who su¤ereda workplace accident, in the second column. Focusing on demographic char-acteristics as potential determinants of accidents, males seem to su¤er moreworkplace accidents than females but that there are no signi�cant di¤erences inthe duration of absence by gender. Statistics related to workplace accident ratesby gender are in line with gender di¤erences argued in several studies on healthand work safety (Hersch, 1998; Leeth and Ruser, 2006; Leombruni et al., 2009).

5The dataset includes also information on personal characteristics, �rms attributes, workingcondition and work organization.

6Given the aim of our analysis, in this study we restrict the dataset to employees aged 18to 65 and exclude from the analysis any workers who are still studying or employed in themilitary or in the agriculture sector. We restrict the sample to estimate the determinants ofworkplace accidents for workers in the labour force and employed in the private sector.

7The European Commission (2004) �nds that about 4% of the workers are victims of anaccident at work during the year

10

However, the spurious correlation between gender and workplace accident ratemay have to do with di¤erences in occupational risk choice: males may be em-ployed in more risky jobs and, consequently, raw di¤erences in safety at work bygender may be attributable to the e¤ect of occupations or sectors on the prob-ability of accident. Workplace accident rates decrease with age; whereas theconsequences of an occupational injury do not change signi�cantly8 . Related tothe correlation between immigration and workplace accident rate, descriptivestatistics show that there is no signi�cant di¤erence between immigrant andnative, in line with �ndings of Bauer, Million, Rotte and Zimmermann (1998);whereas, guestworkers seem to be a¤ected by more serious consequences whenan accident occurs. However, we suspect that this di¤erence can be explained bydi¤erent occupational risk choices: immigrant may su¤er more serious injuriesbecause they are employed in occupations and sectors that are characterized byhigher risk at work. Descriptive statistics show that households are a¤ected byhigher workplace accident rate and by more serious consequences of accident.These results suggest that head of households are more willing to trade risk forwages with respect to their wife/husband. Statistics related to workers with andwithout children are ambiguous since workplace accident rate is higher betweenworkers without children; conversely, the consequences of a workplace accidenton average are more serious for workers with at least one child.Investigating job attributes, we notice that manual workers present higher

workplace accident rates than white collars and service workers. However, theconsequences of an occupational accident are more serious in terms of days o¤work, on average, for service workers. These results seem to suggest that whitecollars are not a¤ected by signi�cant risk at work. Manual and service workers,on the other hand, su¤er from risk at work that seems to entail frequent butnot serious accident among manual workers while service workers are a¤ectedby more dangerous risk in term of consequences. Descriptive statistics con-�rm that workplace accidents occur more frequently in the industrial sectors9 ,namely Manufacturing, Construction and Electricity, Gas and Water Supply.More serious consequences seem to occur in service sectors, namely Wholesaleand Retail Trade, Hotels and Restaurants, Transport and Storage, Post andTelecommunications, Financial Intermediation and Real Estate. CommunitySocial and Personal Services, namely Public Administration and Educationaland Health Services, present lower workplace accident rates and less seriousconsequences since they may not present serious intrinsic risks at work. Rawdi¤erences in terms of average workplace accident rate by size suggest that thesize of enterprise is positively correlated with accident rate. These results maybe explained considering that higher occupational risks are more frequent inenterprises with more than 50 workers; indeed, small enterprises (with less than10 employers) are not generally involved in activities that concern exposure todangerous agents, which is a good proxy for serious occupational risks.Following literature related to �contractual e¤ect�on injury rates (Guadalupe,8We may assume that the duration of absence due to workplace accidents is a good proxy

for the seriousness of occupational injuries.9See OCSE (2007) for aggregate statistics on workplace accident rate by sectors.

11

2003; Amuedo-Dorantes, 2002; Hernanz & Toharia, 2004; Williamson et al.,2009), we present statistics that show the relationship between type of con-tract and occupational accident. Descriptive statistics show that part time jobsare characterized either by lower workplace accident rates or by less seriousconsequences when an accident occurs. Relating to the type of contracts, rawstatistics seem to suggest that there is no signi�cant di¤erences between work-ers with long-term or �xed-term contract in terms of the probability of accidentor the duration of absence when an accident occurs. Statistics related to jobattributes and �rm characteristics may indicate that job categories character-ized by intrinsic risk at work seem to be a¤ected by more frequent and seriousworkplace accident.Investigating raw di¤erences by country, Eastern European countries, namely

Romania, Bulgaria and Czech Republic, present workplace accident rates alwaysbelow of 2 percent; whereas, in Finland, one of the most involved countries inregulating occupational health and safety, more than 7 percent of workers werea¤ected by an occupational injury in the previous 12 months. This result raises apotential weakness of our dataset: di¤erences in knowledge relatively to safetyat work and workplace accident may induce workers in di¤erent countries tomisleadingly report the consequences of a workplace accident. Consequently,the robustness analysis of our results are carried out separately for workersemployed in EU15 countries and for workers employed in Eastern Europeancountries. In addition, Table A2.2 suggests that the duration of absence is notpositively correlated with the probability of a workplace accident among EUcountries.Ambiguous results concern time e¤ect since we �nd that workplace accident

rate is signi�cantly higher in 2000 but with less serious consequences. However,these statistics can be explained easily observing the composition of the datasetin the two waves: in 2000 the questionnaire collects information only on workersin EU15; whereas, in 2005 the dataset presents also information on workers fromEU27, Turkey, Croatia, Switzerland and Norway. Consequently, di¤erences inworkplace accident rates by waves should be ascribed to di¤erences by coun-try. To check the robustness of our analysis, we have to carry out estimationsseparately for workers interviewed in 2000 and in 2005.Our study will be focus on the e¤ects of working conditions and safety at

work on workplace accident rates and on the consequences of occupational ac-cidents. Consequently, to estimate the e¤ect of working conditions, we includevariables related to working time and involvement of the job. With respectto working time, we use information on the arrangement of work time and onworking overtime. Workers were asked whether they work at least once a monthon a Saturday (48.22 percent of the workers interviewed) or a Sunday (23.16percent). In addition, they were asked whether they had to work at least 2 hoursbetween 6.00 pm and 10.00 pm (43.19 percent), or at least 2 hours between 10.00pm and 5.00 am (16.40 percent) and more than 10 hours a day (33.51 percent).Involvement in the job is de�ned as working at very high speed for more thanhalf of the time (50.48 percent) and working to meet tight deadlines for morethan half of the time (50.25 percent).

12

Descriptive statistics suggest that a more demanding job entails higher prob-ability of su¤ering a workplace accident. In other words, jobs which requireworking at least once a month on a Saturday or on a Sunday, at least 2 hours inthe evening or in the night or more than 10 hours a day are a¤ected by higherprobability of su¤ering a workplace accident. In addition, descriptive statisticsshow that workplace accident rates are higher for jobs that require working atvery high speed and to meet tight deadlines. On the other hand, there seemsto be a negative relationship between more demanding working conditions andthe duration of absence because of occupational injuries. One possible expla-nation is that more demanding jobs, for what concern either working time orinvolvement in the job, are more precarious and, consequently, their employeesare more feared of losing their jobs and are reluctant to be absent for a longtime.One of the main contributions of this paper is the fact that we create an

index that evaluates the exposure to physical, chemical or biological agents forestimating the e¤ect of safety at work using data provided by the EWCS. Weuse this index in our empirical analysis in addition to the abovementioned vari-ables. In building the index, our hypothesis is based on the assumption thatworkers perception or knowledge about being exposed to dangerous agents is agood proxy for the existence of intrinsic risks at work. We assume that workersare able to self-evaluate risks that they have to face in their jobs. Consequently,the exposure to physical agents (noise, radiation, vibration, etc.), to chemicalagents and to biological agents can used to de�ne all the possible risks thatmay cause occupational accidents. We use workers�evaluations of expositionsto identify if the employers success to reduce risks that are implicit in enterpriseactivities and to provide safer workplace. The questionnaire asks to evaluatefor what part of working time worker is exposed to: Vibrations from hand tools,machinery, etc.; Noise so loud that you would have to raise your voice to talk topeople; High temperatures which make you perspire even when not working; Lowtemperatures whether indoors or outdoors; Breathing in smoke, fumes (such aswelding or exhaust fumes), powder or dust (such as wood dust or mineral dust)etc.; Handling or being in skin contact with chemical products or substances;Radiation such as X rays, radioactive radiation, welding light, laser beams. Foreach of these risks at work, workers have to rate their exposures on a scale from1 to 7, with 1 meaning that he/she is exposed to risk all the time and 7 thathe/she is never exposed. Principal component analysis is performed to com-bine seven variables into one single factor that evaluates the overall exposure tooccupational risks. With this technique, we construct a continuous index thatidenti�es less dangerous jobs when it increases and, alternatively, the existenceof intrinsic risks at work that the employers ignore or do not deal with, whenit decreases. Table A2.3 shows how workers are exposed to each speci�c occu-pational risk. Workers interviewed are characterized generally by high safety atwork standards: the exposure to occupational risks is limited. Indeed, for eachoccupational risk, more than half of workers interviewed are completely coveredagainst the exposure to speci�c agent. In particular, the exposure to radiationis extremely limited: only 4.99 percent of workers are exposed to radiation for

13

at least one quarter of the time. On the contrary, noise is the occupationalrisk that a¤ects more frequently workers interview: 30.60 percent of workersare exposed to noise for at least one quarter of the time. These stylized factsare in line with statistics presented by various publication edited by EuropeanAgency for Safety and Health at Work, which identi�es that vibration and noiseare the most common risks at work and limited exposures to radiation andhand handling chemical substance. Descriptive statistics, related to di¤erencesin probability and duration of accidents with increasing levels of safety at work,are useful to focus on the primary objective of this study. Figure 1 and Figure 2shows, respectively, the negative relationship between safety at work and work-place accident rate and between safety at work and the duration of absence.Reducing the exposure to physical, chemical or biological agents, the employersprovide safer workplace to workers and decrease either the probability that aworker su¤ers an occupational accident or the seriousness of a workplace injurythat may be measured by the consequences that may occurs in the event of aworkplace accidents. However, descriptive statistics do not allow to concludethat safety at work has an e¤ect in reducing workplace accidents and if theyoccurs, their consequences in term of days o¤ work.

