Malay Validation of Copenhagen Psychosocial Work Environment Questionnaire in Context of Second Generation Statistical Techniques

Psychosocial hazards present in workplaces are being actively investigated by researchers from multiple domains. More research and resources are required to investigate the debilitating consequences of these hazards in the developing and underdeveloped countries where this issue remains one of grave concern. This study aims at investigating the psychometric properties of Malaysian version of Copenhagen Psychosocial Questionnaire for reliability and validity purpose. The Malaysian version of COPSOQ is a multidimensional questionnaire; it comprises of 7 major formative constructs and 28 variables with an additional inclusion of two variables which are organizational loyalty and physiological health biomarkers (blood pressure and body mass index) that explicate a reflective construct which has 93 items all catering to assess psychosocial determinants present in workplace environments. Each formative second-order construct is further categorized into different reflective first-order constructs. The focus of this study was only on first-order reflective constructs. Probability sampling was used for data collection from 300 respondents working in industries with a response rate of 100%; structural equation modeling technique was applied for data analysis. All psychometric analysis performed on reflective constructs gave reliable results which demonstrate the validity of Bahasa Melayu (BM-COPSOQ) and its comprehensiveness of including relevant dimensions particularly in context to Asian region. The BM-COPSOQ will fill up the knowledge gap and provide a bridge between researchers, work professionals and practitioners, and many other workplaces for the best understanding of psychosocial work environment.


Background
Work, regardless of its nature involves certain occupational risks. In the backdrop of rapidly changing economic circumstances, risks present in workplace settings are becoming more apparent. Previously, biological, biomechanical, chemical, and radiological risks were mostly considered critical however in recent times psychosocial risks have also garnered serious attention of researchers. The developing and underdeveloped countries have displayed lack of awareness towards these emerging risks and much research needs to be done to tackle these perils in workplace settings [1].
Petrochemical industries by virtue have hazardous work environments that effect the physical and psychosocial condition of its employees. The Malaysian petrochemical industries are labor-intensive industry that presents many
Majority of the validation studies conducted in the western countries used first generation statistical techniques with limited in-depth analysis. The studies conducted in eastern countries are also limited. In fact, there is no validation study of COPSOQ in the Malaysian context which has used for robust statistical techniques. Therefore, the aim of this research article is to present the Malaysian version of COPSOQ by analyzing the psychometric properties of the instrument. Structural Equation Modeling (SEM), a second generation statistical  [22,23] the names of the petrochemical industries were put in a jar, thoroughly mixed, and the required sample, which is one joint venture and one fully owned firm, from Terengganu and Pahang was randomly drawn. First, the joint venture and fully owned industries in Terengganu were added in the lottery technique followed by Pahang industries that were entered and selected based on the lottery technique.

Selection of Study Subjects from Petrochemical
Industries. The Simple Random Sampling technique was used to collect data from the provided list of the study subjects. The subjects in three industrial zones were equally divided, which means that 50 subjects from each petrochemical industry were chosen as shown in Table 1. To have a maximum representation of subjects from each targeted industry, both morning and evening shifts were targeted from the provided list. Then, 25 subjects were randomly selected from the morning shift and the remaining 25 subjects were selected from the evening shift from the selected industries. The data from all three industrial zones were collected during normal working days over a period of one month, May 2016.

3.3.
Questionnaire. The study constructs were adapted from the second version of COPSOQ II [24], a thorough questionnaire that covers all the aspects that are important to study the psychosocial work environment along with health and wellbeing [16,25].  Table 2.

Translation of Questionnaire.
The current study was conducted in Bahasa Melayu (BM) the national language of Malaysia; therefore, all the study variables were translated into BM from English using the back translation technique [26]. The forward-then-back translation procedure was completed in multiple steps. Translation and back translation of the internationally recognized base questionnaire into BM were carried out with the help of two certified translators located in Kuala Lumpur, Malaysia. In the first step, the English version was translated into BM by one certified translator, and, in the second step, the back translation from BM to English was done by another certified translator. To retain the originality and authenticity of both translations, the two selected translators (unknown to each other) worked independently. To ensure that the contents of each item were cross-linguistically comparable and generated the same meaning, the researchers used both translated languages in a single questionnaire.

