There is an enormous amount of academic work studying which construction site conditions are prevalent when the accident occurs. This literature is reactive in nature as it is aimed to explain probable risk conditions on sites involving accident event reports (Camino López et al., 2011; C.-W. Cheng et al., 2012; C.-W. Cheng et al., 2010; Conte et al., 2011; J. W. Hinze & Teizer, 2011; Liao & Perng, 2008; López Arquillos et al., 2012; McVittie et al., 1997; Nishikitani & Yano, 2008). Another important research line proposes methods to conduct in depth analysis of occupational accidents, although only large companies seem are using these methods (Jørgensen, 2016). Among them we have the “Storybuilder” (Bellamy et al., 2007) in the framework of Worm project, where over 20.000 serious accidents were analysed, jointly with their barriers, and 64 types of hazards were summarised (Ale, 2006; Bellamy, 2010).
Other line of research within health and safety (H&S) has adopted a more preventive approach as it tries to avoid accidents through risk assessment based mainly on site conditions (Memarian & Mitropoulos, 2013; Cambraia et al., 2010; Wu et al., 2010; Yang et al., 2012). However, there is a limited number of studies trying to model the relationship between risk conditions on site and the likelihood of accidents. One example is the study by J. Hinze et al. (2013) where the authors analysed which leading indicators can be utilized to assess safety performance (e.g. works, supervisors, managers, owners, and designers, all them at jobsite). Another example, Sparer & Dennerlein (2013) developed a software to identify sites with high accident risks using leading indicators to measure work conditions that can affect the risks.
More attention to construction tasks is necessary as studies at task level only represent the 2.28% of all research on H&S in the construction industry (Zhou et al., 2015). There is a lack of field risk exposition measurements on sites (Swuste et al., 2012) because most of the research tend to be epidemiological and mainly focused on accidents.
On another level, H&S has been identified as one of the issues that are relevant for company results, competitive advantage and management performance (Teo & Ling, 2006; Argilés-Bosch et al., 2014; Rechenthin, 2004). Most of the research connecting these issues in construction industry has been theoretical. Following Chalos (1992) theoretical framework of cost of safety (COS model), many scholars have analysed, two well-known dimensions of H&S cost, prevention and accidents costs (C.-W. Cheng et al., 2010; C.-W. Cheng, Lin, et al., 2010; Feng et al., 2015; Gurcanli et al., 2015; Harshbarger, 2001; HSE, 2015; Ibarrondo-Dávila et al., 2015; Labelle, 2000). Despite the cost of H&S has been studied in some depth, it is surprising the absence of empirical works trying to understand which is the relationship between those costs and the economic benefit of the firms. The cost of occupational accidents is increasing, and therefore raising safety levels would generate a win-win solution for all parts, including the employees, the firm as well as society (EUROSTAT, 2004; Gavious, Mizrahi, Shani, & Minchuk, 2009; Jørgensen, 2016). Despite this reasonable relationship, to the best of our knowledge, there is just a one single published paper dealing with this issue (Argilés-Bosch et al., 2014). Particularly, these authors found a linear negative relationship between accident rates and firm financial performance one year ahead. Although, this is a very interesting result it may fail to explain a potential more complex relationship between those variables. Our research question is directed to this point: Is it possible a non-linear relationship between benefits and accidents. Can firms support accident costs without affecting their financial performance? As we will see, our empirical research provides evidences regarding the relationships between risk-accidents on the one hand and between accidents-firm performance on the other hand.
There is an extensive body of theoretical models connecting risk with accidents. The traditional “bowtie metaphor” from Visser (1998) can be considered as the one of the first theoretical model of the process of H&S management and the consequences of risks. Following this metaphor, existing uncontrollable hazards converge in the so called “central event” which in turn may evolve and diverge into different risks causing potential damages or accidents. The first role of management in such a scenarios is to interpose some barriers to prevent the conversion of hazards into risks, and the second is to build some protections to prevent risks becoming accidents.
The problematic issue here is that, despite the proposals of these models, the relationship risk-accident is not always contingent because not all risk expositions finally end in accidents and, alternatively, some good safety systems can have some accident. In fact, most of the risk-exposition does not end in accident, or in other words, we do not have as many accidents as it might be expected probably because workers are able to control most of risk situations (Sundström-Frisk, 1985). As Khanzode et al. (2012) concluded, there is a gap in the literature because the study of risk assessment is disconnected of the causality model of accidents that have been proposed. Although this gap is important at the theoretical level, it is more salient at the level of empirical and field research. There is a clear scarcity of exposition measures on site, and there is also a need to identify which main events are related with accidents (Swuste et al., 2012). Only a limited number of empirical researches connect risk conditions on site with accidents results. Most of these studies are contextualised in the assessment of the effectiveness of specific and very focused safety campaigns (Hale et al., 2010; Kines et al., 2010; Laitinen & Päivärinta, 2010; Laitinen & Ruohomäki, 1996; Spangenberg et al., 2002) or in the assessment of the effectiveness of implemented management systems (Yoon et al., 2013). We have only found one study that considers the level of hazard of a project as a moderator variable over the relationship between accidents rates and the total cost of accidents (Feng et al., 2015). From the literature it can be concluded that there is a need to generate more knowledge about the empirical interaction between risk conditions and accidents. Consequently, the first hypothesis is aimed to test whether or not an increasing relationship exists between risk levels and accident rates when you consider a relatively long spam of years:
H1. High levels of risk on sites have a positive effect on accidents rates.
Managers seem to ignore the economic consequences of unsafe practices in the workplace (Harshbarger, 2001). Since they don’t have accurate estimates of the economic impact of accidents, they cannot consequently assess which is the economic contribution of the function H&S management, and consequently, there seem to be low awareness of its strategic value (Labelle, 2000). In order to keep companies being competitive, many contractors try to control short term total operation costs by executing only basic safety measures during construction project implementation (C.-W. Cheng, et al., 2010). It is important to prioritise safety, but other demands as finance, client, production and deadlines change these priorities (Jørgensen, 2016). Due to the uncertainty of overall H&S costs and the daily demands, the companies do not apply several protection measures, reducing direct H&S costs (i.e. cost of prevention and protection measures) but ignoring the amount of indirect H&S costs (i.e. cost of accidents) they will have to afford when accidents occur. H&S costs, both direct and indirect, do exist and they are high (Ibarrondo-Dávila et al., 2015). Companies do not have appropriate accounting systems to estimate these costs. Although, there is not a standard method for estimating direct costs. Gurcanli et al. (2015) have calculated that the cost of safety measures represents a 1.9% of a residential building project budget.