5 Empirical Model

In this section, we discuss the econometric issues that are relevant in the pa-per. The analysis is focused on two dependent variables, accident and absence,which de�ne respectively if worker su¤ers an occupational accident in the past 12months and the duration of the period o¤ work due to workplace accident(s).Given the dichotomous nature of the variable accident, we utilise a discretechoice model, as a probit model, to investigate the determinants of the proba-bility that a worker may su¤er an occupational accident:

P �i = �+X0

i� +W0

i + �si + "i (1)

Pi =

(1 if P �i > 0

0 otherwise

where Pi is the binary variable which captures whether a worker has su¤eredone or more workplace accidents, Xi is the vector of the variables that controlfor personal and �rm-speci�c characteristics, Wi is the vector of the variableswhich provide information on working conditions, si is a safety at work indexand "i is the error term, which is normally distributed, "i � N(0; 1).We concentrate our attention also on the duration of absence for identifying

which determinants in�uence the seriousness of an occupational accident. Forthis analysis, we use the discrete variable absence, which measures the totalnumber of days o¤ work due to workplace accident(s) in the past 12 months.Consequently, the outcome for the major part of workers is equal to zero be-cause they do not su¤er any workplace accident in the past 12 months; for the

14

remaining part of population, the variable absence may assume any value inthe interval 1 to 365. The choice of the econometric technique is falling on amodel for count data due to the nature of the dependent variable. Indeed, itis necessary to use an econometric model which takes into account that thisvariable might be zero for a substantial part of the population (namely thoseworkers that did not su¤er any workplace accident in the last year) and that thevalues of the outcome have a cardinal rather than just an ordinal meaning. Forthese reasons, models for count data are preferred to an ordinary least squaremodel or to order response models. In addition, we did not consider the optionof carrying out a duration analysis because, given the nature of the data avail-able, it is not possible to know if the number of days o¤ work is due to a singleworkplace accident or to several injuries su¤ered during the year. Thus, we usenegative binomial model 10 for investigating the duration of absence because ofoccupational injuries:

Ai = �+X0

i +W0

i � + #si + �i (2)

where Ai is the discrete variable which measures the duration of absencebecause of a workplace accident in term of total number of days o¤ work andall other variables as as in equation (1).Particular attention should be used when interpreting results of (1) and (2)

as causal e¤ects on the probability of workplace accident. In particular, relatingto the variable that captures �rms�expenditure to reduce exposures to danger-ous agents, we have to account for endogenous selection in the provision of safetyat work. Indeed, endogenous selection may arise from two di¤erent sources. Dif-ferences in aversion to risk at work may in�uence workers�career preferencesentailing a selection e¤ect. Firms�behaviour may also induce endogenous se-lection when they choose safety expenditures accounting for national laws andregulations, which mandate to set up occupational health services to guaranteehealth and safety at work standards, and for achieving quality management cer-ti�cations (OHSAS 18001 and ISO 9001) that may reduce insurance premiumsdue to �nance work injury programmes. The existence of endogenous selectionin safety at work would entail that the empirical analysis may produce biasedand inconsistent coe¢ cients for the e¤ect of investing for reducing exposures todangerous agents on the probability of accident and on the duration of absence.We propose two di¤erent empirical strategies to deal with the endogeneity insafety at work11 . When estimating the probability of injury on the workplace,we introduce the instrumental variable probit model. The instrumental variable

10 In ordered to estimate the equation of interest consistently and to predict conditionalprobabilities, Cameron and Trivedi (1986) suggest using the full maximum likelihood analysisof the NegBin I model, that satis�es the condition of overdispesion. In addition, NegBinII model provides a further generalization assuming that the amount of overdispersion isincreasing with the conditional mean. In our estimations, NegBin II model is preferred toNegBin I because it entails a higher loglikelihood value with the same number of parameters.11For accounting for the endogeneity, we need at least one additional moment condition

derived from the availability of an instrumental variables, which is by de�nition correlatedwith the (endogenous) safety at work index but uncorrelated with the error terms "i or �i.

15

probit model is used for �tting probit models with maximum likelihood esti-mation where one or more of the independent variables are endogenous. Theendogenous regressors are treated as linear functions of the instruments andthe other exogenous variables. Maximum likelihood estimator is computation-ally feasible in a large sample, as in our analysis, and it guarantees desirableproperties. Indeed, it is asymptotically normally distributed and asymptoticallye¢ cient; in addition, approximate signi�cance tests of parameters are statisti-cally valid and the tests are easy to compute.When estimating the cumulative duration of absence because of workplace

accidents, we have use a two-step technique12 , which entails that the predictedvalue of safety at work index, si, is estimated introducing a set of instrument,Zi, in the auxiliary equation, as shown in the following model:

Ai = & +X0

i'+W0

i�+ �si + �i (3)

si = $ +X0

i� +W0

i{ + Z0

i%+ � i (4)

The auxiliary equation is estimated by ordinary least square model and,then, the predicted value of the safety at work index is used in the negativebinomial model (3).

6 Result

6.1 Empirical evidences

The probability of workplace accident In this section, we present theresults of our analysis of the e¤ect of working conditions and safety at work onthe probability of workplace accident. Results are presented in Table 1, whichwe structure to show how the estimates of working conditions and safety at workindex vary according to di¤erent speci�cations13 . Since we are interested in themagnitude of the correlations, we report exclusively marginal e¤ects of workingconditions and safety at work index14 .Focusing on the relationship between working condition and the probability

of accident, we analyze how characteristics related to working time and involve-ment on the job are correlated with the probability of accident. Related toworking time, our hypothesis is that more demanding jobs in term of workinghours are characterized by higher rates of workplace accidents since tired work-ers may not pay enough attention provoking or not avoiding accidents. On the

12We assume that the stochastic component of safety at work index is normally distributedand that it may estimate with ordinary least square model; see Woolridge (2002) for theassumption of e¢ ciency of two-stage estimators.13Speci�cation (I) includes only working conditions and safety at work index, controlling

for time and country e¤ects. In speci�cation (II), we introduce control variables related topersonal characteristics of the worker and, in speci�cation (III), we also add variables tocontrol for di¤erences in �rm characteristics.14All coe¢ cient estimates are available upon request.

16

contrary, we do not expect any signi�cant e¤ect related to shifts in workingschedule (namely working in the evening or at night or working on Saturdayor on Sunday). Results con�rm our hypothesis that variations in the arrange-ment of working time do not in�uence signi�cantly the probability of workplaceaccident; conversely, working overtime seems to increase of 1.02 percent theprobability of workplace absence. Indeed, positive and statistically signi�cantcorrelation is highlighted between working at least once a month for more than10 hours a day and the probability of workplace accident, either when we con-trol for personal characteristics15 or �rm attributes16 . This result is in line withour expectations and with the evidence by Wilkins (2004), using cross-sectionaldata: too long working hours result in overly tired workers, who experience adrop in promptness and less care in performing their tasks, thus increasing theprobability that an occupational accident occurs. We also �nd that workingon Saturday is positively correlated with the probability of accident: this e¤ectmay be ascribed to a positive correlation between risks and more demandingworking condition. Jobs that required working on Saturday may be a¤ected byhigher risks at work and, consequently, this correlation is substantially spurious.Focusing on the other aspect of working conditions that is related to in-

volvement on the job, we analyze if working at very high speed or with tightdeadlines for at least half of the time a¤ects the probability of accident. Ourexpectation is that higher physical and mental stress due to working with moreinvolvement on the job may increase the probability of occupational accidents.Surprisingly, we �nd that working at very high speed and the necessity to meettight deadlines do not in�uence signi�cantly the workplace accident rate.Although the relationship between working conditions and the probability

of workplace accident is limited to the e¤ect of working overtime, we assumethat a safer workplace reduces the probability of occupational accident. Thisassumption is reasonable if we consider that, in order to reduce workplace ac-cident rates, safety regulation aims at reducing workers exposures to physical,chemical or biological agents on the workplace. Indeed, results con�rm this as-

15Controlling for personal characteristics of the worker, we �nd exclusively a positive andsigni�cantly correlation between gender and workplace accident rates that is in line withgender di¤erences found in several studies on health and work safety (Hersch, 1998; Leeth andRuser, 2006; Leombruni et al., 2009) and that con�rms results of DeLoire and Levy (2004)and Grazier and Sloane (2006) on di¤erences in occupational risk choices between genders.However, we do not �nd any statistically signi�cant e¤ect related to family composition, thatis against �ndings of DeLoire and Levy (2004) and Grazier and Sloane (2006) on the e¤ect offamily structure in in�uencing occupational risk choices. The lack of statistically signi�cantdi¤erences between native and guestworkers is in line with results of Bauer, Million, Rotteand Zimmermann (1998).16Results related to job and �rm characteristics reveal that there are statistically signi�cant

di¤erences among quali�cations. As expected, manual and service workers are a¤ected byhigher workplace accident rates with respect to white collars. Focusing on �rms characteristics,we do not �nd a statistically signi�cant e¤ect related to sectors, but we �nda limited positiveand statistically signi�cant e¤ect of working in big enterprises (more than 50 employees)with respect to small enterprises (less than 10 employees). In accordance with �ndings ofAmuedo-Dorantes (2002) and Hernanz & Toharia (2004), the �contractual e¤ect�, which isreferred to part-time or �xed-term contract, vanishes when we control for personal and �rmcharacteristics.

17

sumption: higher values of the safety at work index are associated with a higherprobability of a workplace accident. Less exposure to physical, chemical or bi-ological agents contributes to a safer workplace and is associated with a lowerprobability of workplace accident. Moreover, the inclusion of a set of controls,namely personal characteristics or �rms attributes, does not alter the negativerelationship between the safety at work index and the probability of accident;indeed, the coe¢ cient remains statistically signi�cant at the 1 percent level.In column 3, we notice that reducing by 1 percent the exposure to dangerousagents decreases the probability of accident by 1.09 percent. Although an acci-dent at work is a random event that may occur also in jobs that do not presentany intrinsic risk at work, less exposure to dangerous agents may contribute toreduce the probability that an accident occurs.However, particular attention should be used when interpreting the result

related to the exposure to dangerous agents as the causal e¤ect of safety atwork on the probability of workplace accident. Indeed, the empirical analysiswe presented so far may produce a biased and inconsistent estimation of thee¤ect of safety at work if exposures to dangerous agents are endogenously de-termined. This may occur when di¤erences in aversion to risk at work mayin�uence workers�career preferences entailing a selection e¤ect. Risk-adverseworkers may look for occupations that minimize risk at work with more strin-gent safety regulation and innovations in work organization practices. Resultsrelated to the relationship between safety at work and probability of workplaceaccident, presented in Table 1, might re�ect spurious correlation due to selectione¤ect.In addition, �rms�behaviour may also induce endogenous selection if the

�rm chooses safety expenditures according to workers�hidden actions in exertingprecaution e¤ort17 . In a Stackelberg context where the �rm does not observeworkers�precaution e¤ort, the problem of single moral hazard may arise: the�rm proposes a contract to workers where wage and �rm�s safety expenditureare enforceable and, then, workers will choose the optimal precaution e¤ort.The �rm may decide to be over-prudent, investing more in safety expenditurethan optimal, to reclaim the moral hazard implication that workers�precautione¤ort is lower than the optimal level18 . We may assume that the safety at workcan be endogenously determined by the �rm�s expenditure aimed at innovatingwork organization practices.The employers have to account for two di¤erent aspects in determining how

much invest for reducing the exposure to dangerous agents. First, employers areobliged to set up measures mandated by national laws and regulations, inspiredby ILO conventions and EU Framework Directive 89/391, to guarantee healthand safety at work standards. Moreover, national governments may introduce

17This assumption is in line with Lanoie (1991) who argues that the probability of accidentis in�uenced either by the precaution e¤ort of workers or by the �rm�s safety expenditure.The workers� precaution level is related to the time spent in risk-preventing activities orthe unpleasantness of these activities. The �rm�s safety expenditure is related to the safetyequipment distributed to workers or investment in safer technology.18Lanoie (1991).

