Efficiency Assessment Based on Data Envelopment Analysis for Occupational Accidents and Diseases in Furniture Industry of Turkey

• In this study, the provinces in which enterprises operate heavily in the Turkish furniture industry were subjected to an efficiency assessment using the data envelopment analysis (DEA) method based on work accidents and occupational diseases. The data related to the furniture industry was obtained from the Social Security Institution (SSI) and the Ministry of Labor and Social Security (MoLSS). The number of operating businesses and the number of employees insured in each province were evaluated as inputs for the analysis. The total number of workplace physicians and other health personnel in each province, the number of workplace physician and occupational safety centers and A, B, and C-class occupational health and safety specialists were used as outputs. The number of employees who experienced work accidents and occupational diseases, deaths resulting from work accidents and occupational diseases, total temporary disability periods, the number of beneficiaries


INTRODUCTION 1. UVOD
Occupational health and safety (OHS) is an essential element of the business world.At this point, occupational accidents and diseases (OADs) are an issue that particularly concerns employees in many sectors.Sectors such as manufacturing, production, mining, construction, transportation, underground and underwater contain many risks for employees.The issue of OADs is not only a matter of concern for employees, but also affects and concerns employers, the relatives for whom the employee is responsible, and the government.In Turkey, this issue has been taken more seriously since the adoption of the Occupational Health and Safety Law No. 6331 in 2012 within the framework of the European Union (EU) adaptation laws.Although the law contains important provisions for workplaces and employers to take serious measures regarding OHS and to protect employees, and provides for serious sanctions for non-compliance, the rates of OADs in Turkey are still at low levels.According to Serin et al. (2015), in terms of the manufacturing sector, Turkey takes the first place in Europe and second in the world when rankings are examined according to the frequency of occupational accidents.This situation shows the necessity of analyzing occupational accidents seriously in Turkey (Akyüz et al., 2016).
During work life, OHS has a priority that requires a national-level effort involving not only preventive measures taken by employers at the workplace but also contributions from employees and the government.Therefore, as in other countries, various regulations have been implemented and will continue to be implemented in Turkey to continuously improve the working conditions, ensure effective participation of employees in OHS efforts, and plan national-level education on this subject.To measure the effectiveness of such regulations and see their impact, it is necessary to look at the indicators of countries' performance in this area.

KLJUČNE RIJEČI: analiza omeđivanja podataka (DEA); neparametarske metode evaluacije; nezgoda na radu; turska industrija namještaja
Statistical data on OHS and its changes and distribution over the years are essential elements of the business world.
Data on OADs in Turkey are recorded and published by the Social Security Institution (SSI).Until 2013, only accident data for incidents that occurred and were closed within the year were published.However, with the implementation of heavy sanctions for those who fail to report accidents, particularly following Law No. 6331, all reported accidents have been published since 2013.However, it is an issue to be considered that these statistical data only cover the reported OADs.According to Kurttekin and Taçgın (2019), it is known that many occupational accidents are not reported or recorded due to insufficient control and inspection.In addition, identifying occupational diseases is also quite difficult because most occupational diseases emerge either in advanced age or after the worker's retirement, so they are not considered as occupational diseases and are not recorded.Therefore, the accuracy of the data obtained from SSI statistics is accepted as a limitation of the research.
The furniture industry is one of the leading subsectors within the manufacturing industry.According to Magezi and Okan (2023), the Turkish furniture industry has made good use of the opportunities in the last ten years and has become competitive in EU.There are 21,758 businesses in the sector, with 154,829 insured employees (SSI, 2019).Furniture companies are mainly classified as dangerous and highly dangerous based on their risk group.The sector includes many hazardous tools such as cutting, drilling, carving, and crushing equipment, as well as many chemicals that threaten human health, such as wood dust and formaldehyde, paint, and thinner.In addition, there are many heavy tasks and workloads during production and assembly that can harm the musculoskeletal system of workers.
There are many different studies on the subject in the literature.Koç et al. (1998)  This research is aimed to develop suggestions to minimize the work accidents (WA) or occupational diseases (OD) that occur in the provinces where the Turkish furniture industry is concentrated.The provinces determined for this purpose were evaluated in terms of the number of enterprises, employees, workplace physicians (WP) and other health personnel (OHP), and the number of occupational health and safety specialists (OHSS).

