Effect of Health and Safety Management Systems in the Construction Sector

Abstract

Management systems that are recognized as key tools to improve business management and the results associated with it have spread at the business level during the last 50 years. Regarding Safety Management Systems (SMSs), despite having specific international standards, there are no complete studies that analyze the degree of effectiveness of SMS, and even less in construction, a sector that concentrates a large part of registered labor accidents worldwide. The present investigation is an analysis of SMS effectiveness from an empirical study carried out over 48 months in five countries with a total of more than 23 million work hours between 2009 and 2012. Additionally, it is implied that the impact of SMS implementation in a certain organization must be complemented by a statistical qualitative analysis of its effect on the distribution of accidents. Both analyses are developed in the present study, thus contributing relevant implications when assessing both quantitatively and qualitatively the effects of developing and implementing a health and safety management system in this sector. The findings from this research can contribute to understanding how SMS implementation can help reduce accidents in this industry as well as to enhance SMS implementation in a high-risk sector.

1. Introduction

In most developed countries, the construction sector continues to account for at least 20% of all registered fatalities at work [1]. One of the main problems identified in this sector is the absence of effective procedures and Safety Management Systems (SMSs) [2].

In general, an SMS can be defined as

“…a combination of the planning and review, the management organizational arrangements, the consultative arrangements, and the specific program elements that work together in an integrated way to improve health and safety performance” [3].

In this area, there are international standards such as the British standards for health and safety management, OHSAS 18001:2007, Occupational health and safety management systems [4] or, more recently, the ISO 45001, Occupational health and safety management systems-Requirements with guidance for use [5]. Various entities have understood such systems to be crucial for reducing workplace accidents [6,7], and this has led to a marked expansion of SMSs in the construction sector [8].

Regarding the evolution of labor accidents in construction, Hudson’s three-wave model [9] postulates that the accident rate decreases progressively through technological actions (new and safer work equipment), the implementation of SMSs, and the development of a safety culture (management of conduct, leadership, commitment…) (Figure 1).

This theory, later developed by [2], considers that management systems, regardless of their origin, OHSAS 18001 [4], ILO-OSH-2001 [10], ANSI/AIHA Z-10 [11], are an implementation of objectives and practices established by the management of the company. Both in theory and in practice, current management systems have been shaped to develop an organizational culture and total safety regimen in the workplace.

When defining an SMS, a difference must be made between strategy and implementation [12] (Figure 2). The top management is responsible for the strategic part of the SMS, and the implementation represents the execution of the policies within the organization.

Various authors highlighted the need to develop and implement SMSs in specific fields in the construction sector [2,13]. Others have focused on analyzing the advantages and obstacles to their implementation [14,15].

Currently, there are no precedents or accident analysis studies that empirically prove that the implementation of an SMS reduces accident rates and significantly improves working conditions in the construction sector. For this reason, it is necessary to try and advance an analysis of the effectiveness of these management systems both at a quantitative level (empirically analyzing the effect of their implementation on accident rates) and at a qualitative level (analyzing how such implementation influences the distribution of occupational accidents in organizations in this sector).

The remainder of the paper is organized as follows. The background section includes an overview of the effect of an SMS on occupational accident rates and a summary of the scientific studies carried out on this subject. Consequently, the methodology section includes an outline of the research approach and the data analysis. Then, the results are analyzed both quantitatively (in terms of the accident rate evolution) and qualitatively (in terms of accident distribution evolution). Finally, the conclusion section includes the main findings of the research, as well as potential limitations and future research opportunities.

2. Background: Effect of an SMS on Occupational Accidents

Various authors have analyzed the influence an SMS has had on different productive sectors [16,17,18]. Torp and Moen [16] carried out a study in the context of the Norwegian legislation where all companies are required to have an SMS focusing on the effects of implementing or improving the SMS in the work environment, behaviors regarding health and safety, and musculoskeletal disorders in medium-sized companies (100 to 500 employees).

The study shows positive correlations between the implementation or improvement of an SMS and worker satisfaction in health and safety activities and support from colleagues, supervisors, and managers, including participation in Health and Safety (H&S) activities. Regarding SMS effectiveness in reducing accidents, the study shows a slightly significant correlation for the decrease in the section on musculoskeletal disorders, but it was not representative in terms of its effect on sick leave.

