The Causes of Occupational Accidents and Injuries in Romanian Firms: An Application of the Johansen Cointegration and Granger Causality Test

Organizational risks are present in any activity, so it is important to manage them properly. The jobs are dynamic and involve a series of processes and activities. The entire human resource is exposed to several risks. If these risks are approached correctly, the organizational capacity to achieve its objectives and vision will increase considerably. This paper aims to investigate the relationships between work accidents (fatal and non-fatal) and the causes that contribute to their occurrence (causes dependent on the executor, causes dependent on the means of production, workload-dependent causes, and work-dependent causes—the work environment). The augmented Dickey–Fuller (ADF) test is employed to check the data stationarity series, while the Johansen test determines the cointegration relation of variables. The data have been collected from Romanian organizations. The vector error correction model (VECM) and Granger causality test are applied for speed of adjustment, nature, and direction of variables’ relationship. This research demonstrated that both data series are free from the unit-root problem at first difference. The lag length criterions select the third lag for model fitness, and Johansen cointegration declares that variables are cointegrated for the long term. The vector error correction model shows the speed of adjustment from the short to the long run is 83.35% and 42.60% for work and fatal accidents. The study results show that fatal accidents have a series relationship with selected cases for the short run and have a long-run relationship with the means of production. Fatal accidents are directly related to means of production. Fatal accidents are not designed by executors, workload-dependent causes, or work environments in the short run. Fatal accidents are directly related to the means of production and sudden incidents happening in the long run. Fatal accidents are considered by executors, workload-dependent causes, or work environments in the short run. In the long run, fatal accidents are directly related to the means of production and sudden incidents happening.


Introduction
Improving the competitiveness of the business environment contributes to developing new jobs. The business environment dynamics contribute to the permanent development of new jobs that often require new skills or new work procedures. As well as the European dents are registered for outsourced operations [25][26][27]. That is why it aims to train and plan work tasks to reduce the accident rate in the industry [8][9][10][11][12]28]. Accidents at work in recent years have increased, with various causes contributing to increased absenteeism in organizations [29,30]. It is vital to make an inventory of work accidents and contribute to their development because the dynamics and current complex working conditions can develop more accidents. It can be seen that both in the EU and Romania, there is an increasing trend regarding workplaces and work accidents. It can be said that an increase in the number of jobs leads to an increase in the number of accidents.
Accidents at work in different fields of activity fall in specific directions. These directions concern the worker, the work environment, the production process, or the employee [31,32]. Depending on the severity, the two categories for work accidents can be delimited: non-fatal and fatal accidents at work [30,33]. Therefore, these work accidents are caused by causal elements. The systemic principle of approaching the work processes requires expressing the causes on the four elements: the executor, the means of production, the workload, and the work environment. This procedure is also followed if the work process is complex or less complex. The work process elements are characterized by attributes, states, phenomena, and behaviors. These features can lead to a system malfunction. This dysfunction can cause accidents at work and/or occupational diseases, together called risk factors. These risk factors can be potential causes of accidents at work or occupational diseases. If an accident at work or occupational disease has occurred, the causes of these elements must be identified. The government should increase work accident data quality [34].
The source of the data series is the Romanian Ministry of Labor and the Romanian National Institute of Statistics. These data series show the total number of accidents at work (fatal and non-fatal). Therefore, the fields of activity are considered as evaluated by the Ministry of Labor: retail trade, land transport, building construction, wholesale trade; manufacture of road transport vehicles, trailers, and semi-trailers; the food industry; human health activities; industry of metal constructions and metal products, exclusive machines, equipment, and installations; production and supply of electricity and heat, gas, hot water, and air conditioning; manufacture of furniture; public administration; wood processing, and other activities.
The paper presents an analysis of accidents in Romania for the period 2010-2019. At the Romanian level, it is necessary to analyze the number of accidents and a clear presentation of their causes because the existing research is restrictive. The present study advances the research carried out so far by the methodology used and by the long period of analysis (10 years). Thus, an inventory of the current situation is developed by presenting some necessary hypotheses. This study is of interest to researchers, authorities, practitioners, and companies to develop strategies to reduce the likelihood of risks.