Focusing on accidents, they affect costs in many ways at the level of the individuals, the company or the Public Administration: healthcare costs, production losses, delays, lost working days, penalties, etc. (HSE, 2015). In his historical work, Heinrich (1927, 1941) classified accident costs into direct and indirect, concluding that even when the amount of direct cost of accidents are important, indirect costs can be even much higher than direct ones. Direct costs of accidents refers to the expenses directly related with injuries and fatalities, while indirect cost of accidents include productivity losses, disruption in schedules, delays in completion dates, fines and legal expenses, damage in organization image, etc. (Ibarrondo-Dávila et al., 2015; Feng et al., 2015). The complex relationship between the occurrence of accidents and their cost cannot be explained by one single variable (López-Alonso et al., 2013). In this line, (Feng et al., 2015) described up to 13 possible components of indirect costs of accidents based on previous literature review. Moreover, these authors concluded that workplace accident costs of building projects are influenced by accident rates, project hazard level, project size, company size and the involvement of sub-contractors. Feng et al. (2015) have reported that total accidents cost of building projects accounts for 0.25% of total contract sum. Another interesting finding of this research is that the positive effect of accidents rates on total accident costs is moderated by project hazard level, as we have already mentioned above. Hallowell (2011) reported that construction injuries impact firm financial performance by increasing total cost up to 15% in new non-residential projects. Therefore, in the literature there is some results supporting the hypothesis that accidents may have a negative impact on economic results of construction companies via increasing organizational costs.
There is a complex structure of accidents costs with different types and not obvious relationships among them, which makes more complex to stablish a relationship between accident and economic performance of the firms. We have found only one recent empirical study where the incidence of accidents on firm financial performance is estimated. Using panel data for a period of six years (from 1998 to 2003) and a sample of 99 construction firms, Argilés-Bosch et al. (2014) have found evidence of a negative relationship between accidents in one year and firm financial performance one year ahead. In our research we want to check whether or not this hypothesis and results hold for the period of time we are considering and our sample. Therefore, our second and third hypothesis is stated as:
H2. Work accidents have a negative effect on firm financial performance.
H3. There is a quadratic relationship between accidents and firm financial performance. For low levels of accidents there is an increasing positive effect on firm financial performance while for large levels of accidents there is an increasing negative effect on firm performance.
Empirical design
According to our hypotheses we propose to test three models. Following we define all variables and the specification for the models.
In order to regress the accident rate (ACCRATE) on the site risk Index (SRI), we have first obtained the ACCRATE as the proportion of accidents over of total number of workers in the firm. SRI is the average of 10 risk variables, measured on site with CONSRAT (Construction site risk assessment tool) we have created to measure site risk in a broader research project. More levels of this variable means more risk on site (low compliance of H&S plan, bad conditions of order, tidiness or access, low or inefficient protections, high levels of falls of height or other risks, etc.). To regress ACCRATE on SRI, we use a control variable that represents the organizational design complexity (ORGDES). This variable tries to capture different elements related with site organizational structure and planning, related previously with H&S on site (Fang, Huang, et al., 2004; Hinze et al., 2013; López-Alonso, et al., 2013; Manu el al., 2013; Swuste et al., 2012; Yung, 2009). ORGDES is computed with the information from CONSRAT and we have used it in previous empirical studies. A higher value of ORGDES means more complexity on site design (more companies, more contractors, more levels of subcontracting, more works, etc.). We expect a positive relationship between ORGDES and ACCRATE.
In our second and third model we propose to connect firm financial performance as the dependent variable with accident rates as the independent variable using two different specifications, linear and quadratic. In both models we use ROA as a firm profitability measure because is the most common used indicator of firm financial performance in the literature (Tan & Wang, 2010) and adequate for samples such as ours which are usually composed by non-listed firms (Argilés-Bosch et al., 2014). ROA “is the ratio of income before leverage to total assets in percent, indicating firm profitability before leverage relative to its size” (Argilés-Bosch et al., 2014). Following these authors and some others (Bandyopadhyay et al., 2010; Q. Cheng, 2005), past profitability is partially explained by past firm management and characteristics, as it also an explicative factor in part of future profitability. Therefore, we also assume that present firm profitability depends on its profitability in previous period and we control for it in our model. Additionally, Argilés-Bosch et al. (2014) pointed out that profitability also depends on management decisions in the same year which impact organizational efficiency. A common variable used in business literature to capture efficiency is asset turnover (the ratio of firm sales to total assets) (Fairfield & Yohn, 2001). As Argilés-Bosch et al. (2014) did, we have included in our model a variable to control for current firm efficiency change in the period due to present management decisions. This variable (CHASSETURN) is calculated as the difference between a company asset turnover in a given year and in the previous year, relative to asset turnover in the previous year. i.e. is the perceptual change in asset turnover in a given period. We expect a positive relationship with ROA.
All these economic variables are also influenced by market conditions and political decisions among others factors. However, the specific consideration of all these variables are beyond the scope of our study. As in Argilés-Bosch et al. (2014), all those potential explicative factors of ROA, which are not explicitly included in our models, will show their estimated impact either in the lagged ROA variable, the dummies of year variables and also in the error term of our models which, in turn, will determined the percentage of the variance that our models explain.
According to our hypothesis 1, 2 and 3, their corresponding model specifications are as follow:
H1: ACCRATEi,t = ß0 + ß2 . SRIi,t + ß3 . ORGDESi,t + εi,t (Model 1)
H2: ROAi,t = ß0 + ß1 . ROAt-1 + ß2 . ACCRATEt + ß3 . CHASSETURN + εi,t (Model 2)
H3: ROAi,t = ß0 + ß1 . ROAi,t-1 + ß2 . ACCRATEi,t + ß3 . (ACCRATEi,t)2 + ß4 .CHASSETURN + εi,t (Model 3)
Variables, sample and data collection
Variables
As we have mentioned SRI and ORGDES were obtained through CONSRAT. Our tool serves to record responses and assessments on sites related with H&S as well as organizational issues. It contains 97 questions or items to conform 10 variables related to risk, and 10 related to different organizational variables. Each item of CONSRAT has a rating with specific criteria and scoring to allow aggregation with others items. The aggregation rules among item to build risk as well organizational variables are object of previous research (see Paper 1, chapter 1).