Along the same lines, various authors have studied the influence of specific variables both in the industrial and construction sectors, finding a conclusive relationship between the worker’s age, the frequency and severity of the injuries, and the number of accidents depending on the construction phase of the project [19]. However, few studies delved into the relationship between the eventual adoption of SMSs and the evolution of the accident rate in the aforementioned sectors.

Thus, the limited existing empirical studies on the impact of SMSs on accident results in the construction sector [7] focus on comparing the results of companies that have developed SMSs and those that have not. Therefore, it is necessary to analyze, in isolation, the impact that this implementation may have on a given company in terms of the evolution of occupational accidents.

On the other hand, the Canadian Institute for Occupational Safety and Health [20] analyzed this issue, concluding that there was no evidence to demonstrate that an SMS effectively reduces accident rates, understanding its implementation, as a whole, as a neutral action in terms of influence on the results of accidents. Their conclusion was based on the idea that the SMS presented very formal structures that fostered a more bureaucratic than practical management of aspects related to the safety and health of workers. It is important to emphasize that this conclusion was not based on any empirical study.

In any case, more recent studies [21] emphasize the absence of representative empirical studies that allow endorsing the effectiveness of these systems to reduce accidents while stressing the diversity of indicators used when evaluating the performance of an SMS. All this means that the absence of an SMS evaluation tool makes it difficult to make decisions based on observable, measurable, and evaluable data with standardized methods and, therefore, replicable and with comparable results.

This lack of available empirical studies on the specific effectiveness of implementing SMSs coexists with limitations in the analysis that these systems could have at a statistical level of occupational accidents when analyzed as a statistical function. Thus, although the basic theories of occupational accidents and their causal models [22,23] give this phenomenon a clear random component, there are few studies that offer statistical models to characterize occupational accidents and perform systematic analysis.

In general, it has been understood that the inherent randomness of accidents at work means that a probability function follows a Poisson distribution because, theoretically, all individuals have the same probability of suffering an accident, and this probability remains constant during the period of study. This conclusion was reached in specific studies carried out in the construction sector [24,25].

In a complementary way, other investigations [26,27] understand that, as long as there are external factors that affect the random phenomenon of the probability of having an accident, which means the existence of unobservable heterogeneities in mathematical terms, a distribution function will follow a Negative Binomial function.

Based on these analyses that already incorporate aspects related to the influence of an SMS on occupational accidents in a modified damage causation model (MLCM) (see, for example, [28]), a representative case study was used to analyze the eventual modulation that implementing an SMS may have in the accident rate function distribution.

In this scenario, in which the validation of the effectiveness of SMS implementation is especially complex, a double empirical study is proposed. On the one hand, the analysis in a real and sufficiently representative case of the quantitative effect that said implementation has for a multinational company in the construction sector. On the other, the qualitative study at a statistical level on the effect that such implementation has on the distribution of the variable “number of accidents”.

To achieve these objectives, multinational companies in the construction sector were considered as having the ideal characteristics to endorse the mentioned methodologies, firstly because multinational companies provide large samples, presence in different countries, a large number of annual working hours, and a great diversity of workers and types of construction projects. They can, therefore, provide quality empirical evidence on the impact of implementing an SMS on workplace accidents. In addition, large corporations have the necessary resources: company policies that are applied in all their centers and that guarantee uniformity of criteria have SMSs, reliable systems for obtaining observable, measurable, and valuable data, and trained personnel. All this opens the door to the development of applied research and the approach of research methodologies and mathematical models that would allow replicating and comparing results.

3. Research Methodology

To analyze the effectiveness of the SMS in the levels of accidents in construction works, a longitudinal study was developed throughout the period 2009–2012. This period was chosen due to the high activity in the construction sector. The characteristics of this study were:

• To quantitatively and qualitatively assess the effectiveness of the implementation of an SMS in the construction sector. The starting point was the selection of a sufficiently representative sample of works. In this way, and by analyzing this effect in a total of 302 works carried out in five different countries, it is possible to avoid the following possibly distorting obstacles:
  • Small samples vs. large samples. The possibility of having monthly data from five different countries with 302 works analyzed allows for a sufficiently large sample of data.
  • Type of works. Since it is a global construction company that manages various types of works, biases derived from possible concentration on a certain type of work are avoided.
  • Internationalization vs. local effects. The conformed sample includes works in five countries with different legislations and socio-cultural environments and what this implies in terms of geographical representativeness.
  • Standardization of procedures for action and organization and management vs. multiplicity of approaches. An obstacle that is overcome is a sample in which, as the works are managed by a single entity, there is a homogeneity of procedures that avoids distortions that may be present in other studies that analyze results in companies that adopt different approaches in their respective SMSs.
  • Data with real monthly frequencies that generate large panel data and a large number of observations vs. difficulty in obtaining field data.
• As for the implemented SMS, this was carried out based on the most accepted international standard in 2009 OHSAS 18001.
• Carrying out a longitudinal, nomothetic study with a plurality of observation units [29].