Data envelopment analysis (DEA) 1.1. Analiza omeđivanja podataka (DEA)
The data envelopment analysis (DEA) method has been chosen for the evaluation.It is one of the nonparametric evaluation methods.It allows efficiency evaluation with many inputs and outputs, and at the same time, it is easy to include undesirable outputs (UO) in efficiency evaluation with various models.
DEA was developed by Charnes et al. (1978) to measure and compare the technical efficiency of public institutions based on an article by Farrell (1957) on productivity measurement (Safak et al., 2014).DEA is a linear programming-based technique used to measure the performance efficiency of organizational units called Decision Making Units (DMUs).This technique measures how efficiently DMUs use available resources in producing a set of outputs (Charnes et al., 1978).
DEA gets its name from covering observations to set a cutoff value for evaluating investigations that represent the performance of all assets.It can be used to evaluate the performance not only of enterprises but also of public institutions and non-profit organizations such as schools, hospitals, and banks, and also of cities, regions, and countries in various fields (Cooper et al., 2006;Ulutaş, 2016;Depren, 2008).For this purpose, the definition of DMUs has been kept flexible to include any entity that uses similar inputs to produce similar outputs (Cooper et al., 2006).(Pasupathy, 2002).Later, the model was developed by Banker et al. (1984) and the two models, BCC and CCR, have started to be used in the literature.Then, the DEA technique was used and improved by many people (Ulutaş, 2016;Depren, 2008;Charnes et al., 1978;Banker et al., 1984).While the CCR model used the assumption of constant returns to scale only in the measurement of technical efficiency, the BCC model developed the concept of efficient scale and rearranged the linear programming formula in the CCR model for the estimation of the return to scale (Depren, 2008;Besen, 1994).

Classic
DEA is used to determine the "most efficient" or effective decision-making units (DMUs) within a set of DMUs by using the least inputs to produce the most outputs.The accepted efficiency boundary is 1, and the efficiency levels of ineffective DMUs are measured using this boundary as a technical reference.DEA enables obtaining a single efficiency value for each DMU based on multiple input and output variables by using a linear programming model (Depren, 2008).
In a simple DEA model, the efficiency of a random DMU(i) can be defined as:

Weighted sum of outputs for DMU(i) Effectiveness of DMU(i) =
Weighted sum of inputs for DMU(i) The solution of the formula is iterated with linear programming for each DMU under a set of predefined constraints.One of the constraints is the weighting factor.When reaching the solution, the unit is not allowed to choose weights that would cause it to achieve an efficiency greater than 100 %.The weighted sum of the DMUs outputs must be less than or equal to the weighted sum of its inputs.In addition, the weighted sum of the inputs is assumed to be 1 to prevent limitless solutions (Manzoni and Islam, 2009).The principal aim of this equation is the assumption of obtaining the maximum output with the minimum input.
One of the points to be considered for DEA is the number of DMUs.The number of DMUs is crucial as it affects the decision-making ability of the model.For the optimal sample size, the number of DMUs should be greater than the sum of the inputs and outputs.In addition, the sample size is acceptable if the number of fully productive DMUs is not more than one-third of the total number of DMUs in the sample (Cooper et al., 2006;Manzoni and Islam, 2009).

DEA models with undesirable outputs 1.3. DEA modeli s neželjenim izlazima
The DEA approach has a structure that aims to bring the DMU closer to the determined efficiency limit by reducing inputs or increasing outputs after an active limit value is determined in general.However, UO or inputs may also occur in this process (Zhou et al., 2008).UO are the unexpected results of manufacturing processes aimed at maximizing outputs while minimizing inputs.For this reason, the perspective of DEA to approach the effective limit by increasing the output should be handled differently in terms of UO (Seiford and Zhu, 2002).Therefore, many different DEA models have been developed, including UO. Seiford