SRI ranges from 0 (representing no risk) to 1 (signalling maximum risk). It is composed by: Health and safety plan accomplishment (RV1); General condition of the construction work (RV2) (enclosures, circulations, tidiness, cleanliness and illumination, signalling and the electrical system); General conditions of the collective protections (RV3); Specific conditions of phase access (RV4); Falls from height assessment (RV5); Up to 11 other risks identification (RV6); Process evaluation (RV7); Collective (RV8) and Individual (RV9) protections assessment at main stage; Auxiliary resources and machinery adequacy and assessment (RV10).
ORGDES ranges from 0 (representing no complexity or resources) to 1 (signalling maximum complexity or resources). It is composed by: Internal organization structure (OV1) and Job planning and design (OV2). It has been built using a panel of experts, using different weights of the two variables and of all their internal items. See Appendix A-3 for further information. .
Sample
We collected information regarding live conditions visiting a total of 957 sites, mostly building constructions, in Mallorca (Balearic Islands), pertaining to a total of 627 companies. All sites in our sample were selected following random criteria and were collected from 2004 to 2009.
We crossed our initial sample with data on accident rates from Balearic Islands Labour Authority and SABI data base of Bureau van Dijk. As a consequence, we rejected companies with headquarters not located at this region or with not available information in SABI. Table 1 describes the final sample taking in account these issues.
Our sample is composed by all types of firms, including firms with no accidents and firms with any type of accidents (minor injuries, serious injuries and fatalities) without prioritizing by severity. We consider indicative of some preventive problem any kind of accident independently of its seriousness (Jørgensen, 2011). Prioritizing the severity could cause a loss of preventive information of a company (Bellamy, 2015; Khanzode et al., 2012)
Collinearity does not seem to affect the estimations of our models as it can be deduced by the low levels of Pearson correlations between the independent variables of our models. In the context of our Model (1), we have found a significant correlation between our independent variable SRI and the control variable ORGDES (-0.1456, p<0.01), suggesting a possible problem of collinearity between these variables. After conducting a test for collinearity using variance inflation factor (VIF command in STATA) we rejected that collinearity is a problem in our model as we obtained all values to be close to 1, which are far below from 10.
In relation to the other two models (2 and 3), a significant correlation between ROA (profitability) and CHASSETURN (efficiency) is found, (0.0610, p<0.05). As Argilés-Bosch et al. (2014) explained, though this correlation is positive as one would expect, its low magnitude might be caused because CHASSETURN contains information just for one year (it is a year change rate) and ROA includes information regarding managerial decisions of both current and past years. We have also found significant negative correlations between ROAt-1 and both ACCRATE (-0.0513, p<0.1) and ACCRATE2 (-0.0725, p<0.01) but they are very small. This suggests that those firms with less past profitability are associated with a higher current accident rate. The results after using VIF test did not signal any problem with multicollinearity as all variables VIF values were below 3.5.
To test our hypotheses, we have estimated our models using several methods for panel data estimation (pooled, fixed effects and random effects estimators). Additionally, we have added some dummy year variables to explore for time specific effects. In order to control the possible existence of heteroskedasticity, all models were estimated using robust methods. Additionally, we have run Hausman test (Hsiao, 2014) in order to verify which estimation method, random versus fixed effects, better adjusted our data. For the model (1) the Hausman test does not reject the null hypothesis of no correlation between individual effects and the independent variable, therefore individual effects are uncorrelated with the regressors and the random effects estimator is consistent and efficient. For the models (2) and (3) Hausman test rejects the null hypothesis of no correlation between individual effects and the independent variable which implies that individual effects are correlated with the regressors and the fixed effects estimator seems to be more consistent and efficient than the random estimator.
Table 5 shows the results for model (1). The columns 1 to 3 contain the baseline model including only the control variables. The columns 4 to 6 add the explanatory variable, and finally the columns 7 to 9 show the complete model that includes the year dummy variables. As we can see, the coefficients of our explanatory variable, SRI, are positive and significant p<0.05 for pooled and random effects estimators, either with or without dummies of years, while in fixed effects estimators the coefficient is not significant without adding the year dummy variables and significant at p<0.1 with dummy variables. Control variables have in the complete model a similar behaviour compared to the other models. Hausman test (9.03) is not significant at p<0.1 with seven degrees of freedom, it does not reject the null hypothesis of no correlation between individual effects and the explanatory variable, which means that random effects estimator are more efficient and consistent than the fixed effects ones.
----------------------------------------------
INSERT TABLE 5 ABOUT HERE
----------------------------------------------
Following a similar estimation strategy, Table 6 shows the estimations results for model (2) using the three different estimation methods. As we have stated above, this model (2) propose to test the relationship between accident rate (ACCRATE) and firm financial performance (ROA) on construction sector. Columns 1 to 3 of results contain the baseline model including only the control variables. Columns 4 to 6 add the explanatory variable, and finally columns 7 to9 show the complete model that includes the year dummy variables.
Regarding our explanatory variable, ACCRATE, it can be seen in Table 6 that we have obtained a positive and significant effect on ROA for pooled and random effects estimations (p>0.05) and for fixed effects estimation (p>0.01) (model specifications without year dummies, see result columns 4 - 6). These significant results for ACCRATE are not maintained when we add the year dummies into our model specification (see result columns 7 to 9). For that complete model specification, we can see that years 2008 and 2009 have a negative and significant effect on ROA (p>0.01) for all the three estimation methods. Although results are not reported here, we did not find any significant effect between accidents rate on previous year and ROA on current year.
----------------------------------------------
INSERT TABLE 6 ABOUT HERE
----------------------------------------------
As we have mentioned previously, for this model (2) the Hausman test (176.00) is significant at p<0.01 with eight degrees of freedom, therefore fixed effects estimators are more efficient and consistent than random effects ones. As it can be seen in Table 7, the estimations in model (2) presented a significant goodness of fit (p<0.01)
Similarly to model (1) estimation, we observe that 2008 and 2009 dummies have a significant negative effect on ROA (p<0.01) in model (2). However, when year dummies enter into the model (2) specification, the significance of our explanatory variable disappears. Following the same reasons as in model (1) estimation strategy, we have also tested the whole model (2) for a restricted period of years of our panel data (i.e. eliminating 2008 and 2009). Although we don’t report here the results of this restricted model, the explanatory variable was not significant at p<0.1 in any model specification, neither the year dummies.