The process followed for the definition and implementation of the SMS in the company was as follows:

• Analysis of the company. An analysis of the processes and the control in terms of accidents were carried out, identifying the following as opportunities for improvement:

  (i)

  Hazard identification. There was no single system or methodology for hazard identification and risk assessment. Each country and each project left the evaluation criteria and methodologies in the hands of the health and safety officer following the country’s standards.

  (ii)

  Operational control. There was no operational control beyond the legislative needs of the country.

  (iii)

  Monitoring and measurement. It was necessary to define some minimum performance indicators that allow for knowing where we are and where we want to go.

The analysis was carried out after a 2-year international experience, which made it possible to fully understand the international needs of the company.

Design of the management system. Once the analysis was completed, an SMS was designed within the OHSAS 18001 standard under the principle of simplicity. Experience shows that most systems work better if they are kept simple than if they are made complex; therefore, simplicity was a key design goal. The system needed the flexibility to be applied in different countries and by different technicians with minimal training in management systems. The SMS standardized the risk assessment method, using the FINE method as the main tool for developing the Health and Safety Plans (H&S Plans) to be implemented on site. It also included specific procedures for managing the main health and safety duties (H&S training, H&S coordination, H&S planning, H&S inspections…).

2.

Commitment by management. Once the system was completed, a preventive policy was established that was in accordance with the company and its needs, based on the principles established in OHSAS 18001:

3.

Competencies of the managerial team and training. It was necessary to train the health and safety officers of each country in the management system so that they could then carry out the training at all levels of the production process in their own country: work manager, production manager, managers, prevention technicians, subcontractors, and so on. During the four years that the study lasted, more than one hundred technicians and agents involved in the management system were trained. This training program also included the technical site managers, and a complementary training program for laborers was also defined to be implemented at each site.

4.

Monitoring and measurement. The data analysis of the accident rate was carried out on a monthly basis. Those responsible for each country sent the accident rate reports and all the training activities carried out in the previous month to the central services before the 10th of the current month. The report was made in Excel sheets that allowed statistical monthly reporting of the country.

From the result of the analysis of the accident rate of the previous month, the corresponding corrective actions were generated that allowed the establishment of action plans, leading to a reduction of the accident rate.

5.

System feedback. A meeting was held annually in each country with the top managers of the company with the aim of explaining the degree of progress achieved, thus establishing the action plan and objectives to be achieved for the following year. This review made it possible to adapt the SMS to the particularities of each country so that, even with common management procedures, certain actions or controls could be reinforced in those countries where their application did not meet the standards defined in the SMS.

4. Specifications of the Sample

The construction company in which the study was carried out is one of the six largest construction business groups in Spain, with more than 100 years of experience in the sector, a presence in five continents, and a turnover of EUR 2.810 billion.

The sample object of the study comprises 293 civil engineering projects and nine building projects carried out in five countries and 23 million working hours used in all the projects included in the study (most of them by local laborers). With the exception of the US, where the company acted through three subsidiary companies, in the rest of the countries, it was the direct action of the parent company through the establishment of agencies or commercial branches. In the case of Peru, the beginning of the works was notably later than in the rest of the countries (nearly two years after the first projects of the rest of the countries). In any case, the management procedures and organization of the works were consistent with the company’s general management standards. The data collection was on all the projects:

The field study comprised more than 23 million hours worked by 125,000 workers during the period 2009–2012 in five countries (

Table 1. Characteristics of the sample according to country, type and number of works, average billing per project, number of workers per country, and dates of implementation and certification of the SMS.

Country	SMS Implementation	Total Projects	Average Billing Project/Workers	SMS Implementation/Certification
Chile	June 2009	4 Buildings and 9 Civil *	EUR 43 million/720	9 June/10 February
Mexico	June 2009	8 Civil	EUR 347 million/550	9 June/11 September
Peru	March 2010	1 Building and 9 Civil	EUR 53 million/440	10 March/11 February
USA	April 2010	883 Civil	EUR 1.5 million/700	10 April/15 June
Argentina	February 2011	16 Buildings	EUR 14 million/150	11 January/11 September

* Civil engineering works.