MATERIJALI I METODE
Sectoral data in the study were taken from SSI 2019 statistical annuals.First, the provinces where the furniture sector was intense were determined using these data.From the insurance and workplace statistics table in statistical annuals of SSI 2019, the data under the furniture manufacturing title with the code 31 in the NACE coding system have been compiled according to the provinces.In 2019, the number of workplaces with SSI registration in the furniture sector was 21.758, and the number of employees was 154.829.In the study, it was determined that approximately 92 % of the insured employees in the sector are concentrated in 30 provinces (Table 2).The number of enterprises in 30 cities constitutes about 88 % of the total number of enterprises (SSI, 2019).For this reason, the research was limited to 30 provinces, as the representativeness of the data is high.
Table 3 presents the number of WP, OHP in the provinces and the numbers of A, B and C class OHSS.These data were taken from the statistical table on the website of the Ministry of Labor and Social Security (MoLSS, 2021).The information on the Ministry webpage is irregular and does not include the distribution by year.That was assumed as a constraint of the research.
The data used in the study have been compiled from the tables in the SSI 2019 Occupational Accident and Occupational Disease Statistics.These data were the number of employees who have had WA and OD, the number of cases of WA and OD, Temporary Incapacity for Work Duration (TIWD), the number of right holders and the number of permanent incapacity income recipients.Table 4 shows the distribution of these data by province.These data include only data reported and officially registered cases.There was no information about the unreported and unregistered data.That is another limitation of the study.
In this study, the DEA Solver add-on included in the Excel program was used for the efficiency evaluation of the data compiled and tabulated for data envelopment analysis.DEA Solver is a very suitable program for the evaluation of data to a maximum of 30 DMUs and it does not need to be purchased separately as it is included in the Excel program.As a method in the analysis, six different model approaches shown in Table 1 were preferred.The abbreviations in Table 5 can be listed as follows: G1 (number of enterprises), G2 (number of employees insured), O1 (total number of WP and OHP), O2 (number of WPOSC), O3 (A, B, and C-class OHSSs), U1 (number of employees who experienced WA), U2 (number of employees who experienced OD), U3 (number of deaths resulting from WA and OD), U4 (total temporary incapacity for work duration), U5 (number of beneficiaries) and U6 (number of permanent incapacity income recipients).