Table 7 reports model (3) estimation results without the baseline model as is the same that in model (2). For pooled and random effects estimations in the two specifications with and without year dummy variables, our control variables ROAt-1 and CHASSETURN are significant (p<0.01) and positive as expected, however for fixed effects estimate, ROAt-1 loses the significance. Regarding our explanatory variable, ACCRATE, the linear term is positive and significant (p<0.01) and the quadratic term is negative and significant (p<0.01) for all the estimation models without the year dummies (see results columns 1 to 3 in Table 8). These results are consistent with our hypothesis H3. When we introduce into the model (3) all year dummies, we can see that the estimated results for the explanatory variable change. Specifically, the significance of the linear effect of ACCRATE on ROA only holds for the random effects estimations at p<0.1, while the quadratic term remains significant at p<0.05 for the random effect estimation and at p<0.01 for the pooled and fixed effect estimations. Notwithstanding, all the estimated effects for ACCRATE and ACCRATE2 are in the direction we propose in our hypothesis H3. As we have mentioned before, for this model (3) Hausman test (175.62) is significant at p<0.01 with nine degrees of freedom, what suggest that the fixed effects estimations are more efficient and consistent than random effects ones.
----------------------------------------------
INSERT TABLE 7 ABOUT HERE
----------------------------------------------
Table 5. Incidence of risk on site (SRI) on accident rate in the same year. (2004-2009)
|
Variables |
Pooled (1) |
Random effects (2) |
Fixed effects (3) |
Pooled (4) |
Random effects (5) |
Fixed effects (6) |
Pooled (7) |
Random effects (8) |
Fixed effects (9) |
|
SRIt |
|
|
|
0.215** |
0.205** |
0.247 |
0.219** |
0.208** |
0.244* |
|
ORGDES |
0.002** |
0.002** |
0.002 |
0.002** |
0.002*** |
0.003* |
0.002*** |
0.002*** |
0.002 |
|
Intercept |
0.135*** |
0.126*** |
0.132*** |
-0.039 |
-0.040 |
-0.088 |
-0.015 |
-0.014 |
0.0045 |
|
Year 2005 |
|
|
|
|
|
|
0.0245 |
0.024 |
0.0374 |
|
Year 2006 |
|
|
|
|
|
|
-0.0455 |
-0.054 |
-0.127** |
|
Year 2007 |
|
|
|
|
|
|
-0.0405 |
-0.0514 |
-.137** |
|
Year 2008 |
|
|
|
|
|
|
-0.106** |
-0.113** |
-0.184*** |
|
Year 2009 |
|
|
|
|
|
|
-0.116** |
-0.128** |
-0.283 |
|
Goodness of fit |
F( 1, 360) = 4.77** |
Wald chi2(1) = 6.40** |
F(1,251) = 1.60 |
F( 1, 353) = 5.40*** |
Wald chi2(2) = 12.18*** |
F(1,249) = 2.64* |
F(7, 347) = 2.94*** |
Wald chi2(7)= 22.36*** |
F(7,249) = 2.44** |
|
R-squared overall |
0.008 |
0.011 |
0.008 |
0.020 |
0.020 |
0.020 |
0.048 |
0.047 |
0.0414 |
|
No. of observ. |
355 |
355 |
355 |
355 |
355 |
355 |
355 |
355 |
355 |
SRI is index of risk on site. ORGDES represents the organizational design complexity of site.
* Significance level: p < 0.1.
** Significance level: p < 0.05.
*** Significance level: p < 0.01.
Table 6. Incidence of accidents rate (ACCRATE) and control variables on return of assets (ROA) in the same year.
|
Variables |
Pooled (1) |
Random effects (2) |
Fixed effects (3) |
Pooled (4) |
Random effects (5) |
Fixed effects (6) |
Pooled (7) |
Random effects (8) |
Fixed effects (9) |
|
ROAt-1 |
0.351*** |
0.327*** |
0.063 |
0.349*** |
0.321*** |
0.059 |
0.337*** |
0. 322*** |
0.018 |
|
ACCRATE |
|
|
|
4.596** |
4.800** |
7.972*** |
2.0317 |
2.0359 |
1.994 |
|
Chasseturn |
0.181*** |
0.182*** |
0.170*** |
0.183*** |
0.183*** |
0.166*** |
0.1837*** |
0. 183*** |
0.1594*** |
|
Intercept |
0.443 |
0.495 |
1.440*** |
-0.307 |
-.2847301 |
0.143 |
2.1597* |
2.2312** |
4.724*** |
|
Year 2005 |
|
|
|
|
|
|
1.3747** |
1.351 |
0.408 |
|
Year 2006 |
|
|
|
|
|
|
-0.796 |
-0.796 |
-1.361 |
|
Year 2007 |
|
|
|
|
|
|
-1.110 |
-1.142 |
-2.412* |
|
Year 2008 |
|
|
|
|
|
|
-6.329*** |
-6.390*** |
-8.429*** |
|
Year 2009 |
|
|
|
|
|
|
-5.9419*** |
-6.091*** |
-10.014*** |
|
Goodness of fit |
F(2,1236) =20.29*** |
Wald chi2(2)= 46.85*** |
F(2,273)=22.23*** |
F(3,1235) = 14.41*** |
Waldchi2(3)=51.93*** |
F(3,273) =17.87*** |
F(8,1230)= 8.52*** |
b |
F(8,273)=12.21*** |
|
R-squared overall |
0.093 |
0.093 |
0.058 |
0.096 |
0.096 |
0.040 |
0.119 |
0.119 |
0.045 |
|
No. of observ. |
1239 |
1239 |
1239 |
1239 |
1239 |
1239 |
1239 |
1239 |
1044 |
ROA is the percent of return on assets; ACCRATE is the proportional of workers injured with respect to the total firm workers; ACCRATE2 is the quadratic term of ACCRATE; CHASSETURN is the perceptual change rate in asset turnover in a given period.