, Figure 3). Chile, the US, Mexico, and Peru were the countries that contributed the most hours, while Argentina had a more limited role. Chile presented a notable maximum number of hours in 2009, which drastically reduced in 2010; Mexico also fluctuated between 2011 and 2012, while the hours worked in Peru, since its incorporation in 2010, increased notably year on year.

As for the data recorded through the application of a specific tool for the collection and analysis of accident data, they were the following:

-

Number of workers.

-

Number of hours worked.

-

Accidents during work hours with sick leave. Serious accidents. Deadly accidents.

-

Days lost due to work accident with sick leave.

The data were recorded between the years 2009 and 2012 on a monthly basis and collected before the 10th of the current month to analyze them and implement the corresponding preventive measures to avoid repetition. Reporting was carried out by members of the safety department from each of the countries sampled.

Those responsible for the registry followed an incident reporting protocol that included an accident summary sheet, accident investigation, and corrective action. Subsequently, they were processed in spreadsheets to standardize such information (since the definition of accident rates varies from country to country) and to monitor incidence, frequency, and severity rates.

The data corresponded to more than 23 million hours worked for a total of five countries and with a monthly average of around 2500 workers (mostly locals). Considering an average correction factor of the order of 40% due to worker turnover in the construction sector, the average number of workers exceeds 3500 workers/month.

5. Methodology

5.1. Representative Variables to Measure the Effectiveness of the SMS

To evaluate the effectiveness of the SMS, it was decided to use the official occupational accident rates (OSHA) as output variables for the system. Although the definition of these indices is different for some countries, such as Chile and Peru, they were previously homogenized using the following metrics:

• Incidence Index (II): Represents the number of accidents per hundred thousand people exposed: No. of Accidents/No. of Workers × 10 ⁵.
• Frequency Index (IF): This gives the accident rate based on the hours worked: No. of Accidents/No. of Hours worked × 10 ⁶.
• Severity Index (GI): Represents days lost based on hours worked: No. of Days Lost/No. of Hours Worked × 10 ³.

5.2. Quantitative Analysis of the Effect of Implementing an SMS via Panel Data Analysis

In the research carried out, we started with five different countries for which the nine variables were observed on a monthly basis between 2009 and 2012. As it is a study conducted at the country level, a total of 19,440 data entries were available; nine variables were measured monthly for 48 months in five countries.

The following were considered for a model: the effect and significance of each of these factors (qualitative variable) with their respective levels (values adopted by the qualitative variable). The panel data corresponded to different entities (countries, individuals, companies, regions, etc.) and were observed over a time period of four years.

Likewise, it was understood that to carry out the quantitative analysis of the effect of the implementation of an SMS, the most representative indicators were the mentioned accident rates. Alternatively, the evolution of the number of accidents allows for a qualitative analysis of how this implementation affects the probability distribution function of the number of accidents.

Lastly, and in order to obtain the most representative information possible, it was considered appropriate not to limit only to the evaluation of the percentages of annual variation of the chosen indicators but also to use representations in a boxplot with a double purpose: (a)to detect the variations in the interquartile range (p25 and p75) that will confirm in a more relevant way the evolution of the aforementioned indicators and (b) to be able to detect eventual atypical values.

5.3. Qualitative Analysis of the Effect of the Implementation of an SMS

To complement the quantitative analysis of the effect that the implementation of an SMS has on the registered occupational accident rate, a second mathematical and qualitative study was proposed. In this case, we analyzed how the implementation of the SMS affects the dependent variable “number of accidents that occur annually (the variable)” at a statistical level.

Theoretically, the aforementioned variable will follow a Poisson distribution, since these are unlikely events that occur randomly. In fact, in samples of heterogeneities without face-to-face (specific study carried out on 14 construction projects, [28]), the probability distribution function of the number of accidents, f(x), was validated following a Poisson distribution:

$$f\left(x\right)=P \left[X\left(t\right)=x\right]=\frac{e^{-\lambda }{\lambda }^x}{x!}$$

Modeling of the accident distribution function as a probability function following a Poisson distribution [28], where the number of accidents recorded in an interval t, X(t), is modeled starting from the parameter $\lambda$(>0), which is the average number of accidents that occurred in a total of t hours worked.