18

appropriate systems of inspection in order to ensure adequate penalties for lawand regulation violations and to induce the employers to provide optimal safetyat work standards. On the other hand, the employers may decide to investany additional money to reduce the expected costs of workplace accidents, es-pecially when they do not observe workers�precaution e¤ort, or for achievingquality management certi�cations (OHSAS 18001 and ISO 9001) with the aimof reducing insurance premiums due to �nance work injury programmes. In-deed, national governments may introduce �scal incentives to induce employersto improve health and well-being at the workplace19 introducing innovations inbehavioral prevention at the workplace, work organization and design and/orin�uences on health determinants.In order to account for the endogeneity of safety at work, we introduce

two types of instruments which capture these two aspects of �rms�investmentchoice in safety. To identify the e¤ect of variations in measures mandated bynational laws and regulations on safety expenditure, we propose an occupationalhealth and safety index which captures the level of government intervention inpromoting safety at work. The occupational health and safety index measuresthe number of rati�cations of ILO conventions in 2000 for EU15 countries andin 2005 for EU27 countries, Turkey, Croatia, Switzerland and Norway. Thegovernment may intervene to induce �rms to exert the optimal level of safetyexpenditure modifying the accident liability system20 . Accident liability a¤ectsthe �rms�incentives to engage in activities that may cause an accident and toinvest in precaution safety expenditure to reduce risk. Moreover, the liabilitysystem acts as an implicit insurer for accidents victims compensating them withthe �nes paid by injurers. In this sense, the liability system moves the economicrisk of an accident from the workers to the �rms, which may be found guilty ofnot implementing safety precautions. Consequently, the identifying assumptionis that more stringent regulation in�uences positively safety at work, inducingthe �rms to exert the optimal level of safety expenditure reducing exposuresto physical, chemical or biological agents on the workplace. On the contrary,safety regulation does not a¤ect either the probability that a worker may su¤erfrom an occupational accident21 or the consequences of an accident in term ofdays o¤ work.The second aspect in investigating how the �rm chooses safety expenditure

concerns the adoption of innovations in behavioral prevention at the workplace,work organization and design and/or in�uences on health determinants. Theseinnovations may ensure the achievement of quality management certi�cationsentailing a reduction of insurance premiums due to �nance work injury pro-grammes. For identifying the adoption of innovations for workplace health pro-motion, we use indexes that capture which work organization practices wereintroduced by the employer to reduce exposures to physical, chemical or bi-ological agents. We introduce two variables which proxy for the adoption of

19The Luxemburg declaration on Workplace Health Promotion.20According to the tort literature, developed widely by Shavell (1983; 1984; 2005 for a

review).21Curington (1988) and Viscusi (1986) for a review.

19

high performance work organization practices and of Taylor�s principles of taskspecialization, respectively. Several articles, such as Gittleman et al. (1998),Osterman (1994), Ramsay et al. (2000), Truss (2000) and Wood (1999), ar-gue that improvements to the e¢ ciency of work processes and the quality ofproducts and services are directly related to the di¤usion of innovations in or-ganisational practices and arrangements. High-performance work organizationis characterized by a series of practices aimed at increasing employee involve-ment, discretionary autonomy and responsibility for quality control. FollowingArundel, Lorenz, Lundvall and Valeyre (2006), we select six binary variablesthat are related to high-performance work organization: responsibility for qual-ity control (73.12 percent of workers interviewed), problem solving activities(79.93 percent), learning new things at work (70.21 percent), discretion in �x-ing order of tasks (61.52 percent), discretion in choosing methods of work (63.60percent) and discretion in setting speed or rate of work (66.19 percent). We per-form a principal component analysis to construct a factor which identi�es, foreach observation, the relative �rm score in terms of employee involvement, dis-cretionary autonomy and responsibility for quality control. The relationshipbetween high performance work organization and safety at work is depicted inFigure 3, where we show variations of safety at work increasing innovations inwork organization practices. The slope of the trend line suggests that inno-vations in work organization for favouring employee involvement are positivelycorrelated with safety at work.However, work organization framework is not identi�ed exclusively by the

adoption of innovative practices implemented to increase employee involvement,discretionary autonomy and responsibility for quality control, but also by prac-tices that are related to automation and tasks specialization. For this reason,inspired by Taylor�s principles of task specialization, we also construct an in-dex which captures the association between the variables related to constrainson work pace, repetitiveness and monotonicity of working tasks. We use fourvariables: one captures if pace of work is dependent on horizontal, hierarchical,norm-based or automatic constraints (75.11 percent of workers interviewed); an-other ones related to monotony of tasks (43.38 percent) and to repetitiveness oftask, either if the task is repeated every minute or less (28.35 percent) or if it isrepeated every 10 minutes or less (46.03 percent). Principal component analysisis used to construct the factor which measures, for each observation, the relative�rm score in term of involvement in Taylor�s principles of task specialization.Figure 4 shows that constrains on work pace, repetitiveness and monotonicityin working tasks are negatively correlated with the safety at work index, asshown by the negative slope of the trend line. The identifying assumption isthat workers involved in �rms which introduce innovations in the work processare a¤ected by better working conditions and, in particular, by higher levelsof safety at work. On the contrary, the adoption of practices characterized byconstrains on work pace, repetitiveness and monotonicity of tasks may inducehigher exposures to physical, chemical or biological agents on workplace. Inci-dentally, we notice that there is no evidence that practices adopted to modifywork organization are correlated with the probability that a worker may su¤er

20

from an occupational accident.Table 2 presents the estimations of the OLS regression of the safety at work

index on the set of control variables and instruments. Results suggest thatthere is a positive relationship between regulation and the safety at work in-dex: where government intervention in ratifying ILO conventions (proxy forthe enforcement of mandatory occupational health and safety regulations) isconsistent, employers are more constrained in de�ning safety expenditure forequipment and technologies. These constrains ensure that employers guaranteehigher reductions of exposure to physical, chemical or biological agents. Thisresult is in line with our identifying assumption: rati�cations of ILO conven-tions, which state the accident liability of the employer, induce �rms to reduceexposures to dangerous agents. Variations in the liability system in�uence theexpected costs that the �rm may support if an occupational accident occurs. Ifthe Court judges that the �rm is found negligent, in the sense that it causedthe accident distributing unsafe equipment or neglecting any investment in safertechnology, the �nes for compensating victims of accidents will be added to thecost of replacing injured workers. On the other hand, more stringent safety reg-ulation may increase the expected cost of being monitored by the occupationalhealth and safety authority. Thus, the �rm may be induced to increase its safetyprecaution expenditure to avoid the payment of a �ne for the insu¢ ciency ofoccupational health services. In addition, innovation in work organization prac-tices has a positive and statistically signi�cant e¤ect on the safety at work,suggesting that, increasing employee involvement, �rms tend to ensure higherlevels of safety on the job. Increasing employee involvement and responsibil-ity, the �rm may drive workers to exert more precaution e¤ort to reduce theirexposures to dangerous agents. Moreover, we may assume that workers em-ployed in �rms that introduce innovations in work organization practices arebetter informed regarding the health and safety risks related to their perfor-mance of your job. More information on risk at work may help workers toavoid exposures to dangerous agents. On the other hand, the involvement injobs characterized by repetitiveness and monotonicity of tasks has a negativeand statistically signi�cant e¤ect on the safety at work index. Constrains onwork pace and repetitiveness of risky tasks have the consequences of reducingsafety at work because workers are not able to avoid being exposed to dangerousagents. The monotonicity of tasks induce a drop promptness and less care inexerting risky tasks that increase workers�exposures to dangerous agents.In Table 3, we report estimates using the instrumental variable probit model,

which accounts for endogeneity of safety at work, column (2), as well as usingthe analogous model in which safety at work index is treated as exogenous, col-umn (1), in order to facilitate comparisons. Before comparing results of theseestimations, we test if safety at work is endogenously determined, as we expect,or whether we can reject this hypothesis. Wald chi-squared test of exogeneitycon�rm our hypothesis suggesting that safety at work is endogenously deter-mined by occupational health and safety regulations and innovations in workorganization practices. Moreover, Amemiya-Lee-Newey minimum chi-squared

21

statistic test of overidentifying restrictions22 is in favour of combining the useof safety regulation and of work organization practices as instruments for con-trolling for endogeneity in safety at work index. Indeed, safety regulation andwork organization practices a¤ect the safety at work index, whereas they arenot correlated with the probability of an occupational injury.The marginal e¤ect of the safety at work index on the probability of work-

place, presented in Table 3, reveals that the e¤ect of exposure to physical, chem-ical or biological agents is drastically increased when accounting for variationsof safety at work related to di¤erences in safety regulation and innovations inwork organization practices. A 1-percent increase in safety at work decreases aworker�s predicted mean probability of accident by 2.7 percent. Controlling forendogeneity, the coe¢ cient of the safety at work index is higher than the one ofthe exogenous regressor, suggesting that the analysis initial underestimated thee¤ect of safety at work on the probability of workplace accident.The economic intuition of the causal e¤ect of the safety at work index on

the probability of a workplace accident can be found in Lanoie (1991). Accord-ing to Lanoie, the �rm�s precaution expenditure contributes with the worker�sprecaution e¤ort in reducing the probability of a workplace accident. In ad-dition, in the Stackelberg context, the �rm�s precaution expenditure in�uencesthe probability of a workplace accident independently from workers�behaviourin term of safety precaution e¤ort: the �rm may decide to be over-prudent toreclaim the moral hazard implication and to achieve the same probability that isensured in the �rst best solution when �rms and workers cooperate to minimizethe probability of a workplace accident.

The duration of absence from work Investigating the e¤ect of workingconditions and safety at work on the duration of absence we intend to identifyif these variables determine also variations in the seriousness of workplace acci-dents. Since the social expected costs of workplace accidents increase with theirseriousness, investigating the consequences of a workplace accident is funda-mental as much as studying why an accident occurs. Excluding fatal accidents(not accounted for in our dataset) we may study the seriousness of a workplaceaccident investigating the duration of absence. Our assumption is that moreserious accidents entail longer spans of absence from work. To investigate theduration of absence, we have to account for the fact that an accident is a rareevent that entails, generally, light consequences in term of days o¤ work dueto an occupational injury. Results of the marginal e¤ects of working conditionsand safety at work are presented in Table 423 , which we structure to show howthe e¤ects of working conditions and safety at work vary according to di¤erentspeci�cations24 .