Materials and Methods
The causes of work accidents are divided into four categories: causes dependent on the executor-the injured (C1), causes dependent on the means of production (C2), workloaddependent causes (C3), and work-dependent causes-the work environment (C4). Cause 1 refers to falls, omissions, incorrect operations, non-use of protective equipment, or other dangerous conditions. The working conditions that constitute C1 can significantly impact the number of work accidents. The causes dependent on the means of production are relevant for work accidents. Often, production equipment is not adequately balanced, its handling conditions can cause minor or severe accidents, and working positions for their use can contribute to accidents of varying severity. The workload, C3, can be as significant as the first two conditions. Deficiencies in orientation, omissions in work operations, and other conditions are essential for reducing work accidents and assessing significance. Work-related causes, C4, are critical occupational hazards. The workplace climate or the environment's characters are essential conditions that must be managed to correct the occupational risks. The causes of fatal accidents are the four leading causes of the present study: causes dependent on the executor-the injured (C1), causes dependent on the means of production (C2), workload-dependent causes (C3), and work-dependent causes-the work environment (C4). From the C1 perspective, falls, omissions, and some operations can contribute to fatal accidents. Additionally, not using protective equipment can cause fatal accidents. The causes dependent on the means of production are the leading causes of fatal accidents. In some situations, workload, C3, can be fatal to the employee. This high workload can overwhelm the employee and ultimately lead to death. If significant omission occurs, it can be fatal to employees. That is why this primary cause, C3, is considered significant for fatal accidents. The workplace climate or the nature of the environment that are conditions for the main cause of C4 can be the cause of fatal accidents. All the main causes' conditions must be evaluated through efficient risk management.
To reduce the severity of the causes of work accidents or occupational diseases, the elements should be identified at each department's level to specify as clearly as possible the place of action and prevention measures and the employer's responsibility in case of an accident. Following Law 319/2006, the causal elements are divided into four categories. The employer must declare the occurrence of an accident and must identify all the causal elements and working conditions. These data are transmitted through a nationally agreed report. The causal elements that contribute to the occurrence of work accidents are:

•
Causes dependent on the executor-the injured.

•
Causes dependent on the means of production. • Workload-dependent causes. • Work-dependent causes-the work environment A work accident can have one or more causal elements. The State Committee for Labor Protection establishes the classification of cases. Causes dependent on the executorthe injured-involve the improper performance of process operations. As a result, dangerous and harmful conditions develop. Causes depending on production refer to the machines' functional movements. Mechanisms triggered contraindicated incompletely described procedures for functional movements and other activities depending on production. Workload-dependent causes refer to the improper distribution of executors in the workplace. In some companies, no training is provided in occupational safety and health, or this training is provided at a low level. Work-dependent causes-the work environment-include the work climate. Depending on the field of activity, the work environment may have different particularities.

Data Description
This research is based on quarterly data of work and fatal accidents (mentioned the specification of firms/companies/road accidents) from 1 January 2010 to 31 December 2019. The data used in this research were collected from the Romanian Ministry of Labor and the National Institute of Statistics. The employer immediately reports an event that has contributed to the body's death or injury to the Territorial Labor Inspectorate.
The work and fatal accidents are considered dependent variables, while four main causes of work conditions are declared independent (see Figure 1). The main causes cover different work conditions, which are explained in detail in Table 1. The main causes are computed by the weighted average method for analysis.
The number of accidents considered in the present research is presented in the following table. Each accident had one or more causal elements. These are accidents reported by employers.

Cause-4
Work-dependent causes-the work environment Physical causes (air temperature, air humidity, low light level) Psychosocial climate The special character of the environment Other causes The number of accidents considered in the present research is presented in the following table. Each accident had one or more causal elements. These are accidents reported by employers.

Research Methods
The research methodology for this article is alienated into four sub-sections that are (1) stationarity test or unit-root test. (2) Cointegration test is employed after confirmation of integration order and stationarity of data series. In this research, the Johansen cointegration test is employed to check long-run relationships or long-run equilibrium among series with a linear combination of variables. (3) The vector error correction model (VECM) is used to determine the causality direction of variables after confirmation of the cointegration relation. The fourth method is the Granger causality test used to check individual variable direction and relationships with another individual. A flowchart of the complete methodology is pictorially presented in Figure 2. In this study, Eviews-10 is used to run the Johansen cointegration model, VECM, and Granger causality test for data analysis.