In this way, as long as all the individuals in the sample had the same probability of having an accident, and this probability also remained constant during the study period, the aforementioned variable had to be adjusted to a Poisson distribution.

When external factors affect the probability and, despite the fact that individuals have constant probabilities over time, these vary from one subject to another due to the external factors. Thus, the negative binomial distribution would be the most adequate to model for the number of accidents that occurred in a certain period. This situation is explained in statistical terms because the presence of unobservable heterogeneities externally distorts the randomness, generating an overdispersion in the counting variable [30].

Starting from the theoretical base, a temporal analysis of the evolution of the observed proportions of the counting variable was carried out, analyzing the adjustment of the results obtained previously and at 12, 24, 36, and 48 months after the implementation of the SMS in the company. This analysis served the purpose of analyzing how the distribution that models the number of accidents occurred in a given period varies, if any.

The analysis also made it possible to assess the starting situation prior to the implementation of the SMS (before 2010) and how the distribution varied as the SMS was implemented.

6. Obtained Results

6.1. Quantitative Analysis of the Results in the Evolution of the Number of Accidents

From the graphic analysis of the evolution of the total number of accidents (Figure 4), it should be noted that, in all countries, there is a concentration of accidents in the first half of the period (the first 2 years) of the field study. Additionally, it is noteworthy that the most drastic improvement is observed in Chile since it has the highest rate of decline because, in 2009, it had the highest number of accidents. At the opposite extreme is Peru, which, from the outset, had the lowest accident rate.

Despite the heterogeneity of the curves for each country, a clear pattern of decline can be seen in general terms in the number of total accidents recorded in each country.

The response depending on the severity of the accidents is presented in Figure 5, revealing a certain homogeneity between countries regarding the behavior of the three types of accidents—minor, serious, and fatal—since, in all cases, it seems that the decrease mainly affects minor accidents, considerably less for serious accidents, and very rarely for fatal accidents.

6.2. Results in the Evolution of the Incidence Rate (II)

The results show a continued decrease in the Incidence Index (II) and the Accumulated Incidence Index (IIac) for the entire period and for all countries, which is presented in Figure 6. Consequently, the values of the p50 percentiles and p75, that is, the values of the II and IIac below which 50 and 75% of the records of those indexes are included, decrease throughout the period. This is evident in Figure 7, which clearly shows the decrease in p50 and p75 in the period 2010–2011 compared to 2009–2010.

From the analysis of the boxplot graphs, it can be deduced that the values of the p50 and p75 percentiles of the II and IIac decrease progressively in the total sample analyzed. In addition, the decrease in rates accelerates as time passes from the start of the SMS implementation.

In any case, the analysis of the annual variation of II by country (Figure 8) shows a marked and progressive decrease in this indicator, with year-on-year decreases ranging from 7% to 24% in the third year, accumulating decreases of up to 43% in two years in the case of Chile.

6.2.1. Observed Results in the Evolution of the Frequency Index (IF)

As with the Incidence Rates, the results (Figure 9) show a downward trend in the frequency rate for all countries throughout the 48-month field study.

Again, we observed decreases in p50 and p75 in the second part of the period compared to the first part of the period (Figure 10). The notable difference between the frequency rate and the cumulative frequency rate is that the latter includes all accidents and all hours worked from the beginning of the period until the end, while the former only accounts for the values of the two variables corresponding to the current month. Thus, the information provided by the boxplot is more representative since the variations in the interquartile range better define the real improvement in the levels of accidents.

The differences between countries and the interannual variations are shown in Figure 11. As was the case with the incidence indices, the downward trend in the values of the IF and IFac frequency indices continues.

6.2.2. Observed Results in the Evolution of the Severity Index (GI)

Unlike what happens with the incidence and frequency indices, which depend solely on the number of accidents, the severity index includes an assessment of the severity of the registered accidents. This index represents lost working days because of accidents with sick leave for every thousand hours worked. In this case, and unlike the rest of the indices analyzed, the evolution of the severity index, even though it is favorable in global terms for the company, does not present such an unequivocal behavior (Figure 12).

If the results of the present study are analyzed in the interannual variation figures (Figure 13), relevant decreases in the GI are verified. In this case, except in the case of Peru, whose late incorporation into the system makes it unrepresentative, the annual improvements in this indicator are concentrated around 8% in the third year, accumulating improvements in two years of full application of the SMS. The totals ranged from a limited 3% in the US to 36% in Chile, a country where, as was the case with II and IF, a greater effect of SMS implementation was recorded.