22The test of overidentifying restrictions is estimated by Newey�s minimum chi-squaredtwo-step estimator, which provides similar estimates of coe¢ cients, as compared to maximumlikelihood estimator.23All coe¢ cient estimates are available upon request.24Speci�cation (I) includes only working conditions and safety at work index, controlling

for time and country e¤ects. In speci�cation (II), we introduce control variables related to

22

Results related to the e¤ect of working time on the duration of absence,controlling for personal characteristics25 and �rms attributes26 , con�rm whatwe �nd analyzing the probability of workplace accident. Working more than10 hours a day has a positive and statistically signi�cant e¤ect on the durationof absence. A drop in promptness and less care in performing tasks, whichcharacterize who works overtime, do not entail higher workplace accident ratesbut also an increase of 15.38 percent of the duration of workplace absence. Inaddition, investigating the e¤ect of working time on the duration of accident, we�nd a positive correlation between working in the evening and on Saturday andthe seriousness of accident; conversely, the e¤ect of working at least once a timeat night is positive and statistically signi�cant. However, these e¤ects may bedue to spurious correlation with di¤erences in risk at work that not controlledfor in our analysis and that are not captured by safety at work index. Focusingon involvement on the job, we expect that more physical and mental stress dueto working with higher involvement on the job may increase the duration ofabsence. Surprisingly, we �nd that working at very high speed has a positiveand statistically signi�cant e¤ect on the duration of absence and that working tomeet tight deadlines does not in�uence signi�cantly the seriousness of workplaceaccidents. However, we do not have enough information for providing a possibleexplanation for this e¤ect; further analyses may explain this ambiguous result.Focusing on the e¤ect of the safety at work index on the duration of ab-

sence, we assume that exposures to physical, chemical or biological agents arepositively correlated with the seriousness of consequences of accidents. Occupa-tional injuries due to excessive exposure to dangerous agents may leave workersinvalid or may entail fatal consequences. Results presented in Table 4 con�rmour hypothesis: reducing the exposure to physical, chemical or biological agentshas a negative e¤ect on the consequences of workplace accidents. If �rms pro-vide increased safety at work, the consequences of a workplace accident will beless serious and the duration of absence will be signi�cantly lower. The mar-ginal e¤ect of a reduction of 1 percent in safety at work reduces the durationof absence of 17.77 percent. The statistically signi�cant e¤ect of the safety at

personal characteristics of the worker and, in speci�cation (III), we also add variables tocontrol for di¤erences in �rm characteristics.25Focusing on personal characteristics, the positive and statistically signi�cant e¤ect of

gender con�rms what argued previously highlighting that gender di¤erences do not concernexclusively the probability of accident but, also, the seriousness of occupational injury, con-�rming results of DeLoire and Levy (2004) and Grazier and Sloane (2006) on di¤erences inoccupational risk choices between genders. Relating to family composition, we �nd that work-ers with at least one child are absent for an accident more days with respect the correspondentworkers without child. The negative e¤ect of being native is in line with results of Bauer,Million, Rotte and Zimmermann (1998) that do not �nd di¤erences in probability of accidentbetween native and guestworkers but argue that the latter are employed in industries whichface higher risks in term of technological safety and, consequently, more serious workplaceaccidents.26Results related to job and �rm characteristics con�rm �ndings of analysis of the prob-

ability of accident: there are statistically signi�cant di¤erences among quali�cations and alimited positive and statistically signi�cant e¤ect of working in big enterprises (more than 50employees) with respect to small enterprises (less than 10 employees).

23

work index is robust to inclusion in the model of a set of controls related to per-sonal characteristics and �rm speci�c characteristics: in all speci�cations, thecoe¢ cient of the safety at work index is statistically signi�cant at the 1 percentlevel.However, as shown in the previous paragraphs, we have to pay particular

attention in interpreting the results related to safety at work as the causal e¤ectof reduction of exposure to dangerous agents on the seriousness of workplaceaccidents. The motivation of our assumption of endogeneity of safety at workin the analysis of the determinants of the seriousness of workplace accidents canbe found in results of safety at work on the probability of workplace accident,which con�rm our hypothesis of endogenous selection of safety expenditure. Weinclude in the analysis variables related to safety regulation and work organiza-tion practices as instruments to control for endogeneity in safety at work sinceneither variations in occupational health and safety regulation or di¤erences inwork organization practices may a¤ect the seriousness of occupational accidentsfor the same reasons discussed previously.Table 5 reports the estimates of negative binomial regression in which safety

at work index is treated as exogenous, column (1), and results of the two-stagetechnique used to account for endogeneity in safety at work, column (2)27 . Re-sults reveal that, controlling for endogeneity, the e¤ect of exposure to physical,chemical or biological agents is drastically increased. Accounting for di¤erencesin safety regulation and in work organization practices, a decrease of exposuresto dangerous agents of 1 percent reduces the duration of absence of 45.5 per-cent28 . Based on the empirical results presented, any attempt at reducing theexposure to physical, chemical or biological agents on workplace, either ratifyingILO conventions, which ensure a coherent enforcement of laws and regulations,or promoting innovation in work organization practices that increase employeeinvolvement, induces, at the same time, a marginal decrease of workplace acci-dent rate and the consequences that occur when workers su¤er an occupationalinjury.These results are particularly important because they show the existence of

a trade-o¤, for the �rm, between the precaution expenditure and the expectedcost of a workplace accident. If a workplace accident occurs and the �rm isfound negligent, the �rm has to pay the cost for replacing injured workers andthe �ne for compensating victims of accidents. Moreover, we may assume thatthe expected cost of a workplace accident increases either on the probabilityor on the duration of absence. Indeed, if the probability of accident and theduration of absence increase, the �rm has to pay an increasing cost either for

27As robustness check of our two-stage technique, we perform also generalized linear modelusing IRLS (maximum quasi-likelihood), introduced by Hardin, Schmiediche and Carroll(2003). This methodology allows accounting for negative binomial distribution of errors andcorrecting for endogeneity but it does not provide marginal e¤ects. Coe¢ cient estimates areavailable upon request.28Results of generalized linear model using IRLS (maximum quasi-likelihood) introduced

by Hardin, Schmiediche and Carroll (2003) con�rm that safety at work has a negative andstatistically signi�cant e¤ect on the duration of absence, also when we control for endogenousselection in safety expenditure.

24

the �ne, which is related to the seriousness of an injury, or for replacing injuredworkers. Cost/bene�t analysis induces the �rm to choose safety expenditurethat minimizes the expected cost of a workplace accident. Consequently, gov-ernment intervention, acting on the expected cost a workplace accident, mayplay a crucial role in inducing the �rm to invest in safety expenditures. Indeed,the government may intervene increasing the monitoring activity of �rms, withthe purpose of checking the adequacy of their investment for supporting occu-pational health service. On the other hand, the government may adjust the�nes that injurers must pay for compensating the victims of accidents. Boththe possible interventions may induce the �rm to increase the optimal safetyprecaution expenditure to reduce the expected cost a workplace accident.

6.2 Discussion

In this section, we focus on estimating how the causal e¤ect of safety at workmay vary according with di¤erences in personal characteristics, �rm attributesand working conditions. In other word, our aim is to de�ne for what job cate-gories a reduction of exposures to physical, chemical or biological agents entailsstronger e¤ect on workplace accident rates29 . Consequently, we intend to per-form disaggregate analyses to compare results related to the e¤ect of safety atwork according to di¤erences in personal characteristics, �rm attributes andworking conditions. Results of disaggregate analyses are presented in Table 6,where we report marginal e¤ects of separate estimations, accounting for endo-geneity of safety at work.Descriptive statistics by gender seem to suggest the existence of asymme-

tries that do not concern exclusively the remunerations but also safety at work.DeLoire and Levy (2004), Grazier and Sloane (2006) and Galizzi and Boden(2003b) argue that di¤erences in labour market by gender are explained by dif-ferent occupational choices. Women are less willing to trade risk for wages thanmale and this implies that, according with literature on the wage compensationdi¤erentials for risk at work, men earn higher wages with respect to women.Descriptive statistics presented in the previous section suggest that males area¤ected by higher workplace accident rate and the empirical analysis con�rmthis hypothesis, since we �nd a positive and statistically signi�cant e¤ect ofgender on the probability of accident. These results and descriptive statistics ofsafety at work by gender (Table A2.4) con�rm hypothesis that there is signi�cantdi¤erences in occupational risk choices by gender. We expect that the disag-gregating analysis by gender highlights that reducing exposures to dangerousagents produces an higher e¤ect in decreasing the probability of workplace acci-dent for males. Results presented in Table 6 con�rm our hypothesis: 1-percentincrease of safety at work standards decreases the probability of accident of5.69 percent for males. Investigating the e¤ect for females, we �nd that safetyat work does not in�uence signi�cantly female workplace accident rates: this

29We limit this discussion to the probability of accident since we do not have enough ob-servations to perform disaggregate analyses of the duration of absence consistently.

25

result con�rm the hypothesis that females�occupational choices are addressedto jobs that do not imply exposures to dangerous agents but that also providelower remunerations, explaining part of gender gap in remunerations.We inspect further how vary the e¤ect of safety at work performing separate

estimations by working quali�cations and �rm characteristics. The aim of theseanalyses is to test the hypothesis that investing in safety at work may producebetter e¤ects for jobs that are characterized by intrinsic risks at work. Followingthis purpose, we perform a disaggregate analysis by occupations, distinguishingbetween high-skill, as white collars and clerks, and low-skill workers, as manualand service workers. Descriptive statistics show the existence of signi�cant dif-ferences between high-skill and low-skill workers related to safety at work: theexposure to dangerous agents is limited for high-skill quali�cations; whereas,manual and service workers are a¤ect by higher risks at work. Our hypothesisis that high-skill workers are employed in occupations that do not required tobe exposed to extremely dangerous agents, as radiations or chemical agents,and they may face only limited risks, as high or low temperatures or noise.Consequently, we expect to �nd that an increase in safety at work standardsreduces workplace accident rates only for low-skill workers that are exposedmore frequently and for a longer period of time to dangerous agents. Resultsof the disaggregated analysis con�rm our hypothesis �nding that safety at workdoes not contribute to reduce the probability of accident for high-skill workers.Increasing of 1 percent safety at work standards reduces the probability of ac-cident of 3.75 percent for manual and service workers. Di¤erences in the e¤ectof the safety at work index are ascribed to the peculiarity of occupations. Gen-erally, high-skill quali�cations are characterized by high safety standards andtheir involvements do not entail to be exposed frequently to physical, chemicalor biological agents. On the contrary, exposures to dangerous agents may beessential to �nalize speci�c tasks for low-skill workers employed especially inindustrial sectors. Consequently, a marginal increase in the safety expenditurefor reducing exposures to dangerous agents has a negative and statistically sig-ni�cant e¤ect on the probability of workplace accident exclusively for low-skillindividuals.The existence of di¤erences in the e¤ect of safety at work between high-skill

and low-skill workers suggests investigating also di¤erences among sectors. In-deed, since manual and service workers are employed especially in industrial andservice sectors, we expect to �nd that the e¤ect of safety at work is stronger inthese sectors with respect to Community Social and Personal Services, whichare characterized by lower exposures to dangerous agents. Descriptive statisticspresented in Table A2.4 sustain our hypothesis since we observe that industrialand service sectors present, on average, lower safety standards, which suggestthat their employees are exposed more frequently to occupational risks; instead,risk at work is extremely limited in Community Social and Personal Services.Results of disaggregate analysis con�rm our hypothesis since we �nd that in-creasing safety at work has no signi�cantly e¤ect on the probability of accidentin Community Social and Personal Services. The existence of intrinsic risks atwork in industrial and service sectors implies that, reducing the exposure to