Cointegration Test
After checking the order of integration and stationarity of series, the second stage of analysis is tested. The cointegration test measures long-run relationships or long-run equilibrium among many time series datasets with a linear combination of variables. This research employs the Johansen cointegration test to check the stability and long-term equilibrium relationship between variables by the following equation: The above equation shows Π is an indicator of the adjusted disequilibrium matrix. The stacking coefficient A is boosted up endogenous factor's speed of change counter to

Unit-Root Test
The unit-root test is used to verify the order of variables' integration. As per existing literature, the augmented Dickey-Fuller (ADF) is a well-known test for checking the integration order. This research is also employed the ADF test for unit root by the following equation: ρ represents that the number of lags is empirically determined using Schwarz information criteria (SIC), and e t is an error term with zero mean and variance. The coefficient term Y t−1 is included later for testing the significance of coefficient [35][36][37]. The augmenting process is completed with the possible removal of autocorrelation among error terms. If: Calculated Value > Critical Value = Accept H 1 and H 1 = Y t has stationarity . (3)

Cointegration Test
After checking the order of integration and stationarity of series, the second stage of analysis is tested. The cointegration test measures long-run relationships or long-run equilibrium among many time series datasets with a linear combination of variables. This research employs the Johansen cointegration test to check the stability and long-term equilibrium relationship between variables by the following equation: The above equation shows Π is an indicator of the adjusted disequilibrium matrix. The stacking coefficient A is boosted up endogenous factor's speed of change counter to disequilibrium. The sign Γ is used to capture the short-term dynamic adjustment [37][38][39]. This test progression can declare the association of variables with their positions in the matrix and featuring roots.

Vector Error Correction Model
The vector error correction model (VECM) is used to determine the causality direction of variables after confirmation of cointegration relation [40,41]. The VECM framework is structured as follows: where the coefficients λ 1 − λ 7 are indicated the error correction term, the homoscedastic disturbance term is denoted by ε 1t − ε 4t , and ECM t−1 represents long-run equilibrium and speed of adjustment.

Granger Causality Test
The Granger causality test estimated by the following equation: In the above equation, it is assumed that the term ε 1t and ε 2t are uncorrelated with each other as E(ε 1t , ε 2t ) = 0 = E(ε 2t ε 2s ) . . . ..s = t. The unidirectional causality from fatal and work accidents to 4 specific causes is shown in the equation. If estimated coefficient α 2i is statistically significant, α 2i = 0 then Y → Granger causes → X [42]. If X is cause variable for Y and β 2h is statistically significant, i.e., β 2h = 0 [43,44]. The significance of α 2i and β 2h confirms mutual dependency of two specific variables. The term Y and X will be independent if α 2i and β 2h are not other than zero.