Quantitative Demands.
Quantitative Demands (QD) was measured by a 4-item scale, coded by QD1, QD2, QD3, QD4, having items like "does your workload pile up due to uneven distribution?" translated into BM "Adakah beban kerja anda semakin bertimbun disebabkan pembahagian tidak sekata?"   (MAP) is defined as average blood pressure in an individual during a single cardiac cycle as shown in the following equation: In this equation, SBP is the systolic blood pressure and DBP is the diastolic blood pressure. The unit of mean arterial pressure (MAP) is measured in mmHg. MAP is used to approximate the pressure gradient (ΔP) of the subjects and includes the effect of systolic and diastolic pressure. Measurements from left arm was taken and measured as "Systolic blood pressure reading 1, left arm" and "Diastolic blood pressure  Table 2. However, in this study, we only aimed to examine the first-order reflective constructs for psychometric properties via applying the second generation tool, i.e., PLS-SEM. The firstorder reflective constructs of Copenhagen are discussed in Table  2 under dimensions section. In order to examine the quality of reflective constructs in terms of reliability and validity, we used the following standards as suggested that the item loading should be greater than 0.60 or at minimum value of 0.40, the Cronbach's alpha and composite reliability are to be at 0.70 or greater, and the convergent validity also termed average variance extracted (AVE) must exceed the value 0.50.
Another criterion to examine the reflective construct is discriminant validity which is defined as to what extent each latent variable is different from other variables in a study model. Few authors added that the AVE of each variable must be greater than the highest squared correlation of variables with any other latent variables in the model to form discriminant validity [29,30].
The study analyses were conducted using WarpPLS 6.0 [31] software as it provides us many options for the assessment BMI as endogenous non-invasive biomarker variable was calculated with a single item. Weight was measured in kilograms "How much do you weigh?" translated into BM "Berapakah berat anda?" Height was measured in meters "How tall are you?" translated into BM "Berapakah ketinggian anda?" 3.4. Statistical Approach. In this study the complexity of the model is high and therefore we have used Partial Least Squares Structural Equation Modeling (PLS-SEM) technique to evaluate the psychometric properties of Copenhagen psychosocial work environment questionnaire [29]. They further argued, a model involving reflective and formative constructs is reflected as a multifaceted measurement model. Copenhagen scale consists of reflective constructs on first-order level and formative constructs on second-order level. For instance, demands at work is a second-order formative construct which is based on three first-order reflective constructs such as emotional demands, work pace, and quantitative demands as shown in  We have added a construct "further parameter" having two variables in it which are "organizational loyalty" and "biomarkers" Organizational loyalty is measured by three items while biomarker is made up of blood pressure and body mass index which are physiological health measures. The details of measurement of each physiological variable are already explained in Sections 3.3.29 and 3.3.30. Organizational loyalty (Intention to leave) is predicted as another outcome of working condition which results from psychosocial hazards.
Several strengths of this study need to be highlighted. Firstly, the inclusion of two physiological health biomarkers (BP and BMI). In behavioral studies, the use of biomarkers in context to the psychosocial work environment factors is largely lacking. The inclusion of two non-invasive health biomarkers was used as a screening tool to measure the physiological health of the workers besides psychological such as stress and burnout in the psychosocial work environment. The screening tools will help workers to keep updated with their health conditions due to psychosocial risks emerged around them. Secondly, the use of SEM for evaluating the psychometric properties of BM-COPSOQ and to ensure the robustness of the results which eventually we have found by achieving the validity and reliability of the items well above their minimum cut off values. Thirdly, the probability sampling technique to reach out the technical samples working in different states of highly hazardous petrochemical industries ensured greater confidence in results.
There were some limitations to this study. Firstly, the study sample is based on workers working only in the petrochemical industries of Peninsular Malaysia which should be expanded to different regions, branches and professions in the country. Secondly, this study is limited to only evaluate the psychometric properties of BM-COPSOQ. Our upcoming research article focuses on COPSOQ III where we have thoroughly evaluated the impact of all the higherorder constructs such as Demands at Work; Work Organization and Job Content; Interpersonal Relations and Leadership; Work-Individual Interference; & Health and Wellbeing (inclusive of health non-invasive biomarkers) in lieu to CORE and additional items. These higher-order constructs will be used as reflective-formative constructs to widen the scope of COPSOQ [33,34].

Data Availability
Data of 277 study samples used will be provided on request.

Conflicts of Interest
The authors declare no conflicts of interest regarding the publication of this paper.

Acknowledgments
Center for Organizational & Social Research in Energy (COSRE). of outer model parameters and calculating the latent variables scores (LVs). Using WarpPLS 6.0, we performed the algorithm, i.e., PLS regression for outer model to assess the first-order reflective measurement (outer) model parameters and LVs. The results of first-order reflective measurement model comprising factor loadings, Cronbach's alpha, composite reliability (CR), AVE, and discriminant validity are discussed in Tables  3 and 4, respectively. 3.5. Results. Using the WarpPLS, we assessed the first-order reflective dimensions for their reliability and validity in terms of indicator loadings, Cronbach's alpha, composite reliability, convergent validity, and discriminant validity. The results shown in Table 3 reported that the indicator loadings have exceeded the critical value of 0.40 to retain an item. However, few items did not meet threshold value of 0.40, for instance, QD4 of quantitative demands and T1, T2, and T6 of trust variable. Next, we checked the reliability tests comprising Cronbach's alpha and composite reliability and found that all constructs are reliable as they met the critical value of 0.70. We also evaluated the convergent validity via using the criteria of AVE and resulted that all reflective constructs have achieved the AVE value of 0.50, hence verifying that all constructs had met the requirement of convergent validity, see Table 3.
Another criteria to examine the reflective constructs of BM-COPSOQ, is by means of discriminant validity. While assessing the discriminant validity, we compared the square root of AVE with the correlation of latent variables. As a result, Table 4 showed that there is no discriminant validity issue as the square root of AVE of all constructs is greater than the correlation of other variables as shown in Table 4.
We also calculated the full Collinearity (FVIF) which refers to the vertical and lateral Collinearity of one construct in association with other variables [32]. They further suggested that FVIF is another source to establish the discriminant validity and the critical value of FVIF should be equal to or less than five. As shown in Table 3, we have found that none of FVIF is greater than the threshold value.

Discussion and Conclusion
We have evaluated the psychometric properties of BM-COPSOQ using a sample of Malaysian Petrochemical Industry workers with the help of probability sampling for data gathering and SEM technique for data analysis. Main goal of this study was to provide a valid and reliable psychosocial work environment questionnaire for an eastern industrially developing country Malaysia in Bahasa Melayu.
The BM-COPSOQ is a standardized self-report measure which is comprised of 7 constructs having 28 variables with 93 items designed for the assessment of psychosocial workplace environment. The details of each construct followed by variables and number of items are already summarized in aforementioned Table 2. The 4 items, i.e., one from the "quantitative demands" and 3 from "trust" were dropped due to low factor loadings. The composite reliability of all the items is well above 0.7 so as average variance extracted which is more than 0.5 for all variables.