26

physical, chemical or biological agents of 1 percent, workplace accident ratesdecrease of 3.66 percent. Results of separate analyses by sector contributes toprovide more robustness to the hypothesis that investing in safety at work hasa causal e¤ect exclusively where there is an intrinsic risks that characterized anoccupation or a sector. Consequently, since industrial and service sectors maypresent high level of exposures to dangerous agents, as, for example, exposuresto radiation and chemical agents in Manufacturing or noise and vibration inConstruction, which may not be eliminated completely. In these sectors, an in-crease in safety at work standards, with more stringent occupational health andsafety regulation or with incentives to workplace health promotion with inno-vations in work organization practices, may reduce signi�cantly the probabilitythat an accident occurs.A disaggregate analysis by size of enterprise is useful to identify if the e¤ect

of safety at work is stronger when intrinsic risks at work are higher, which oc-curs in medium and big enterprises, or when, due to limited size of enterprise,investment in safety at work is lower. Descriptive statistics show that work-ers in medium and big enterprises are exposed more frequently to dangerousagents. Indeed, speci�c tasks, which entail high level of exposures to dangerousagents, may be completed exclusively when enterprise is characterized by givenworkforce or business dimensions. On the other hand, small enterprises maynot invest too much in safety at work for many reasons: less union strengthin bargaining safety at work standards, less serious implications of workers�moral hazard in precaution e¤ort, simpli�ed national laws and regulations, etc.Results of the disaggregate analysis suggest that investing in safety at workreduce signi�cantly the probability of accident exclusively in medium and bigenterprises. An increase of 1 percent in safety at work standards reduces theprobability of accident of 3.15 percent. In small enterprises, the probability ofaccident is not a¤ected by variation in safety at work because intrinsic risk atwork is limited. Indeed, small enterprises are presented, in particular, in low-risk sectors and they employ a limited number of manual and service workers.Dangerous activities may not be completed in enterprises with limited numberof employees; consequently, policies that induce small enterprises to invest insafety at work may not reduce workplace accident rates but may have harmfulconsequences for their business.Finally, we investigate how the e¤ect of safety at work may vary according

with working conditions. For this purpose, we construct two indexes that iden-tify if a job is working time and involvement demanding. We assume that ajob is working time demanding if it requires working at least once a month onSaturday, on Sunday, in the evening, in the night or more than 10 hours a day.Our hypothesis, consistent with descriptive statistics presented in Table A2.4,is that overly tired workers, who experience a drop in promptness and less carein performing their tasks, are more exposed to risks, which characterize theirjobs. Consequently, increasing safety at work standards may mitigate positivecorrelation between working time demanding jobs and accident at work. Re-sults of disaggregate analysis con�rm our hypothesis showing that the e¤ect ofsafety at work on the probability of accident for more demanding jobs in term of

27

working time is negative and statistically signi�cant. Increasing safety at workstandards reduce of 3.28 percent the probability that an occupational accidentoccurs when a job is working time demanding. On the contrary, safety at workdoes not in�uence the probability of accident when workers are employed injobs that are not working time demanding, highlighting the existence of a posi-tive correlation between more demanding working conditions and risk at work.Relating to involvements in the job, we de�ne a dummy variable that identi�esif workers are requested to work at very high speed or to meet tight deadlinesfor more than half of the time. According with results of disaggregate analysisby working time, we expect that the e¤ect of safety at work is stronger whenjobs are more involvement demanding. Our hypothesis is that there is a posi-tive correlation between involvement in the job and risk at work, as shown indescriptive statistics and for the same reasons discussed above. Results con�rmthat for workers employed in more involvement demanding jobs investing insafety at work reduce the probability of accident of 3.79 percent. The e¤ect ofsafety at work is not statistically signi�cant on the probability of accident whenwe consider jobs that are not involvement demanding. These results con�rmcorrelation between more demanding working condition and risk at work.Results presented in this section help us to de�ne more clearly the patterns of

risk at work and to identify how the causal e¤ect of safety at work may vary ac-cording with occupational risk choices. This analysis allows drawing importantimplications for government policy evaluation. Indeed, government intervention,either introducing more restrictive occupational health and safety regulations orpromoting innovations in work organization practices, should pursue the goal ofincreasing safety at work standards especially in sectors and occupations that arecharacterized by dangerous and intrinsic risks. In order to reach this purpose,identifying variations of the e¤ect of safety at work by �rm and job characteris-tics ensures well-de�ned and targeted policies. Results suggest that a marginaldecrease of exposures to dangerous agents, investing in safer technology or intro-ducing innovations in work organization practices, has stronger e¤ects on theprobability of accident for high-risk jobs. Thus, government intervention hasto be �nalized in identifying job categories that are characterized by intrinsicrisks at work to de�ne targeted policies; otherwise, policies that introduce morestringent occupational and health regulations or promote innovation in work or-ganization practices may entail increasing social costs, when they are addressedalso to low-risk jobs, which do no present signi�cant exposures to dangerousagents. Performing a cost/bene�t analysis, government may �nd that policiesfocused also on sectors and occupations that do not su¤er for intrinsic risks atwork may entail costs of implementing more stringent safety regulations or ofpromoting innovations in work organization are not compensate by signi�cativebene�ts in term of reduction of workplace accident rates.

6.3 Sensitivity analyses

EU15 versus new entrants The existence of di¤erences in the prob-ability and the duration of absence among European countries, as shown in

28

descriptive statistics, suggests to test the sensitivity of our results for EU15countries and for countries that entered in the European Union later, namelyin 2004 and 2007.Table 7 shows di¤erences in the e¤ects of the safety at work on the probabil-

ity of a workplace injury between EU15 countries and new entrants. Comparingthe results of the e¤ect of the probability of workplace accident, we �nd thatsafety at work has a statistically signi�cant and negative e¤ect for workers em-ployed in EU15 countries; whereas, this e¤ect vanishes for new entrants. Ananalysis of the duration of absence cannot be performed since data does notprovide enough information on days o¤ work due to occupational accidents innew entrants countries. Results on the probability of accident and the lack ofinformation for performing estimations of the duration of absence con�rm thehypothesis that the analysis should be limited to EU15 countries. These resultscan be explained by di¤erences in knowledge relative to workplace accidents andtheir consequences. Workers employed in new entrants may be induced to mis-leadingly report information on a workplace accident. More detailed analysis ofthe e¤ect of working conditions and safety at work on the probability of acci-dent and on the duration of absence for new entrants requires a more numerousdataset.For testing the robustness of our results, we also perform estimations on the

probability of accident and duration of absence excluding each time only onecountry from the model of interest. The purpose of this test is to prove that thee¤ect of safety at work is consistent also if we exclude one country. Estimationscon�rm our hypothesis, proving the validity of our results.30

7 Conclusion

Results suggest that increasing safety at work, reducing exposures to physical,chemical or biological agents, has a negative e¤ect either on the probabilityof accident or the duration of absence, con�rming empirically what assumedby Lanoie (1991). An increase of 1 percent of safety at work standards mayreduce the probability of accident of 2.7 percent and the duration of absence45.5 percent31 . Since workplace accident is essentially a random event thatmay occur also in low-risk jobs, it is crucial to minimize risk at work withpolicies that induce the employers to guarantee higher standards of occupationalhealth services. These policies may contribute to prevent especially workplaceaccidents that occur because workers are not well informed on potential risksand well trained for avoiding them or because the employers are not committed

30Coe¢ cient estimates available upon request.31Performing a sensitivity analysis between EU15 countries and countries that entered in

the European Union in 2004 and 2007, we �nd that results are consistent considering workersemployed in EU15 countries. In new entrants, the safety at work does not a¤ect the prob-ability and the duration of a workplace accident. Possible explanation concerns di¤erencesin knowledge relative to workplace accidents and their consequences that may be induced toreport misleadingly an accident at work. These results suggest of limiting our analysis onlyto workers interviewed in EU 15 countries.

29

in replacing the dangerous tasks by the non-dangerous or the less dangerous oradopting new and safer technological productive processes.Our analysis suggests that government interventions should be �nalized in

in�uencing �rm�s behaviour acting on two di¤erent aspects: mandating, withnational laws and regulations, to set up measures to guarantee health standardsand safety at work or promoting innovations in work organization practices with�scal incentives for �rms that are awarded with quality management certi�ca-tions (OHSAS 18001 and ISO 9001). Results con�rm our hypothesis �nding thateither promoting innovations in work organization practices or imposing morestringent safety regulation, as suggest by Shavell (1983; 1984, 2005), governmentmay induce the employers to reduce exposures to physical, chemical or biologicalagents. In addition, results suggest that promoting innovations in work orga-nization may provide higher e¤ect with respect to more stringent occupationalhealth and safety regulation, in particular, when workplace health promotionis �nalized to replace work organization practices inspired by Taylor�s princi-ples of task specialization with high-performance work organization practices.The existence of stringent safety regulation in many European countries mayexplained why the e¤ect of ratifying ILO conventions, which is good proxy foran increase in occupational health and safety regulation, is lower with respect tothe e¤ect of innovation in work organization practices. Intervening with �scalincentives for �rms that introduce innovations in work organization practicesmay in�uence �rms� cost/bene�t analysis related to occupational health andsafety: investing in safety at work may become more convenient with respect toexpected costs of workplace accidents.Analyzing separately the e¤ect of safety at work on the probability of ac-

cident by personal characteristics, �rm attributes and working condition, wenotice that the causal e¤ect of reducing exposures to dangerous agents is consis-tent for job categories that are characterized for intrinsic risks at work. Whereoccupational risks are limited and workers are almost never exposed to dan-gerous agents, increasing safety at work standards does not reduce workplaceaccident rates. This result suggests that government interventions for imposingmore stringent safety regulation may produce negative e¤ects for low-risk jobs.For instance, higher requirements in term of occupational health services mayentail excessive costs for small enterprise, which may undermine their business.Consequently, government should intervene identifying risky jobs and, then,adopting policies in accordance with the nature of the activities and the size ofthe enterprise. A possible solution should be found introducing policies in favourof innovations in work organization practices that may produce more desirablee¤ects: in low-risk jobs, they may entail an increase of labour productivity, asargued by Bloom and Van Reenen (2010); in high-risk jobs, they may also con-tribute in improving occupational health and safety and in reducing workplaceaccident rate.

30

References

[1] Aigner, D.J., and Cain, G.G., 1977, Statistical Theories of Discriminationin the Labor Market. Industrial and Labor Relations Review 30(2):175�87

[2] Amuedo-Dorantes C., 2002,Work Safety in the Context of Temporary Em-ployment: The Spanish Experience. Industrial and Labor Relations Review,55(2), 262-284.

[3] Arundel, A., Lorenz, E., Lundvall, B.A. and Valeyre, A., 2006, The Orga-nization of Work and Innovative Performance: A comparison of the EU-15, DRUID Working Papers 06-14, DRUID, Copenhagen Business School,Department of Industrial Economics and Strategy/Aalborg University, De-partment of Business Studies.