Results
In Table 2, augmented Dickey-Fuller (ADF) results for integrating work and fatal accidents with four main causes are presented. The null hypothesis of a unit root in series has failed to reject at a level even at a 10% level of significance [45]. Both dependent variables series, i.e., work and fatal accident, also obtained stationarity at the first difference at 1%. The stationarity of the series under work and the fatal accident is confirmed after the first difference. After taking the first differences, both conditional series have shown 1% significance level for all causes except C-2 of a fatal accident. The C-2 of the fatal accident of ADF is significant at a 5% level of confidence. The last column contains the order of integration, which is representative of the integrated variables order [46,47]. The numeric values 0 and 1 are used for level and first difference. The below table showed I (1) in front of all causes because the results support the first difference's significance, which is a fundamental requirement to run the cointegration test.  Table 3 shows the results of a second step in which an appropriate lag length is selected. Based on VAR lag order selection criteria, lag three is chosen for work accidents by three information criteria, i.e., Akaike information criterion (AIC), Schwarz information criterion (SC), and the Hannan-Quinn information criterion (HQ), while in fatal accidents, AIC and HQ chose lag three and SC chose lag 0 as an appropriate lag. After selecting the appropriate lag length, the next step is to check long-run relationships among variables used in research. The Johansen cointegration test is employed to check the long-run relationship among variables and is presented in Table 4. The results are in two portions, i.e., the Johansen trace test and max eigenvalue. The results indicated that variables are cointegrated with none 83.83758 for trace (work accident), 76.77265 for trace (fatal accident) at a 1% level of significance. The value of the maximum eigenvalue for work accidents is 41.83049, and the value for a fatal accident is 36.07142, which are significant at 1% and 5% levels, respectively. The cointegration results confirmed the long-run relationship between work and fatal accidents with four leading causes.
In the next step, the trace and max eigenvalues test results generated a cointegration equation based on log-likelihood (LL) ratio. A linear combination between selected causes with work accident and the fatal accident could be scrutinized from the cointegration equation. This cointegration equation is also used to check long-run relationships, and it is confirmed for this research. In Table 5, The normalized work accident and fatal accident equations showed a mixed relationship-the significant positive relationship for C-3 and C-4 and a negative relationship for C-2 with work accidents. Simultaneously, the fatal accident is positively related to C-2 and C-3 and has a significant relationship with C-4.
In Table 6, the vector error correction model (VECM)'s results are presented with multiple time series' long-run and short-run relationship between a dependent (work and fatal accidents) and independent variable (four main causes). The short-run causality test shows unidirectional causality between work accident and cause-3 with a first difference (WA −1 => C3 −1 = WA −1 ) at a 10% level of significance. A unidirectional causality confirmed between cause-1 and cause-3 (C1 −1,−2,−3 => C3 −1,−2,−3 = C1 −1,−2,−3 ) with three different structures at a 1% level. The cause-1 unidirectional related to cause-4 (C1 −1 => C4 −1 = C1 −1 ) at a 10% level of confidence. The unidirectional causal effect of cause-1 and cause-3 on work accidents (C2 with different subscripts is confirmed at a 1% level of significance. The error correction term (ECT t-1 ) meets the requirement of negativity and significance at a 1% level, supporting the relationship between dependent and independent variables. The ECT value declares the 83.35% adjustment speed of short-run causality into the long run. The ECT of cause-1, 2, and 4 is positively significant, while cause-3 is negatively insignificant. The unidirectional relationship is observed between fatal accident and cause-2 with one difference (FA −1 => C2 −1 = FA −1 ) at 1% and cause-2 with a difference of three (C2 −3 => FA −3 = C2 −3 ) significant at a 5% level of significance. A bidirectional relationship observed between fatal accident and cause-3 with a difference of one at a 5% level of significance and unidirectional causality with a difference of three (C3 −3 => FA −3 = C3 −3 ) is significant at the 1% level. With a difference of two and three, a unidirectional relationship between cause-1 and cause-2 is found (C1 −2,−3 => C2 −2,−3 = C1 −2,−3 ) at a 1% level of significance. The unidirectional relationship between cause-1 and cause-4 (C1 −2 => C4 −2 = C1 −2 ), cause-3 and cause 1(  Table 6. The speed of adjustment from short-term equilibrium to long-term equilibrium for fatal accidents is 42.60%, with a negative sign and a 10% significance level. Table 7 is a summary of Table 6 and cross-check of directional causality. The VECM declared the long-run relationship among variables and the speed of variables' adjustment from the short-run to the long-run equilibrium level. The unidirectional relationship is shown with "→" and the opposite or bidirectional relationship with "~" in Table 7. The total observations of the research are 37.     Study results show a relationship between work accidents, causes dependent on the executor, causes dependent on the means of production, and work-dependent causes with 1% and 5% significance levels. In fatal accidents, cause-1 (dependent on the executor) and cause-3 (workload-dependent causes) have shown a positive and significant relationship with a fatal accident at a 5% level. Cause-2 (dependent on the means of production) and cause-4 (work-dependent causes) have shown a positive and insignificant relationship with fatal accidents. Figure 3 is a graphical representation of the short-run relationship between work accidents and fatal accidents with selected independent causes, i.e., the injured person (executor), means of production, workload, and work environment. Figure 3 is a graphical representation of the short-run relationship between work accidents and fatal accidents with selected independent causes, i.e., the injured person (executor), means of production, workload, and work environment.

Discussion
Organizational activities contain many hazardous steps collectively, which are known as organizational risk. Improper management of organizational risk can decrease an organization's production, affecting the speed of the production process and work accidents [48]. This research is focused explicitly on accidents (fatal and non-fatal) as dependent variables, while different causes of accidents are considered as independent variables [49][50][51]. Working equipment is also vital in workplace accidents [52]. Many causes will become the reason for accidents in an organization, but our selected causes are related