** Significance level: p < 0.05.
*** Significance level: p < 0.01.
b Not reported
Table 7. Incidence of accidents rate (ACCRATE and ACCRATE2, quadratic) and control variables on return of assets
|
Variables |
Pooled (1) |
Random effects (2) |
Fixed effects (3) |
Pooled (4) |
Random effects (5) |
Fixed effects (6) |
|
ROAt-1 |
0.350*** |
0.321*** |
0.060 |
0.339*** |
0.322*** |
0.020 |
|
ACCRATE |
10.698*** |
11.069*** |
16.161*** |
6.099 |
6.101* |
6.158 |
|
ACCRATE2 |
-4.850*** |
-4.953*** |
-6.287*** |
-3.165* |
-3.156** |
-3.102* |
|
Chasseturn |
0.1845*** |
0.184*** |
0.165*** |
0.185*** |
0.184*** |
0.159*** |
|
Intercept |
-0.936 |
-0.928 |
-0.715 |
1.660 |
1.738 |
4.207*** |
|
Year 2005 |
|
|
|
1.400 |
1.375 |
0.441 |
|
Year 2006 |
|
|
|
-0.780 |
-0.781 |
-1.354 |
|
Year 2007 |
|
|
|
-1.073 |
-1.107 |
-2.364 |
|
Year 2008 |
|
|
|
-6.220*** |
-6.2873*** |
-8.327*** |
|
Year 2009 |
|
|
|
-5.673*** |
-5.835*** |
-9.764*** |
|
Goodness of fit |
F(4,1234)= 10.78*** |
Wald chi2(4) =52.80*** |
F(4,273) = 13.84*** |
F(9,1229) =7.87*** |
(b) |
F(9,273)=11.04*** |
|
R-squared overall |
0.099 |
0.099 |
0.039 |
0.120 |
0.120 |
0.046 |
|
No. of observ. |
1239 |
1239 |
1239 |
1239 |
1239 |
1239 |
ROA is the percent of return on assets; ACCRATE is the proportional of workers injured with respect to the total firm workers; ACCRATE2 is the quadratic term of ACCRATE; CHASSETURN is the perceptual change rate in asset turnover in a given period.
* Significance level: p < 0.1.
** Significance level: p < 0.05.
*** Significance level: p < 0.01.
b Not reported
One of the most relevant differences of our study with traditional literature is that we try to explain the accident rate using a leading indicator with our variable SRI (Grabowski et al., 2007; J. Hinze et al., 2013; Rozenfeld et al., 2010; Sparer & Dennerlein, 2013). This is because, SRI contains relevat safety barriers on site, and assessing them we can provide better leading indicators (Jørgensen, 2016). All the literature recognises that accidents cause disturbs at work, interfere in the normal development of tasks and finally entail direct and indirect costs (Fang et al., 2015; Ibarrondo-Dávila et al., 2015; López-Alonso et al., 2013). However, it does not seem to be a broad knowledge about the mechanisms that regulates the mutual relationship between the cost of accidents, on one side, and the costs of prevention and protection measures, on the other side. At least in theory, it has been proposed that a trade-off exists between the cost of accidents and the cost of prevention. But the problem in the practical application is that, although a high level of safety will always implies huge prevention costs, a low level of safety will not always be paired with high cost of accidents when they do not occur. In other words, since we always face the non-contingent nature of accident, it is important to adopt a longitudinal approach to show that sustained high level of risk on site increases the probability and the occurrence of accident.
All variables in our model (1) obtained the expected behaviour. ORDGES control variable is positive and significant (p<0.05) in the baseline and complete model, and SRI shows a significant (p<0.05) positive influence on accident rates (ACCRATE) (see Table 5, columns 7 and 8). More site risk generates more accidents according to pooled and random robust estimations with our complete model specification. The year dummies effects for 2008 and 2009 are significant. Since the world economic crisis that started on 2008 was especially important in Spain, and more acute in the construction sector, the significant estimations for the effects of 2008 and 2009 suggest that the recession might be affecting the relationships we are analysing. To verify this influence, we replicate the estimation of model (1) for the period 2004-2007. We confirm a significant positive relationship between SRI and ACCRATE (p<0.01) in pooled and random effects robust estimations (see Table 6, columns 7 and 8). All those robust results confirm our first hypothesis (H1). According to our results, we found empirical evidence of the existence of a lasting relationship between site risk level and the accident rates. Despite of the contingency dilemma between risk and accident and the fact that all likely accidents do not finally occur (Sundström-Frisk, 1985) , we can conclude that when risk expositions are repeated along the time, the rate of accidents increases. We think that the present study can be taken as a first step in the direction of narrowing the knowledge gap regarding the connections between risk assessment processes and causality models of accidents (Khanzode et al., 2012). This result is an additional reason to decrease those risk expositions which has become so common in our site works that we have got used to hem and we do not properly think about them (Jørgensen, 2016). However, as our results have showed the consequences of risk exposure will arrive sooner or later.
Regarding model (2) the behaviour of our control variables in the baseline models is as we expected. ROAt-1 is positive related with ROAt (pooled and random estimations, p<0.01), and CHASSETURN, has a positive significant impact on ROA (p<0.01) for all alternative specifications and all estimation methods. But our study does not replicate the findings of Argilés-Bosch et al. (2014) as we do not find a significant negative linear relationship between accident rate on site and firm profitability. Despite our explanatory variable (ACCRATE) is positive and significant at p<0.01 for the fixed effects estimations (see Table 7, column 6), this significance disappears when year dummies are incorporated in the model specification (see Table 7 last column).
Our results are interesting in the sense that they do not replicate previous evidence. Argilés-Bosch et al. (2014) argued that accidents are unexpected and disturb works and they found evidence that the effect of accidents on firm profitability is produced one year ahead. We have also failed to reproduce these effects of accident in a previous year on current ROA. One possible explanation for the different results we have obtained can be in the particular period of data we have utilized and the specific sample of both studies. We have considered (2004-2009), a period during which the construction sector in Spain had an abnormal evolution, evolving from a rapid and intense growth of activity to a sudden severe stop and decline of the activity. Differently, Argilés-Bosch et al.'s (2014) studied a panel data from 1998 to 2003, where the sector faced a more stable environment. Although the type of construction firms considered in both studies were similar, local firms with headquarters in the region under analysis (Catalonia, and Balearic Islands in our case), both regions show a relevant difference in its accident rates. In fact, Balearic Islands accident rate has historically been one of the highest in Spain. One other aspect that differentiate both studies is that Argilés-Bosch et al. (2014) used a stratified quasi random sample formed by all firms with fatal and serious accidents completed by a random sample of firms reporting minor accidents. Our sample, on the contrary, has included a random sample of construction firms with and without accidents independently of their level of seriousness.