[4] Bauer, T.K., Million, A., Rotte, R. and Zimmermann, K.F., 1998, Immi-gration Labor and Workplace Safety, IZA Discussion Papers 16, Institutefor the Study of Labor (IZA).

[5] Baugher, J. E Roberts, T., 1999, Perceptions and Worry about Hazardsat Work: Unions, Contract Maintenance, and Job Control in the U.S.Petrochemical Industry. Industrial relations, 38(4), 522-541.

[6] Berger, M. and Gabriel, P., 1991, Risk Aversion and the Earnings of U.S.Immigrants and Natives. Applied Economics, 23, 311�18

[7] Bloom, N. and Van Reenen, J., 2010, Human Resource Management andProductivity, NBER Working Papers 16019, National Bureau of EconomicResearch, Inc.

[8] Boone, J. and van Ours, J.C., 2002, Cyclical Fluctuations in WorkplaceAccidents, CEPR Discussion Papers 3655, C.E.P.R. Discussion Papers.

[9] Boone, J. and van Ours, J.C., 2006, Are recessions good for workplacesafety?, Journal of Health Economic, Elsevier, vol. 25(6), pages 1069-1093,November.

[10] Cameron, C.A. and Trivedi, P.K., 1986, Econometric Models Based onCount Data: Comparisons and Applications of Some Estimators and Tests,Journal of Applied Econometrics, John Wiley and Sons, Ltd., vol. 1(1),pages 29-53, January.

[11] Crichton, S., Stillman, S. and Hyslop, D., 2005, Returning to Work from In-jury: Longitudinal Evidence on Employment and Earnings, IZA DiscussionPapers 1857, Institute for the Study of Labor (IZA).

[12] Curington, W.P., 1986, Safety Regulation and Workplace Injuries, SouthernEconomic Journal, Vol. 53, No. 1 (Jul., 1986), pp. 51-72

31

[13] DeLeire, T. and Levy, E., 2004, Worker Sorting and the Risk of Deathon the Job, Journal of Labor Economics, University of Chicago Press, vol.22(4), pages 925-954, October.

[14] DeLeire, T. and Levy, H., 2001, Gender, Occupation Choice and the Riskof Death at Work. NBER Working Papers, 8574.

[15] Fenn, P. and Ashby, S., 2004, Workplace Risk, Establishment Size andUnion Density, British Journal of Industrial Relations, Blackwell Publish-ers Ltd/London School of Economics, vol. 42(3), pages 461-480, 09.

[16] Galizzi, M. and Boden, L.I., 2003a, The return to work of injured workers:evidence from matched unemployment insurance and workers�compensationdata, Labour Economics, Elsevier, vol. 10(3), pages 311-337, June.

[17] Galizzi, M. and Boden, L.I., 2003b, Income Losses of Women and MenInjured at Work, Journal of Human Resources, XXXVIII(3):722-757 (2003);

[18] Garcia-serrano, C., Hernanz, V. and Toharia, L., 2008, Mind the gap,please! The e¤ect of temporary help agencies on the consequences of workaccidents, ISER working papers 2008-28, Institute for Social and EconomicResearch.

[19] Gittleman, M., Horrigan, M. and Joyce, M., 1998, Flexible� WorkplacePractices: Evidence from a nationally Representative Survey, Industrialand Labour Relations Review, Vol. 52, No. 1.

[20] Grazier, S. and Sloane, P.J, 2006, Accident Risk, Gender, Family Statusand Occupational Choice in the UK, IZA Discussion Papers 2302, Institutefor the Study of Labor (IZA).

[21] Guadalupe, M., 2003, The hidden costs of �xed term contracts: the impacton work accidents, Labour Economics, Elsevier, vol. 10(3), pages 339-357,June.

[22] Hämäläinen, R. M., K. Husman, K. Räsänen, P. Westerholm and JormaRantanen, 2001, Survey of the Quality and E¤ectiveness of OccupationalHealth Services in the European Union and Norway and Switzerland.Finnish Institute of Occupational Health, Helsinki.

[23] Hamermesh, D., 1998, Immigration and the Quality of Jobs. NBER Work-ing Papers, 6195.

[24] Hardin, J.W., Schmiediche, H., Carroll, R.J., 2003, Instrumental variables,bootstrapping, and generalized linear models, Stata Journal, StataCorp LP,vol. 3(4), pages 351-360, December.

[25] Hernanz, V. and Toharia, L., 2004, Do temporary contracts increase workaccidents? A microeconometric comparison between Italy and Spain, Work-ing Papers 2004-02, FEDEA.

32

[26] Hersch, J., 1998, Compensating Di¤erentials for Gender-Speci�c Job InjuryRisks, American Economic Review, June 1998, v. 88, iss. 3, pp. 598-627

[27] Hoskins, A.B., 2005, Occupational Injuries, Illnesses, and Fatalities amongWomen, Monthly Labor Review, October 2005, v. 128, iss. 10, pp. 31-37.

[28] Lanoie, P., 1991, Occupational Safety And Health: A Problem Of DoubleMoral Hazard, Journal of Risk and Insurance, 58 (1991) : 80:100.

[29] Leeth, J. and Ruser, J., 2006, Safety Segregation: The Importance of Gen-der, Race, and Ethnicity on Workplace Risk. Journal of Economic Inequal-ity, 4(2), 123-52.

[30] Leombruni R., Bena, A. and Giraudo, M, 2009, Lavoro e salute. Primeevidenze da un sistema informativo integrato sugli infortuni da lavoro. LRRwp no. 86, 2009

[31] Litwin, A.S., 2000, Trade Unions and Industrial Injury in Great Britain,CEP Discussion Papers n�0468, Centre for Economic Performance, LSE.

[32] OECD, 2007, Society at a glance: OECD Social Indicator, 2006 Edition

[33] Osterman, P., 1994, How Common is Workplace Transformation and WhoAdopts It ? Industrial and Labor Relations Review, Vol. 47.

[34] Ramsay, H., Scholarios, D. and Harley, B., 2000, Employees and High-Performance Work Systems, Testing inside the Black Box, British Journalof Industrial Relation, 38: 4, pp. 501-531.

[35] Reville, R.T. and Schoeni, R.F., 2001, Disability from Injuries at Work:The E¤ects on Earnings and Employment, Working Papers 01-08, RANDCorporation Publications Department.

[36] Ruhm, C.J., 2000, Are Recessions Good for Your Health?, Quarterly Jour-nal of Economics, 2000 - MIT Press - May 2000, Vol. 115, No.2, Pages617-650

[37] Scha¤ner S. and Kluve, J., 2007, Gender Wage Di¤erentials and the Occu-pational Injury Risk - Evidence from Germany and the US, Ruhr EconomicPapers 0028, Rheinisch-Westfälisches Institut für Wirtschaftsforschung,Ruhr-Universität Bochum, Universität Dortmund, Universität Duisburg-Essen.

[38] Shavell, S., 1983, Liability for Harm Versus Regulation of Safety, NBERWorking Papers 1218, National Bureau of Economic Research, Inc.

[39] Shavell, S., 1984, A Model of the Optimal Use of Liability and Safety Reg-ulation, RAND Journal of Economics, The RAND Corporation, vol. 15(2),pages 271-280, Summer.

33

[40] Shavell, S., 2005, Liability for Accidents, NBER Working Papers 11781,National Bureau of Economic Research, Inc.

[41] Truss, C., 2001, Complexities and Controversies in Linking HRM with Or-ganisational Outcomes, Journal of Management Studies, 38: 8, pp. 1120-1149.

[42] Viscusi, W.K, 1986, The Impact of Occupational Safety and HealthRegula-tion, 1973-83, University of Chicago - George G. Stigler Center for Study ofEconomy and State 40, Chicago - Center for Study of Economy and State.

[43] Viscusi, K.W. and Aldy, J., 2003, The Value of a Statistical Life: A CriticalReview of Market Estimates

[44] Williamson, A., Bohle, P.; Quinlan, M. e Kennedy, D., 2009, Short Tripsand Long Days: Safety and Health in Short-Haul Trucking. Industrial andLabor Relations Review, 62(3), 415-429.

[45] Wood, S., 1999, Getting the Measure of the Transformed High-PerformanceOrganisation, British Journal of Industrial Relations, 37: 3, pp. 391-417.

[46] Wooldridge, J., 2002, Econometric Analysis of Cross Section and PanelData, MIT Press

A Appendix

A.1 ILO Conventions concerning occupational health andsafety

The rati�cation of ILO Conventions concerning occupational health and safetycan be used as a valid proxy to measure the e¤ort of European governmentstowards avoiding workplace accidents over the last years.Among the ILO Conventions concerning occupational health and safety, the

Occupational Safety and Health Convention (c.155), the Occupational HealthServices Convention (c.161) and the Protocol of 2002 to the Occupational Safetyand Health Convention (p.155) gain in importance for their general purpose. In-deed, the Convention 155 prescribes that each member state has to formulate,implement, enforce and periodically review a coherent set of laws and regulationsconcerning occupational safety, occupational health and the working environ-ment. Moreover, an appropriate system of inspection has to be implementedin order to ensure adequate penalties for violations of the laws and regulationsimplemented. The Convention 161 advances these arguments mandating thateach member state undertakes the development of progressively adequate andappropriate occupational health services for workers who su¤er from speci�c oc-cupational risks. The Protocol 155 ensures that the competent authority shallestablish and periodically review requirements and procedures for the recording

34

and the noti�cation of occupational accidents, occupational diseases and dan-gerous occurrences, commuting accidents and suspected cases of occupationaldiseases.In addition, International Labour Organization introduces the Prevention

of Major Industrial Accidents Convention (c.174) to prevent major accidentsinvolving hazardous substances and to limit the consequences of such accidents.The competent authority has to provide special policies to guarantee occupa-tional health and safety by establishing a system for the identi�cation of ma-jor hazard installations, consulting the major organizations of employers andworkers, based on a list of hazardous substances and their respective thresh-old quantities. The Promotional Framework for Occupational Safety and HealthConvention (c.187) sets the goal of promoting continuous improvement towardsprogressively achieving a safe and healthy working environment to prevent occu-pational injuries, diseases and deaths. This goal should be reached by developingnational policies, systems and programs inspired by the promotional frameworkfor occupational safety and health set out by ILO Conventions.Most of the ILO Conventions related to occupational health and safety con-

cern the enforcement of laws and regulations to ensure that any worker doesnot su¤er an occupation accident due to a speci�c occupational risk. The Radi-ation Protection Convention (c.115) prescribes that all appropriate steps shallbe taken to ensure e¤ective protection, to restrict the exposure of workers toionising radiations to the lowest practicable level and to avoid any unnecessaryexposure. The Benzene Convention (c.136) provides the safety precautions toguarantee occupational hygiene and the technical measures to ensure e¤ectiveprotection of workers exposed to benzene or to products containing benzene.Moreover, the convention prescribes the replacement of benzene, or productscontaining benzene, with harmless or less harmful substitute products. Accord-ing to the Asbestos Convention (c.162), each member state has to prescribethe measures to prevent and protect workers against occupational exposures toasbestos and to review periodically the relative laws and regulations in the lightof technical progress and of advances in scienti�c knowledge. The ChemicalConvention (c.170) prescribes that each member state has to provide a coher-ent enforcement of laws and regulations to regulate the use of chemicals onthe workplace and that, on safety and health grounds, the competent authorityshall restrict the use of certain hazardous chemicals or require advance noti�ca-tion and authorization before such chemicals are used. The prohibition of usingwhite lead, sulphate of lead and all products containing these pigments in theinternal painting of buildings, is stressed by theWhite Lead (Painting) Conven-tion (c.13); exceptions are justi�ed where the use of these products is considerednecessary, as in the case of railway stations or industrial establishments. TheOccupational Cancer Convention (c.139) prescribes that each member statehas to periodically list the carcinogenic substances and agents to which workersshould not be exposed or which use should be subject to authorization or con-trol. Furthermore, the convention calls for the replacement by non-carcinogenicor less harmful substances of those carcinogenic substances and agents to whichworkers may be exposed. The Working Environment (Air Pollution, Noise and