It is worth noting that the evolution of the GI does not always correspond to those of the II and IF. For example, in the case of the US, the interannual variations 2010–2011 were around −22% for the II and IF (Figure 8 and Figure 10), while for the GI, it was 4%. In the aggregate analysis, however, a clear global decline is observed in the company (Figure 14).

In this way, we observed, at the global company level, how the implementation of the SMS notably reduces the number of accidents with sick leave and, therefore, notable improvements in the severity rate are recorded at the aggregate level. Minor accidents had a greater reduction in number, while serious accidents seemed to follow a pattern that was less dependent on the SMS. Starting from this base, we then proceeded to analyze how the implementation of the SMS affects the probability of occurrence of an accident.

6.3. Analysis of the Accident Rate Distribution

Once the quantitative results of the research carried out were obtained, which showed an evident improvement in the accident rate after the implementation of the SMS, the modeling of the variable “number of accidents” was analyzed as a complementary aspect. The present study was used to deepen the analysis of the phenomenon of occupational accidents to verify whether said variable fits a certain type of probability distribution function, which will allow a deeper analysis of the effect of the SMS on the number of accidents.

As mentioned before, it must be assumed that the number of accidents follows a Poisson distribution [25] since these are non-deterministic and unlikely events that occur randomly.

However, continuing on a theoretical level, the negative binomial distribution would be more adequate than the Poisson distribution to model the number of occupational accidents that occurred in a certain period. This is justified because, in a Poisson distribution, it is assumed that all individuals have the same probability of having an accident and that this probability remains constant during the study period. However, this assumption may not take into account the impact of environmental factors that affect the accident rate: socio-labor, economic, and legislative factors, among others). Therefore, despite the fact that individuals have constant probabilities over time, these vary from one subject to another due to environmental factors.

One of the main environmental factors will be the implementation of an SMS, and understanding that said system will affect the probability of an accident occurring, neutralizing environmental factors as much as possible. Therefore, if the probability distribution of having an accident is close to a Poisson distribution, it will be in a scenario where accidents are kept under control, depending only on the parameter λ: where λ represents the number of times an event is expected to occur (in our case, an accident) during a certain period of time.

The graphs that follow show the observed proportions together with the Poisson and Negative Binomial proportions that the counting variable would follow. Figure 15, Figure 16, Figure 17, Figure 18, Figure 19, Figure 20 show the adjustment of the results obtained at 12, 24, 36, and 48 months, respectively, after the implementation of the SMS based on OHSAS 18001 in the company.

As can be seen, initially (both before 2010, Figure 15, and 12 months after the SMS application, Figure 16), the fit between what was observed and the Negative Binomial is perfect, especially in Figure 16. This means that it can be concluded that there are causes assignable to the environment that justify that a number of accidents do not follow a Poisson distribution.

As the implementation of the SMS progressed, it became very clear how the observed distribution modulates towards a Poisson distribution. Thus, until 36 months, the external heterogenization factors of the environment were neutralized, and consequently, the probability distribution met the theoretical assumption for the variable “number of accidents”, which is a Poisson distribution.

This progressive modulation and adjustment of the distribution of the variable “number of accidents” as the implementation of the SMS advances qualitatively confirms what has already been pointed out by the quantitative analysis carried out.

7. Conclusions

In the present study, an in-depth analysis of the effect an SMS has on the occupational accident rate of a company in the construction sector was carried out, delving into two aspects of analysis: quantitative and qualitative.

On the quantitative side, the effects that the implementation of an SMS has on the interannual evolution of the main occupational accident rate indicators were analyzed, including the interannual evolution of the interquartile range of said indicators. Thus, beyond the results derived from the direct observation of the mentioned evolution, a more in-depth evaluation of the variation of such indicators was also carried out.

On the qualitative side, an analysis of the modulation that such an SMS has on the distribution of the probability of occurrence of an accident confirmed what was described at the theoretical level by previous authors, as long as heterogenizing external factors concur that fit a Negative Binomial distribution. The methodology used made it possible to analyze the evolution of the distribution as the degree of implementation of the SMS advanced, modulating such distribution function clearly towards a Poisson distribution.