Despite these considerations, our empirical results provide some evidence that a positive relationship might exist between accidents rate and ROA. Obviously, we do not propose that this counter-intuitive finding can be linear. As we have discussed, we think that a quadratic specification might explain the increasing behaviour of ROA as accidents grow for a range of low levels of accident rates and, at the same time, a decreasing tendency of ROA when accident rate is at relatively high levels. This quadratic relationship between accidents and ROA may exist if there is a trade-off between the cost of safety measures (Gurcanli et al., 2015) and the accident cost (Feng et al., 2015), as we have discussed above. This quadratic specification can be also consistent with theoretical contributions from other scholars as for example, Behm et al. (2004), Chalos (1992) or (López-Alonso et al., 2013).
In order to analyse those questions, we have proposed to estimate our model (3) that hypothesises a quadratic relationship between accident rate and return on assets. As we can see in Table 8, our results are consistent with hypotheses H3a and H3b across the different estimations methods with the model specifications that only include control variables. However, these results do not maintain for the complete model specification that includes the year dummies. In those model specification we found evidence (at p<0,1 for pooled and fixed effects estimations and at p<0,05 for random effects estimations) for a significant negative quadratic effect (supporting H3b) but we only found a significant positive linear effect (supporting H3a) for the random effects estimates (p<0,1). In all models, the sign of the coefficients remains positive for the lineal regressor, ACCRATE, and negative for quadratic regressor, ACCRATE2, which is in with of our hypothesis H3. Taking into account all these results, we can conclude that we partially confirm our hypothesis H3, as we have found a robust effect of the quadratic term of accident rate on ROA, even though when the significance level is not very strong.
The confirmation of the decreasing part of our adjusted predicted ROA as a function of accident rate, makes compatible our results with those in Argilés-Bosch et al.'s (2014) study. Due to the range restriction in the sample these authors used, because they did not include any firm without accidents in their study, we deduce that they might be focused on the decreasing right side part of our quadratic model. However, as it follows from our results, it seems that at least under the environmental conditions of our empirical study (high growth in construction activity followed by a rapid sudden decline), it is possible to find a positive relationship between economic firm performance and the number of accidents they report. This increasing behaviour of ROA is only manifested for an inferior range of accident rate values, while for superior values in that variable the tendency turns to be decreasing.
Another possible argument to explain the increasing part of our adjusted quadratic model (3), in addiction to COS theory, is that most construction firms were so extremely highly profitable during the years of the real-state bubble that they could absorb any kind of costs levels related with their operations, with includes H&S cost. This argument jointly with the prevailing absence of a culture of prevention in the sector and the low levels of professionalization would explain the evidence we have found of a positive relationship between accident rates and ROA. In fact, our results are partially in line with Ibarrondo-Dávila et al. (2015) who conclude that total H&S costs are substantial but they are not so onerous to erode financial firm performance.
One interesting practical and political implication is that if we allow that construction firms make decisions from a pure economic perspective, they will probably never invest enough in safety preventive measures. According to our results, only after arriving to a too high accident rate of 0.974, companies will start to suffer a negative impact on their financial results. This suggests that may result economically profitable maintaining high accident rates, especially in context of increasing production and consequently profits, which implies a clear conflict of interest from a social perspective. We claim that social and private interests should be aligned in order to reduce the high accident rates in the construction sector. This can be done mainly in two ways: investing higher amount of resources (financial, technical, human, etc.) and implementing several policy interventions based in two axis, safety promotion and safety control. For the first course of action, we think that Public Administration should promote more effective awareness campaigns of H&S, offer aids to make more efficient the safety management in firms or promote more appropriate and extensive training to workers and managers, among others. Related to the second course of action, we believe that might be helpful to implement harder supervision mechanisms and imposing stronger sanctions to those companies who does not strictly follow safety regulations. Facilitating and coercive approaches will be aimed to the same final goal: reduce the expositions to risks and make unprofitable for companies any deviation from the social optimal level of safety.
Facing the two intervention ways, it would be preferable to turn the focus on more proactive approach by looking at leading indicators on site such as safety barriers and organizational structure on site (Bellamy et al., 2008; Jørgensen, 2016; and paper 2 chapter 2). Therefore, we think that those policies that direct the focus towards the verification of an adequate level of resources and structure on sites and also towards the periodical checking of the sites live conditions, are preferable than policies drived to penalise high accident rates. The later are reactive actions which do not avoid the accidents, they just penalise its occurrence in economical terms. We think that this kind of measures must be applied only for extremely cases because they are clearly indicating that all preventive system is failing.
Balearic Islands Labour Authority for provide us data of accidents that has made possible our research.
Ale, B. J. M. (2006). The Occupational Risk Model – Final Report of the Workgroup on ORM. s.l.: Technical University Delft.
Argilés-Bosch, J. M., Martí, J., Monllau, T., Garcia-Blandón, J., & Urgell, T. (2014). Empirical analysis of the incidence of accidents in the workplace on firms’ financial performance. Safety Science, 70, 123–132. http://doi.org/10.1016/j.ssci.2014.05.012
Bandyopadhyay, S. P., Chen, C., Huang, A. G., & Jha, R. (2010). Accounting Conservatism and the Temporal Trends in Current Earnings’ Ability to Predict Future Cash Flows versus Future Earnings: Evidence on the Trade-off between Relevance and Reliability*. Contemporary Accounting Research, 27(2), 413–460. http://doi.org/10.1111/j.1911-3846.2010.01013.x
Behm, M., Veltri, A., & Kleinorge, I. K. (2004). The cost of safety. Professional Safety, 49(4), 22–29.
Bellamy, L. J. (2010). Which management system failure are responsible for occupational accidents? Saf. Sci. Monit., 14 (1).