35

Vibration) Convention (c.148) and the Hygiene (Commerce and O¢ ces) Con-vention (c.120) are implemented to prevent and control occupational hazardsrelated to working environment, such as air pollution, noise and vibration, andto hygiene in the workplace, respectively. The Maximum Weight Convention(c.127) prescribes that no worker shall be required to engage in the manualtransport of a load which, by reason of its weight, is likely to jeopardise healthor safety. The Guarding of Machinery Convention (c.119), on the other hand,sets out the conditions to deal with all machinery, either power-driven or man-ual, which present a risk of injury to the worker.Finally, ILO Conventions concerning occupational health and safety call for

the enforcement of laws and regulations to avoid occupational accidents in spe-ci�c labour markets. The Safety Provisions (Building) Convention (c.62) andThe Safety and Health in Construction Convention (c.167) concern the adop-tion of adequate provisions to preserve safety and health in construction andbuilding work where the probability of serious accident is higher. The Safetyand Health in Mines Convention (c.176) provides the adequate prescription toprevent any fatalities, injuries or ill health a¤ecting workers which might arisefrom mining operations. The Underground Work (Women) Convention (c.45)prohibits that any female, whatever her age, be employed on underground workin any mine. Lastly, the Safety and Health in Agriculture Convention (c.184)prescribes that each member state shall provide a coherent national policy onsafety and health in agriculture.

36

37

Table A1.1: Estimated marginal effects of safety at work and working conditions from the probability of workplace accident equation

European country Legislation concerning OHS Authority responsible for legislation

Austria Health and Safety at Work Act (1995) Federal Ministry of Labour, Health and Social Affairs; Federal Ministry of Social Security and Generations

Belgium Well-being of Employees at Work (1996) Federal Ministry of Employment and Labour

Denmark Work Environment Act (1975) Ministry of Labour and Ministry of Industry and Trade

Finland Occupational Health Service Act (1978); Labour Safety Act (which includes the content of OHS) and the Supervision of Labour Protection Act (which includes the inspection of employers’ organization of OHS services)

Ministry of Social Affairs and Health

France Civil service code, district codes or public health code (1995)

Ministry of Labour

Germany Occupational Health and Safety Act (1974) Federal Ministry of Labour and Social Affairs

Greece Presidential Decree 17/96 for the integration of European directives with national law

Ministry of Employment and Social Security

Ireland Safety, Health and Welfare at Work Act (1989) and General Application Regulation

Ministry of Labour

Italy Health and Safety at Work Acts 547/1955, Law 277/1991, Law 626/1994 and Law 242/1996; D.Lgs.359/99

Ministry of Labour and Social Affairs

Luxembourg Law of Transposition of Framework Directive 89/391/EEC (1995)

Ministry of Health

Netherlands Dutch Working Act (1998) Ministry of Social Affairs and Employment

Norway Work Environmental Act (1977) Ministry of Local Government and Regional Development; Labour Inspectorate

Portugal Law 26/1994; Law 133/1999 Transposition of European Union directives

-

Spain Law 31/1995, Prevention of Risks at Work; General Health Law and Industry Law

Ministry of Labour and Social Affairs; Ministry of Labour and Social Affairs

Sweden - Ministry of Industry Employment and Communications

Switzerland Law 832.30 (1983), (revised 1997); Law 822.213 (1993); Regulation No.3 of the Labour Act; Accident Insurance Law

Federal Coordination Commission for Work Safety; State Secretariat for Economic Affairs; Swiss Accident Insurance Fund

United Kingdom Health and Safety Act (1974); Management of Health and Safety at Work Regs (1999); Safety representatives and Safety Committes Regulations (1978); Consultation with Employees Regulations (1995)

Source: Hämäläinen, R. M., K. Husman, K. Räsänen, P. Westerholm and Jorma Rantanen (2001)

38

A.2 Data description

Table A2.1: Variables used in the analysis Obs. Mean Std. Dev. Min Max Dependent Variables Workplace accident rate 16043 0,0395 0,195 0 1 Cumulative duration of absence 634 26,09 45,20 1 365 Explanatory Variables Safety at work Index 16043 0,0001 1,0005 -4,9178 0,8390 Working conditions Nights worked 16043 0,1641 0,3703 0 1 Evenings worked 16043 0,4319 0,4954 0 1 Sundays worked 16043 0,2316 0,4218 0 1 Saturdays worked 16043 0,4822 0,4997 0 1 More than 10 hours worked 16043 0,3352 0,4721 0 1 Working at high speed 16043 0,5049 0,5000 0 1 Working to tight dealines 16043 0,5026 0,5000 0 1 Personal Characteristics Male 16043 0,5466 0,4978 0 1 Age 16043 37,62 1115,10 18 65 Age squared 16043 1539,32 881,20 324 4225 Native 16043 0,9557 0,2058 0 1 Child 16043 0,4398 0,4964 0 1 Household 16043 0,6364 0,4811 0 1 Firm Characteristics Manual workers 16043 0,2998 0,4582 0 1 Service workers 16043 0,4495 0,4975 0 1 White collars 16043 0,5828 0,4931 0 1 Industry 16043 0,3679 0,4823 0 1 Service sector 16043 0,4772 0,4995 0 1 Community Social and Personal Services 16043 0,1549 0,3618 0 1 Small enterprise (less than 10 employees) 16043 0,3492 0,4767 0 1 Medium enterprise (10-50 employees) 16043 0,3443 0,4751 0 1 Big enterprise (more than 50 employees) 16043 0,3065 0,4610 0 1 Part-time 16043 0,1565 0,3634 0 1 Indefinite contract 16043 0,8607 0,3463 0 1 Countries Variables Belgium 16043 0,0552 0,2283 0 1 Czech Rep. 16043 0,0135 0,1155 0 1 Denmark 16043 0,0489 0,2156 0 1 Germany 16043 0,0642 0,2451 0 1 Estonia 16043 0,0111 0,1048 0 1 Greece 16043 0,0330 0,1786 0 1 Spain 16043 0,0599 0,2373 0 1 France 16043 0,0645 0,2456 0 1 Ireland 16043 0,0461 0,2096 0 1 Italy 16043 0,0486 0,2149 0 1 Cyprus 16043 0,0081 0,0897 0 1 Latvia 16043 0,0191 0,1370 0 1 Lithuania 16043 0,0130 0,1134 0 1 Luxembourg 16043 0,0219 0,1463 0 1 Hungary 16043 0,0186 0,1352 0 1 Malta 16043 0,0080 0,0890 0 1 Netherlands 16043 0,0592 0,2359 0 1 Austria 16043 0,0545 0,2270 0 1 Poland 16043 0,0148 0,1209 0 1

39

Portugal 16043 0,0585 0,2346 0 1 Slovenia 16043 0,0134 0,1150 0 1 Slovakia 16043 0,0186 0,1352 0 1 Finland 16043 0,0499 0,2178 0 1 Sweden 16043 0,0504 0,2188 0 1 UK 16043 0,0496 0,2170 0 1 Bulgaria 16043 0,0155 0,1234 0 1 Croatia 16043 0,0170 0,1293 0 1 Romania 16043 0,0170 0,1293 0 1 Turkey 16043 0,0034 0,0579 0 1 Norway 16043 0,0194 0,1381 0 1 Switzerland 16043 0,0252 0,1569 0 1 Time Variables Year 2000 16043 0,4763 0,4995 0 1 Instrumantal variables OHS regulation 16043 4,6387 2,1059 1 10 High-performance work organization 16043 0,0000 1,0002 -2,2851 1,0485 Taylor’s principles of task specialization 16043 -0,0010 1,0001 -1,3470 1,7094

Table A2.2: Descriptive statistics Workplace accident rate Duration of absence Personal Characteristics Gender Female 0,0213 25,52 Male 0,0546 26,27 Age class 18-30 0,0435 20,28 31-45 0,0389 30,64 more than 45 0,0358 26,51 Child No 0,0409 23,94 Yes 0,0377 29,06 Household No 0,0307 23,65 Yes 0,0446 27,05 Native No 0,0464 43,06 Yes 0,0392 25,15 Firm-specific characteristics Occupation Manual workers 0,0721 24,90 Service workers 0,0308 29,61 White collar 0,0162 20,40 Sector of the main paid job Industry 0,0556 26,30 Service sectors 0,0314 26,98 Community Social and Personal Services 0,0266 21,79 Size of enterprise Small enterprise (less than 10 employees) 0,0327 27,14 Medium enterprise (10-50 employees) 0,0407 24,94 Big enterprise (more than 50 employees) 0,0460 26,38 Type of contract Full time 0,0425 26,53 Part time 0,0235 21,75 Fixed-term/temporary contract 0,0385 23,91 Indefinite contract 0,0397 26,43 Country

40

Belgium 0,0678 26,68 Czech Rep. 0,0046 10,00 Denmark 0,0408 23,31 Germany 0,0728 15,55 Estonia 0,0112 3,50 Greece 0,0132 52,86 Spain 0,0416 25,45 France 0,0600 30,65 Ireland 0,0365 19,56 Italy 0,0257 25,90 Cyprus 0,0308 37,25 Latvia 0,0163 42,40 Lithuania 0,0191 31,50 Luxembourg 0,0598 47,95 Hungary 0,0167 16,40 Malta 0,0391 62,40 Netherlands 0,0242 30,35 Austria 0,0561 19,08 Poland 0,0042 100,00 Portugal 0,0309 60,83 Slovenia 0,0558 18,25 Slovakia 0,0201 50,17 Finland 0,0749 19,85 Sweden 0,0358 22,03 UK 0,0365 13,97 Bulgaria 0,0000 0,00 Croatia 0,0183 9,80 Romania 0,0037 14,00 Turkey 0,1111 6,67 Norway 0,0160 35,00 Switzerland 0,0222 28,56 Year 2000 0,0602 24,53 2005 0,0207 30,20 Working Conditions Working at night No 0,0366 26,74 Yes 0,0543 23,84 Working in the evening No 0,0349 27,40 Yes 0,0456 24,77 Working on Sunday No 0,0385 26,29 Yes 0,0428 25,49 Working on Saturday No 0,0341 24,70 Yes 0,0454 27,20 More than 10 hours per day No 0,0330 26,97 Yes 0,0524 24,98 Working at very high speed No 0,0337 28,81 Yes 0,0452 24,09 Working to tight deadline No 0,0312 29,90 Yes 0,0477 23,62