Based on the field study carried out in five countries (Argentina, Chile, Mexico, the USA, and Peru), encompassing more than 300 construction projects of a multinational company in the construction sector, with more than 20 million hours worked and a monthly average of 3500 workers during the 48 months between 2009 and 2012, the following general conclusions can be drawn:

→

Once the SMS was implemented, there was a clear drop in the frequency and incidence rates and an appreciable drop in the severity (both monthly and cumulative). In this way, and at a global level, there was a clear improvement in the evolution of the different indicators used. This improvement is clearly evolutionary, accelerating in the second and third years after the beginning of the implantation and decelerating such improvement from the fourth year.

→

Likewise, it is necessary to highlight the consistency of the result in the decrease in the accident rate, given that the decreases are appreciable in all the countries despite the heterogeneity of the projects that were carried out, the number of workers involved, and the diversity in their socio-cultural environments. Notwithstanding this, the present study verified how in the countries that established a work methodology from the beginning, as is the case of Peru, better results were achieved both in the index of frequently as in the severity index than in those countries that implement this methodology years after it was establishment.

→

The decreases in the accident rate applies to all the countries and practically all of the indicators used (partial and accumulated frequency, incidence and severity indices), despite the heterogeneity of the projects that were carried out, the number of workers involved, and the diversity in the legal frameworks of application.

→

This positive evolution, which supposes cumulative decreases of up to 43% in the Incidence Rate in some of the countries, is confirmed at a statistical level through the analysis carried out on a boxplot graph. This complementary analysis confirms a marked improvement in the registered occupational accident rate at a global level based on the evolution of the interquartile ranges analyzed.

→

Based on this, and in contrast to previous studies [20], it can be concluded that the implementation of an SMS in this type of construction project reduces occupational accidents in global terms.

→

Additionally, as time elapses from the start-up of the SMS, the present study out demonstrates how the distribution of the probability of occurrence of an accident changes from a Negative Binomial distribution to a Poisson distribution (random). This means that the SMS neutralized the possible reasons for that increase in the number of accidents, making their occurrence not random but due to assignable causes. This conclusion reinforces, from a qualitative point of view, the effectiveness of the implementation of an SMS in this sector.

→

Finally, and based on the double analyses carried out at the quantitative and qualitative levels, the Hudson accident reduction model [9] is also confirmed, which locates in its second wave the reduction derived from the implementation of an SMS.

Additionally, the results allow us to reach the following particular conclusions:

→

Although in global terms there is a significant decrease in occupational accidents, it should be noted that this decrease is notably concentrated in minor accidents, while serious accidents and fatal accidents follow more random patterns. This conclusion is in line with what has already been pointed out, for example, by [7], that understanding that the implementation of an SMS especially improves accident rates by avoiding minor and serious accidents, but that fatal accidents follow another distribution that is difficult to reduce.

→

As for the decreases in accident rates, logically, they are more marked in the frequency and incidence rates, while the severity rates take longer to decrease and do so less markedly.

→

Another noteworthy aspect of the descriptive analysis performed is the type of statistic considered for the study variables. Unlike many of the existing studies, which focus on the analysis of the mean, this analysis is enriched by using additional statistics, the median and the p25 and p75 percentiles. Thus, a series of boxplots was obtained that provided more information, including statistical values such as the minimum, the interquartile range, or the extreme values, giving valuable information regarding the traditional analysis models based on the study of the mean value while reinforcing the significance of the results obtained.

→

In this way, the loss reduction model based on “waves” proposed by Hudson is also qualitatively confirmed.

→

As an essential aspect obtained from this investigation, the level that should be highlighted is the progressive modulation registered in the probability distribution function of an accident at a global level. Thus, it is verified how, year on year, the distribution changes from an almost full adjustment with the Negative Binomial until it ends up adjusting to a Poisson distribution. It is thus confirmed that the external aspect on which action was taken, the SMS, was representative when modifying the randomness of the distribution and that, as this heterogenizing effect is neutralized, the randomness of such distribution is reduced.

Despite the results of the present study, more in-depth analyses could be considered for the rest of the actions that make up the Hudson accident reduction model, such as the influence of human factors. These were also identified as a path for improvement [31]. Additionally, and for future research, it should be borne in mind that the effectiveness of an SMS can also be estimated by considering its effect on the avoidance of risk situations. This also leads to an interest in carrying out broader studies in which the influence of actions by other agents, for example, the client, can have on the accident rate.