REZULTATI I RASPRAVA
Table 6 shows the distribution of the efficiency scores according to the provinces obtained as a result of DEA, in which UO are modeled with the six different models shown in Table 1.
First, the model of Faere et al. (1989) was used.In the model, UOs are ignored.In the classical DEA, the increase of the outcomes is assumed by keeping the inputs.As a result of the analysis, Aydın, Diyarbakır, Muğla and Zonguldak came to the fore as the most effective cities.Kayseri was the most ineffective province with an efficiency score of 0.1143.The average efficiency of the states is 0.6199.
In the second model, UOs were included in the analysis as inputs, as proposed by Dyckhoff and Allen (2001).In the DEA analysis, UOs were defined as inputs by inverting them according to the addition process (U=-u).As a result of the analysis, Aydın, Diyarbakır, Manisa, Muğla, Tekirdağ and Zonguldak were seen as the most effective provinces.With an efficiency score of 0.1443, Kayseri was again the most ineffective city.The average efficiency score is 0.6339.Seiford and Zhu (2002) stated that including UO as input into DEA like in the second model does not reflect the actual production process.For this reason, in Therefore, in the third model, according to this approach, UOs were improved and included in DEA as output.According to the DEA result, Aydın, Diyarbakır, Kahramanmaraş, Malatya, Muğla and Zonguldak came to the fore as effective provinces.Kayseri was in the last place in the efficiency evaluation as the most ineffective city with an efficiency score of 0.1071.In this model, the average efficiency score is 0.69100.
In the fourth model, as suggested by Dyckhoff and Allen (2001), undesirable factors are transformed into desired outputs by applying a nonlinear monotonous decreasing conversion (U=1/u).Then, they are included in DEA as natural outputs.In Table 6, where the results of the analysis are shown, Aydın.Çanakkale, Diyarbakır, Kahramanmaraş, Malatya, Muğla, Tekirdağ and Zonguldak are seen as the most effective cities.With an efficiency score of 0.1443, Kayseri is in the last place as the most ineffective province.The average efficiency score was calculated as 0.7561.
A linear monotone decreasing transformed model was applied as the fifth model.This model is suggested by Seiford and Zhu (2002).The undesirable factors were improved and added in DEA with natural outputs.As seen in Table 6, Aydın, Çanakkale, Diyarbakır, Düzce, Eskişehir, Kahramanmaraş, Malatya, Muğla, Ordu, Sakarya, Samsun, Sivas, Trabzon and Zonguldak were the most efficient provinces.While the average efficiency score of the provinces was calculated as 0.7871, Kayseri was seen as the most inefficient city with an efficiency score of 0.1443.
According to Yang et al. (2008), UOs can be transformed into desired outcomes by curing them with the Shephard distance function.Hua and Bian (2007) stated in their study that undesired factors can be transformed into desired outputs by using the directional distance function and that DEA results can better reflect the actual production process.This approach model of Hua and Bian (2007) was used as the sixth model.According to the DEA result in Table 6, Aydın, Çanakkale, Diyarbakır, Eskişehir, Malatya, Muğla, Samsun, Trabzon and Zonguldak were the most influential cities. Kayseri took the last place again as the most ineffective province with an efficiency score of 0.1447.The average efficiency score is 0.6844.
Coping with UO is very important in DEA.In this study, the data with UO were improved and transformed according to six different models in Table 1.Then, they were included in DEA and subjected to an efficacy evaluation.Table 6 shows that Model 2, Model 3 and Model 6 produce similar results in terms of DMUs.When Table 6, Table 2, Table 3 and Table 4 are evaluated together, it can be observed that optimal results in terms of efficiency are obtained with the Model 6. Hua and Bian (2007) state that in this model, since the direction vector is strongly affected by the weights determined by the users, optimum results for DMUs can be achieved more effectively.
According to the OHS Law, an OHSS must be assigned to every 1000 employees for the less hazardous class, every 500 employees for the dangerous group, and every 250 employees for the very precarious class.Likewise, there is a requirement to appoint a workplace physician for every 2000 employees in the less dangerous business class, every 1000 employees in class, and every 750 employees in the very precarious enterprise.According to the current law, these numbers in the sector are sufficient.
The 30 provinces within the scope of the study are about 88 % of the total number of enterprises or employees in the furniture sector.Also, they cover approximately 84 % of the total number of enterprises or employees in Turkey (Table 2).Istanbul is the city with the highest number of enterprises and employees countrywide.One out of every 38 employees in Istanbul had a work accident.One out of every 20757 employees died in a work accident (Table 2 and 4).Besides, the number of employees per one WP or OHP in Istanbul is approximately 380.The number of enterprises per  2 and 3).As a result of DEA, Aydın, Diyarbakır, Muğla, and Zonguldak emerge as the most effective provinces in all models.Zonguldak is a significant city in both the furniture and mining sector.One of every 16 employees in Zonguldak has a work accident.One of every 8685 employees died in a work accident (Table 2 and  4).The employees' number per WP or OHP in Zonguldak is about 263.The number of enterprises per workplace physician and occupational safety center is determined as 245 (Table 2 and 3).
The death rate due to work accidents or occupational diseases in Turkey is 8.01 per 100,000 employees.The average of 30 provinces is 7.27.In the European Union (EU), the average rate is 1.77.In Romania, where the highest number of deaths occurred, this rate is 4.33, or in Luxembourg 4.22.When examining the ratio of work accidents, while it is 2951 per 100,000 in Turkey, the EU average is 1650.This rate is around 3450 in France.The average of 30 provinces within the scope of the research is 3100 (Eurostat, 2021), (Table 2 and 4).