41

Table A2.3: Descriptive statistics of exposures to physical, chemical or biological agents

Vibrations Noise High

temperatures Low

temperatures Breathing fumes

Handling chemicals Radiation

All of the time 6,11 6,65 3,35 1,94 5,66 2,62 0,84 Almost all of the time 5,08 5,72 3,11 1,98 4,66 2,68 0,80 Around 3/4 of the time 2,62 3,72 2,34 1,89 2,57 1,37 0,46 Around half of the time 3,75 5,22 5,55 5,68 3,43 2,40 0,88 Around 1/4 of the time 7,13 9,29 9,07 9,19 6,53 5,60 2,01 Almost never 12,88 17,24 17,90 18,29 12,73 13,81 8,03 Never 62,42 52,16 58,69 61,03 64,40 71,52 86,98

Table A2.4: Descriptive statistics of exposures to physical, chemical or biological agents by personal and firm characteristics and working conditions

Safety at work By gender

Male -0,2621 Female 0,3161 By occupations

White collars and clerks 0,4277 Manual and service workers -0,3050 By sector

Industry and service sectors -0,0444 Community Social and Personal Services 0,2427 By size of enterprise

Small enterprise (less than 10 employees) 0,1542 Medium (10-50 employees) and Big enterprise (more than 50 employees) -0,0826 By working conditions

Working time demanding Yes -0,0470

No 0,0981 Involvement demanding

Yes -0,1591 No 0,2701

42

Figure 1: Workplace accident rate by safety at work index

Figure 2: Duration of absence and safety at work index

0.1

.2.3

.4.5

.6W

orkp

lace

acc

iden

t rat

e

-5 -4 -3 -2 -1 0 1Safety at work

05

1015

2025

30D

urat

ion

of a

bsen

ce

-5 -4 -3 -2 -1 0 1Safety at work

43

Figure 3: Safety at work index and high-performance work organization

Figure 4: Safety at work index and Taylor’s principles of task specialization

-2.5

-2-1

.5-1

-.5

0.5

1S

afet

y at

wor

k

-.4 -.3 -.2 -.1 0 .1 .2 .3 .4High-performance work organization

-2-1

01

2S

afet

y at

wor

k

-.4 -.2 0 .2 .4 .6Taylor's principle of work organization

44

Table 1: Estimated marginal effects of safety at work and working conditions from the probability of workplace accident equation

Specification Specification Specification (I) (II) (III) Safety at work -0,0164 *** -0,0146 *** -0,0109 ***

(0,0011)

(0,0011)

(0,0012)

Working condition

Nights worked 0,0017

0,0002

-0,0018

(0,0037)

(0,0035)

(0,0032)

Evenings worked 0,0010

0,0009

0,0023

(0,0030)

(0,0029)

(0,0028)

Sundays worked -0,0015

-0,0005

0,0006

(0,0038)

(0,0034)

(0,0033)

Saturdays worked 0,0074 ** 0,0074 *** 0,0069 **

(0,0029)

(0,0029)

(0,0028)

More than 10 hours worked 0,0113 *** 0,0087 *** 0,0102 ***

(0,0030)

(0,0029)

(0,0029)

Working at high speed -0,0016

-0,0008

-0,0011

(0,0028)

(0,0028)

(0,0026)

Working to tight deadlines 0,0026

0,0023

0,0030

(0,0028)

(0,0028)

(0,0027)

Personal Characteristics No Yes Yes Firm-specific Characteristics No No Yes Time Fixed-Effect Yes Yes Yes Country Fixed-Effect Yes Yes Yes Log likelihood -2350,1435 -2336,9976 -2304,494 N° observations 15795 15795 15795

Note: * Significant at 0,100; ** Significant at 0,50; *** Significant at 0,01, Standard errors are in parentheses

Table 2: Estimated marginal effects of safety at work and working conditions from the duration of absence equation

Specification Specification Specification

(I) (II) (III) Safety at work -0,3166 *** -0,2684 *** -0,1778 ***

(0,0354)

(0,0346)

(0,0293)

Working condition

Nights worked 0,0135 -0,0617 -0,1246 **

(0,0978) (0,0826) (0,0566)

Evenings worked 0,0795 0,1152 0,1270 **

(0,0848) (0,0825) (0,0691)

Sundays worked -0,0105 -0,0055 0,1341

(0,0930) (0,088) (0,0902)

Saturdays worked 0,3235 *** 0,31651 *** 0,2258 ***

(0,0885)

(0,0835)

(0,0677)

More than 10 hours worked 0,1365 ** 0,1353 ** 0,1539 ***

(0,0736) (0,0749) (0,0638)

Working at high speed -0,1744 ant -0,1562 *** -0,1534 ***

(0,0734) (0,0694) (0,0556)

Working to tight deadlines -0,0896 -0,1005 -0,0457

(0,0693)

(0,0670)

(0,0549)

Personal Characteristics No Yes Yes Firm-specific Characteristics No No Yes Time Fixed-Effect Yes Yes Yes Country Fixed-Effect Yes Yes Yes Log likelihood -5221,938 -5212,3639 -5182,2153 N° observations 16043 16043 16043

Note: * Significant at 0,100; ** Significant at 0,50; *** Significant at 0,01, Standard errors are in parentheses

45

Table 3: Estimated coefficients of instrumental variables from the safety at work equation Safety at work

Safety regulation 0,0466 *** (0,0015)

High-performance work organization 0,0329 *** (0,0047)

Taylor’s principles of task specialization -0,1317 *** (0,0045)

Personal Characteristics Yes Firm-specific Characteristics Yes Working conditions Yes Country Effect Yes Year Effect Yes Adj R-squared 0,3530 N° observations 16043

Note: * Significant at 0,100; ** Significant at 0,50; *** Significant at 0,01, Standard errors are in parentheses

Table 4: Estimated marginal effects of safety at work and working conditions from the probability of workplace accident equation – Exogenous versus Endogenous safety at work index specifications

Exogenous safety at work

Endogenous safety at work

Safety at work -0,0109 *** -0,0270 ***

(0,0012)

(0,0104)

Working condition

Nights worked -0,0018

-0,0043

(0,0032)

(0,0036)

Evenings worked 0,0023

0,0021

(0,0028)

(0,0030)

Sundays worked 0,0006

0,0008

(0,0033)

(0,0035)

Saturdays worked 0,0069 ** 0,0058 **

(0,0028)

(0,0030)

More than 10 hours worked 0,0102 *** 0,0108 ***

(0,0029)

(0,0031)

Working at high speed -0,0011

-0,0046

(0,0026)

(0,0035)

Working to tight deadlines 0,0030

0,0001

(0,0027)

(0,0033)

Personal Characteristics Yes Yes Firm-specific Characteristics Yes No Time Fixed-Effect Yes Yes Country Fixed-Effect Yes Yes Log likelihood -2304,494 -21270,834 N° observations 15795 15795

Note: * Significant at 0.100; ** Significant at 0.50; *** Significant at 0.01. Standard errors are in parentheses. Wald test of exogeneity (/athrho= 0): chi2(1) = 3.40; Prob > chi2 = 0.0653. Test of overidentifying restrictions (Amemiya-Lee-Newey minimum chi-sq statistic): 0.858 Chi-sq(2); P-value = 0.6512

46

Table 5: Estimated marginal effects of safety at work and working conditions from the duration of absence equation – Exogenous versus Endogenous safety at work index specifications

Exogenous safety at work

Endogenous safety at work

Safety at work -0,1778 *** -0,4550 ***

(0,0293)

(0,0346)

Working condition

Nights worked -0,1246 ** -0,1502 **

(0,0566) (0,0634)

Evenings worked 0,1270 ** 0,1520 **

(0,0691) (0,0707)

Sundays worked 0,1341 0,1155

(0,0902) (0,0962)

Saturdays worked 0,2258 *** 0,2065 ***

(0,0677)

(0,0732)

More than 10 hours worked 0,1539 *** 0,1869 ***

(0,0638) (0,0711)

Working at high speed -0,1534 *** -0,1866 ***

(0,0556) (0,0760)

Working to tight deadlines -0,0457 -0,0674

(0,0549)

(0,0734)

Personal Characteristics Yes Yes Firm-specific Characteristics Yes Yes Time Fixed-Effect Yes Yes Country Fixed-Effect Yes Yes Log likelihood -5182,2153 -5194.5858 N° observations 16043 16043

Note: * Significant at 0,100; ** Significant at 0,50; *** Significant at 0,01, Standard errors are in parentheses

Table 6: Estimated marginal effects of safety at work from the probability of workplace accident equation – Disaggregate analyses

Safety at work Log

likelihood N°

observations By gender

Male -0,0569 *** -12845,891 8571

(0,0189)

Female -0,0003

-6986,1065 6453

(0,0112)

By occupations

White collars and clerks -0,0067

-5591,886 5707

(0,0096)

Manual and service workers -0,0376 *** -13704,139 9104

(0,0174)

By sector

Industry and service sectors -0,0367 ***

-17.981 13145

(0,0136)

Community Social and Personal Services 0,0059

-2493,0703 2086

(0,0209)

By size of enterprise

Small enterprise (less than 10 employees) -0,0070

-6309,6775 5037

(0,0168)

Medium (10-50 employees) and Big enterprise (more than 50 employees)

-0,0315 *** -14255,921 10259

(0,0128) By working conditions

Working time demanding

Yes -0,0328 ***

-14719,086 10672

(0,0133)

No -0,0123

-6063,6599 4795

(0,0141)

47

Involvement demanding

Yes -0,0379 ***

-14150,85 9811 (0,0160)

No -0,0120

-6263,0062 5552 (0,0116)

Note: * Significant at 0,100; ** Significant at 0,50; *** Significant at 0,01, Standard errors are in parentheses

Table 7: Estimated marginal effects of safety at work and working conditions from the probability of workplace accident equation – EU15 vs. New entrants

EU15 New entrants Safety at work -0,0365 *** 0,0244

(0,0130)

(0,0607)

Working condition

Nights worked -0,0053 -0,0031

(0,0047) (0,0066)

Evenings worked 0,0035 0,0085

(0,0039) (0,0140)

Sundays worked -0,0016 0,0095

(0,0045) (0,0105)

Saturdays worked 0,0069 * 0,0032

(0,0039)

(0,0080)

More than 10 hours worked 0,0136 *** 0,0068

(0,0042) (0,0108) Working at high speed -0,0061

0,0100

(0,0045) (0,0211) Working to tight deadlines 0,0004 -0,0005

(0,0043)

(0,0067)

Personal Characteristics Yes Yes Firm-specific Characteristics Yes Yes Time Fixed-Effect Yes Yes Country Fixed-Effect Yes Yes Log likelihood -16628,68 -3153,8429 Observations 16043 2384

Note: * Significant at 0,100; ** Significant at 0,50; *** Significant at 0,01, Standard errors are in parentheses