Bellamy, L. J. (2015). Exploring the relationship between major hazard, fatal and non-fatal accidents through outcomes and causes. Safety Science, 71, Part B, 93–103. http://doi.org/10.1016/j.ssci.2014.02.009
Bellamy, L. J., Ale, B. J. M., Geyer, T. A. W., Goossens, L. H. J., Hale, A. R., Oh, J., … Whiston, J. Y. (2007). Storybuilder—A tool for the analysis of accident reports. Reliability Engineering & System Safety, 92(6), 735–744. http://doi.org/10.1016/j.ress.2006.02.010
Bellamy, L. J., Geyer, T. A. W., & Wilkinson, J. (2008). Development of a functional model which integrates human factors, safety management systems and wider organisational issues. Safety Science, 46(3), 461–492. http://doi.org/10.1016/j.ssci.2006.08.019
Brody, B., Létourneau, Y., & Poirier, A. (1990). An indirect cost theory of work accident prevention. Journal of Occupational Accidents, 13(4), 255–270. http://doi.org/10.1016/0376-6349(90)90033-R
Camino López, M. A., Ritzel, D. O., Fontaneda González, I., & González Alcántara, O. J. (2011). Occupational accidents with ladders in Spain: Risk factors. Journal of Safety Research, 42(5), 391–398. http://doi.org/10.1016/j.jsr.2011.08.003
Chalos, P. (1992). Managing cost in today’s manufacturing environment. Prentice Hall, Englewood Cliffs, NJ.
Cheng, C.-W., Leu, S.-S., Cheng, Y.-M., Wu, T.-C., & Lin, C.-C. (2012). Applying data mining techniques to explore factors contributing to occupational injuries in Taiwan’s construction industry. Accident Analysis & Prevention, 48, 214–222. http://doi.org/10.1016/j.aap.2011.04.014
Cheng, C.-W., Leu, S.-S., Lin, C.-C., & Fan, C. (2010). Characteristic analysis of occupational accidents at small construction enterprises. Safety Science, 48(6), 698–707. http://doi.org/10.1016/j.ssci.2010.02.001
Cheng, C.-W., Lin, C.-C., & Leu, S.-S. (2010). Use of association rules to explore cause–effect relationships in occupational accidents in the Taiwan construction industry. Safety Science, 48(4), 436–444. http://doi.org/10.1016/j.ssci.2009.12.005
Cheng, Q. (2005). The Role of Analysts’ Forecasts in Accounting-Based Valuation: A Critical Evaluation. Review of Accounting Studies, 10(1), 5–31. http://doi.org/10.1007/s11142-004-6338-4
Conte, J. C., Rubio, E., García, A. I., & Cano, F. (2011). Occupational accidents model based on risk–injury affinity groups. Safety Science, 49(2), 306–314. http://doi.org/10.1016/j.ssci.2010.09.005
EUROSTAT. (2004). Eurostat, 2004. Statistical Analysis of Socio-economic Costs of Accidents at Work in the European Union. s.l.: European Commission. Retrieved from http://appsso.eurostat.ec.europa.eu/nui/show.do?dataset=hsw_aw_inaws&lang=en
Fairfield, P. M., & Yohn, T. L. (2001). Using Asset Turnover and Profit Margin to Forecast Changes in Profitability. Review of Accounting Studies, 6(4), 371–385. http://doi.org/10.1023/A:1012430513430
Fang, D., Jiang, Z., Zhang, M., & Wang, H. (2015). An experimental method to study the effect of fatigue on construction workers’ safety performance. Safety Science, 73, 80–91. http://doi.org/10.1016/j.ssci.2014.11.019
Fang, D. P., Huang, X. Y., & Hinze, J. (2004). Benchmarking Studies on Construction Safety Management in China. Journal of Construction Engineering & Management, 130(3), 424–432. http://doi.org/10.1061/(ASCE)0733-9364(2004)130:3(424)
Feng, Y., Zhang, S., & Wu, P. (2015). Factors influencing workplace accident costs of building projects. Safety Science, 72, 97–104. http://doi.org/10.1016/j.ssci.2014.08.008
Fernández-Muñiz, B., Montes-Peón, J. M., & Vázquez-Ordás, C. J. (2009). Relation between occupational safety management and firm performance. Safety Science, 47(7), 980–991. http://doi.org/10.1016/j.ssci.2008.10.022
Gavious, A., Mizrahi, S., Shani, Y., & Minchuk, Y. (2009). The costs of industrial accidents for the organization: Developing methods and tools for evaluation and cost-benefit analysis of investment in safety. Journal of Loss Prevention in the Process Industries, 22(4), 434–438. http://doi.org/10.1016/j.jlp.2009.02.008
Grabowski, M., Ayyalasomayajula, P., Merrick, J., & McCafferty, D. (2007). Accident precursors and safety nets: Leading indicators of tanker operations safety. Maritime Policy and Management, 34(5), 405–425. http://doi.org/10.1080/03088830701585084
Gurcanli, G. E., Bilir, S., & Sevim, M. (2015). Activity based risk assessment and safety cost estimation for residential building construction projects. Safety Science, 80, 1–12. http://doi.org/10.1016/j.ssci.2015.07.002
Hallowell, M. (2011). Risk-Based Framework for Safety Investment in Construction Organizations. Journal of Construction Engineering and Management, 137(8), 592–599. http://doi.org/10.1061/(ASCE)CO.1943-7862.0000339
Harshbarger, D. (2001). Managing safety from the executive suite. Occupational Health & Safety (Waco, Tex.), 70(10), 28–30, 32, 34–36.
Heinrich, H. (1927). The incidental cost of accidents.
Heinrich, H. (1941). Industrial Accident Prevention a Scientific Approach. (second ed.)McGraw-Hill Book Company, London (1941).
Hinze, J., Thurman, S., & Wehle, A. (2013). Leading indicators of construction safety performance. Safety Science, 51(1), 23–28. http://doi.org/10.1016/j.ssci.2012.05.016
Hinze, J. W., & Teizer, J. (2011). Visibility-related fatalities related to construction equipment. Safety Science, 49(5), 709–718. http://doi.org/10.1016/j.ssci.2011.01.007
Hollnagel, E. (2008). Risk + barriers = safety? Safety Science, 46(2), 221–229. http://doi.org/10.1016/j.ssci.2007.06.028
HSE. (2015). Costs to Britain of workplace fatalities and self-reported injuries and ill health, 2013/14. Retrieved from http://www.hse.gov.uk/statistics/pdf/cost-to-britain.pdf
Hsiao, C. (2014). Analysis of Panel Data (3rd ed.). Cambridge: Cambridge University Press. Retrieved from http://ebooks.cambridge.org/ref/id/CBO9781139839327
Ibarrondo-Dávila, M. P., López-Alonso, M., & Rubio-Gámez, M. C. (2015). Managerial accounting for safety management. The case of a Spanish construction company. Safety Science, 79, 116–125. http://doi.org/10.1016/j.ssci.2015.05.014
Jørgensen, K. (2011). A tool for safety officers investigating ‘simple’ accidents. Safety Science, 49(1), 32–38. http://doi.org/10.1016/j.ssci.2009.12.023
Jørgensen, K. (2016). Prevention of ‘simple accidents at work’ with major consequences. Safety Science, 81, 46–58. http://doi.org/10.1016/j.ssci.2015.01.017
Khanzode, V. V., Maiti, J., & Ray, P. K. (2012). Occupational injury and accident research: A comprehensive review. Safety Science, 50(5), 1355–1367. http://doi.org/10.1016/j.ssci.2011.12.015
Labelle, J. E. (2000). What do accidents truly cost? - ProQuest. Professional Safety, 45(4), 38–42.