ZAKLJUČAK
Previous studies show that the work environments of enterprises in the furniture sector are not suitable enough in terms of OHS.For example, according to Koç et al. (1998), 59 % of the enterprises were not adequately ventilated.Likewise, Karademir and Koç (2020) found in their study that 30 % of the enterprises were completely without air, while 40 % were not properly ventilated.Although 20 years elapsed between the two studies, it seems that the work environments of the enterprises have not been brought to a good point in terms of OHS.At this point, it is known that the sector enterprises should be effectively vised and kept under control, especially in terms of employees' safety and health.
In the study, according to Model 6, 9 of 30 provinces were found to be effective and 21 of them were found to be ineffective.The most ineffective is Kayseri with an efficiency value of 0.1447.At this point, Kayseri should reduce the number of its employees by 10 % compared to Diyarbakır, for example, and 14 % compared to Zonguldak.Likewise, it should increase the number of WPOSC by 40 % compared to Diyarbakır and by 52 % compared to Zonguldak.
OHSSs' field tours, risk analyses, environmental measurements, periodic inspections and workups, and feedback from employees are very important for accelerating the detection and improvement of malfunc-tions in the workplace (Ulutaş. 2016).Reports and records kept by the occupational physician, other health personnel, occupational physician and occupational safety centers and OHSSs increase the accuracy and reliability of statistical data on occupational accidents and diseases.
Although the number of OHSSs seems sufficient according to the current laws, the data in the field shows that these numbers are insufficient to prevent OADs.Despite being the most efficient province in all analyses, the number of employees per WP is high even in Zonguldak.Herein, according to the hazard classes, work accidents that have occurred from the past to the present should be examined.According to accident frequency, causes of accident and death statistics, a sectoral status evaluation should be made in the enterprises.Optimal numbers should be determined according to assessment, and workloads should be recalculated.The relevant articles of the OHS law should be updated.
Increasing numbers and reducing workloads will play a significant role in preventing occupational accidents and deaths.However, it is not enough by itself.At this point, training programs with the participation of all stakeholders such as business owners, white and blue-collar employees, WP, OHP and OHSSs should be planned and periodically implemented.
Up till today, full cooperation between industry and universities has not been established in practice.For having a good collaboration, a legal basis must be established between the parties, and this legal ground must be a forcing factor.Universities must be involved as a partner in projects preparation, training, auditing and consultancy services.
According to Lombardi et al. (2019), learning from accidents is one of the steps that can be taken to prevent possible future accidents.The most substantial step for the learning approach is to create a detailed accident database on a sectoral basis across the country.The variables of all work accidents that occur should be determined, and the variables interrelationships should be calculated.In this way, the most probable accident causality models for each accident should be revealed and the accident proximal and distant specific features should be defined.As a result, risk profiling for each accident will provide foremost and practical information for the safety precautions.
As to the European Agency for Occupational Safety and Health (OSHA), the key to sustainable economic recovery is the employees' well-being.Moreover, OSH is seen as a significant aspect of responsible and sustainable development of rapidly growing new technologies (nanotechnologies and green jobs).Therefore, there is a need to establish an OHS culture, namely a cultural evolution, to increase the value of occupational health among an extensive range of stake-holders, from decision-makers to OHS experts, employers, workers and their representatives, and ultimately the whole society (Gagliardi. 2012).
examined OADs in the furniture industry and developed recommendations for their prevention.Akyüz et al. (2016) studied accident statistics in the forest products industry sector and made general evaluations regarding the issue.Gedik and İlhan (2014) analyzed OHS issues among furniture manufacturers in Sakarya province.Sevim Korkut and Gedik (2010) examined occupational safety in the Turkish forest products industry sector.Ulutaş (2016) evaluated provinces in Turkey in terms of OADs.Bilim and Bilim (2015) conducted a statistical study on OHS in Turkey and analyzed the issue.Lombardi et al.
and Zhu (2002) or Faere et al. (1989) can be shown as the most widely known and used DEA models.Whereas Seiford and Zhu's (2002) model focuses on data transformation, Faere et al. (1989) show an approach that ignores UO (Zhou et al., 2008; Seiford and Zhu, 2002).Table 1 lists six specific models that are most frequently used and recommended in the literature.

Table 2
Distribution by provinces with intense furniture industry, number of enterprises and employees by provinces Tablica 2. Pregled pokrajina u kojima je industrija namještaja bila intenzivna te broj poduzeća i zaposlenika u njima

Table 5
Models used in DEA analysis and their inputs and outputs Tablica 5. Modeli primijenjeni u DEA analizi te ulazi i izlazi u tim modelima