Discussion
Organizational activities contain many hazardous steps collectively, which are known as organizational risk. Improper management of organizational risk can decrease an organization's production, affecting the speed of the production process and work accidents [48]. This research is focused explicitly on accidents (fatal and non-fatal) as dependent variables, while different causes of accidents are considered as independent variables [49][50][51]. Working equipment is also vital in workplace accidents [52]. Many causes will become the reason for accidents in an organization, but our selected causes are related to executors, means of production, workload, and work dependent on the work environment. Analytical data were collected from the Romanian firms. First of all, the ADF is used to check the stationarity of the data series. The stationarity of both series under work and the fatal accident is confirmed after taking the first difference. Both conditional series have shown their significance at a 1% level for all causes except C-2 of the fatal accident. The C-2 of the fatal accident is significant at a 5% level of confidence. Secondly, the appropriate lag length is selected for both time series by considering lag length criterions, i.e., Akaike information criterion (AIC), Schwarz information criterion (SC), and the Hannan-Quinn information criterion (HQ). For work accidents, all criterions support lag three as an appropriate lag length.
In contrast, fatal accidents receive support for lag three by two criterions (AIC and HQ), and SIC chose lag 0 as an appropriate lag. Thirdly, the Johansen cointegration test is employed to check the long-run relationship among variables. The Johansen cointegration confirmed a long-run integration among data series. Lastly, the vector error correction model (VECM) was applied to measure the speed of adjustment from short run to long run, which is higher for work accidents (83.35%) than fatal accidents (42.60%). As per our analysis, work accidents are riskier and more impactful than fatal accidents, and a close and strong relationship between work accidents and their causes is observed. Heinrich's law has also supported the results of this research as Heinrich's law categorized 0.3% accidents as having majorly injured the victim, 8.8% having been minorly injured, and 90.9% having no injuries. According to this pre-mentioned rule, the ratio between work accidents and injuries and fatal accidents and injuries comparatively increases at the same rate [53]. The short-run causality test shows unidirectional causality between work accident and cause-3 with a first difference (WA −1 => C3 −1 = WA −1 ) at a 10% level of significance. The unidirectional causal effect of cause-1 and cause-2 on work ac- with different subscripts is confirmed at a 1% level of significance. The unidirectional relationship observed between fatal accident and cause-2 with one difference (FA −1 => C2 −1 = FA −1 ) at 1% and cause-2 with a difference of three (C2 −3 => FA −3 = C2 −3 ) is significant at a 5% level of significance. A bidirectional relationship observed between fatal accident and cause-3 with a difference of one at a 5% level of significance and unidirectional causality with a difference of three (C3 −3 => FA −3 = C3 −3 ) is significant at the 1% level. The unidirectional relationship between cause-4 and fatal accidents (C4 −1,−2 => FA −1,−2 = C4 −1,−2 ) is observed.

Study Limitations and Future Research Directions
As per our best knowledge, this is the first research work that pointed out the relationship direction between work and fatal accidents and their causes by utilizing the data of Romanian organizations. Due to the innovative nature of research and direction, this research is limited in its context. Firstly, the study is focused only on Romanian industries as a whole, while some productive activities are riskier in different organizations than others. In microanalysis, future research can be focused on accidental risk in the different industries separately. The microanalysis will determine the exact practical place or activity that shows accidental risk management's quick impact. Secondly, the macro impact of this research can be checked in future research by enhancing the data series, focused on all types of industries, including manufacturing industries, comparative analysis of manufacturing industries and rest of all industries, and comparative analysis of different Romanian industries with other developed countries. Thirdly, the ARDL methodological approach can analyze future prediction capability and check the psychological impact of work accidents on industries employees using the sentiment index approach. Future research will address the costs of occupational and fatal risks. This research employed the Johansen cointegration model, vector error correction, and Granger causality models. The GMM, panel regression analysis, and dynamic ARDL models can be employed concerning work accidents in future research. The same study can be done with various databases from different countries and compare them.

Conclusions
Accidents at work are of great importance to organizations. These must be addressed through risk management to be planned, evaluated, analyzed, and controlled correctly. Regardless of the field of activity, human resources are exposed to the causes that contribute to risks. The treatment of the causes contributes to reducing the probability of occurrence of the risks and implicitly to the diminution of the consequences. Once a risk has arisen, the foreseeable consequences must be considered. In this research, the augmented Dickey-Fuller (ADF) test is employed to check data series stationarity. This research demonstrated that both data series are free from the unit-root problem at first difference. The lag length criterions select the third lag for model fitness, and Johansen cointegration declares that variables are cointegrated for the long term.
The short-run directional nature of variables shows unidirectional causality between work accidents and workload-dependent causes, with the first difference at a 10% significance level. The unidirectional relationship observed between fatal accident and the injured person with one difference at 1% and means of production with a difference of three significant at a 5% level of confidence. A bidirectional relationship was observed between fatal accidents and workload-dependent causes with a difference of one at a 5% level of significance. A unidirectional causality of workload toward fatal accidents with a difference of three is confirmed. The unidirectional relationship between work environment and fatal accidents is observed. The VECM declares the speed of adjustment from the short to the long run is 83.35% and 42.60% for work and fatal accidents. Future research will address the costs of occupational and fatal risks.