Laitinen, H., & Päivärinta, K. (2010). A new-generation safety contest in the construction industry - A long-term evaluation of a real-life intervention. Safety Science, 48(5), 680–686. http://doi.org/10.1016/j.ssci.2010.01.018
Laitinen, H., & Ruohomäki, I. (1996). The effects of feedback and goal setting on safety performance at two construction sites. Safety Science, 24(1), 61–73. http://doi.org/10.1016/S0925-7535(96)00070-7
Liao, C.-W., & Perng, Y.-H. (2008). Data mining for occupational injuries in the Taiwan construction industry. Safety Science, 46(7), 1091–1102. http://doi.org/10.1016/j.ssci.2007.04.007
López-Alonso, M., Ibarrondo-Dávila, M. P., Rubio-Gámez, M. C., & Munoz, T. G. (2013). The impact of health and safety investment on construction company costs. Safety Science, 60, 151–159. http://doi.org/10.1016/j.ssci.2013.06.013
López Arquillos, A., Rubio Romero, J. C., & Gibb, A. (2012). Analysis of construction accidents in Spain, 2003-2008. Journal of Safety Research, 43(5–6), 381–388. http://doi.org/10.1016/j.jsr.2012.07.005
Manuele, F., & Main, B. W. (2002). On acceptable risk. Occupational Hazards, 1, 57–60.
Manu, P., Ankrah, N., Proverbs, D., & Suresh, S. (2013). Mitigating the health and safety influence of subcontracting in construction: The approach of main contractors. International Journal of Project Management, 31(7), 1017–1026. http://doi.org/10.1016/j.ijproman.2012.11.011
Ma, Y., Zhao, Q., & Xi, M. (2016). Decision-makings in safety investment: An opportunity cost perspective. Safety Science, 83, 31–39. http://doi.org/10.1016/j.ssci.2015.11.008
McVittie, D., Banikin, H., & Brocklebank, W. (1997). The effects of firm size on injury frequency in construction. Safety Science, 27(1), 19–23. http://doi.org/10.1016/S0925-7535(97)00048-9
Nishikitani, M., & Yano, E. (2008). Differences in the lethality of occupational accidents in OECD countries. Safety Science, 46(7), 1078–1090. http://doi.org/10.1016/j.ssci.2007.06.017
Rechenthin, D. (2004). Project safety as a sustainable competitive advantage. Journal of Safety Research, 35(3), 297–308. http://doi.org/10.1016/j.jsr.2004.03.012
Rozenfeld, O., Sacks, R., Rosenfeld, Y., & Baum, H. (2010). Construction Job Safety Analysis. Safety Science, 48(4), 491–498. http://doi.org/10.1016/j.ssci.2009.12.017
Spangenberg, S., Mikkelsen, K. ., Kines, P., Dyreborg, J., & Baarts, C. (2002). The construction of the Øresund Link between Denmark and Sweden: the effect of a multi-faceted safety campaign. Safety Science, 40(5), 457–465. http://doi.org/10.1016/S0925-7535(01)00013-3
Sparer, E. H., & Dennerlein, J. T. (2013). Determining safety inspection thresholds for employee incentives programs on construction sites. Safety Science, 51(1), 77–84. http://doi.org/10.1016/j.ssci.2012.06.009
Sundström-Frisk, F. (1985). Kurs i Säkerhetsanalys. Arbetarskyddsstyrelsen, Stockholm.
Swuste, P., Frijters, A., & Guldenmund, F. (2012). Is it possible to influence safety in the building sector?: A literature review extending from 1980 until the present. Safety Science. http://doi.org/10.1016/j.ssci.2011.12.036
Tan, J., & Wang, L. (2010). Flexibility–efficiency tradeoff and performance implications among Chinese SOEs. Journal of Business Research, 63(4), 356–362. http://doi.org/10.1016/j.jbusres.2009.04.016
Teo, E., & Ling, F. (2006). Developing a model to measure the effectiveness of safety management systems of construction sites. Building and Environment, 41(11), 1584–1592. http://doi.org/10.1016/j.buildenv.2005.06.005
Visser, K. (1998). Developments in HSE management in oil and gas exploration and production. Safety Management, The Challenge of ChangePergamon, Amsterdam, pp. 3–66.
Wu, W., Gibb, A. G. F., & Li, Q. (2010). Accident precursors and near misses on construction sites: An investigative tool to derive information from accident databases. Safety Science, 48(7), 845–858. http://doi.org/10.1016/j.ssci.2010.04.009
Yang, H., Chew, D. A. S., Wu, W., Zhou, Z., & Li, Q. (2012). Design and implementation of an identification system in construction site safety for proactive accident prevention. Accident Analysis & Prevention, 48, 193–203. http://doi.org/10.1016/j.aap.2011.06.017
Yoon, S. J., Lin, H. K., Chen, G., Yi, S., Choi, J., & Rui, Z. (2013). Effect of Occupational Health and Safety Management System on Work-Related Accident Rate and Differences of Occupational Health and Safety Management System Awareness between Managers in South Korea’s Construction Industry. Safety and Health at Work, 4(4), 201–209. http://doi.org/10.1016/j.shaw.2013.10.002
Yung, P. (2009). Institutional arrangements and construction safety in China: an empirical examination. Construction Management & Economics, 27(5), 439–450. http://doi.org/10.1080/01446190902855633
Zhou, Z., Goh, Y. M., & Li, Q. (2015). Overview and analysis of safety management studies in the construction industry. Safety Science, 72(0), 337–350. http://doi.org/10.1016/j.ssci.2014.10.006
Papers relacionados
