Confirmatory Factor Analysis (CFA) For the Instrument of Personality, Safety Climate and Safety Performance in The Malaysia Manufacturing Sector

Confirmatory Factor Analysis (CFA) For the Instrument of Personality, Safety Climate and Safety Performance in The Malaysia Manufacturing Sector. (17), – Abstract This paper aims to draw on the application of Confirmatory Factor Analysis (CFA) in Structural Equation Modeling (SEM), to test the validity and reliability of instruments in the study of personality, safety climate, and safety performance in the Malaysia manufacturing sector. Exploratory factor analysis (EFA) was employed to determine the best sub-factors and items for the instrument, while confirmatory factor analysis (CFA) was performed to test and validate the measurement model. Confirmatory Factor Analysis (CFA) using Structural Equation Modeling (SEM) Partial Least Square (PLS), has been used to test the validity and reliability of the instruments. Various tests i.e., construct validity analysis, construct reliability, validity convergent as well as discriminatory validity to filter the best items that can represent the constructs in the study. Results from CFA indicated that two items from the Safety Performance Scale (SPS) had to be discarded to confirm that the model was fit. Meanwhile, all items from the Safety Climate Scale (SCS) and Mini-International Personality Item Pool (IPIP) were maintained. Overall, the final version of the instrument consisted of Safety Climate Instruments (46 items), Big Five Personality Instruments (20 items), and Safety Performance Instruments (37 out of 39 items).


Introduction
Research instruments refer to the tools used by researchers to collect data. For quantitative data, the questionnaire is usually used as an instrument to collect data as in this study. To measure the personality among employees of production operators in the manufacturing sector (electrical and electronics) the researcher used a questionnaire Mini-International Personality Item Pool (IPIP) by Donnellan et al (2006). This Mini-IPIP questionnaire comes from a 50-item questionnaire International Personality Item Pool Five-Factor Model (IPIP-BF) by Goldberg (1999) which was later shortened to 20 items (Baldasaro et al., 2013). Next, the safety climate in this study was measured using the Safety Climate Scale (SCS) by Wu et al (2007). This instrument has 46 questions and consists of five dimensions namely CEO safety commitment and action, manager safety commitment, employee safety commitment, perceived risk, and emergency response. Safety performance is measured using the Safety Performance Scale (SPS) by (Wu et al., 2008). This scale has 39 items with six dimensions namely safety organization and management, safety equipment and measures, safety training practices, safety training evaluation, accident investigation, and accident statistics.

Methodology
In this study, the researcher has tested the measurement model on the study instrument through SEM PLS (Partial Least Square) analysis, using SmartPLS software version 3. Analysis of the measurement model on this study instrument was conducted by performing construct validity analysis, construct reliability, convergent validity as well as discriminatory validity to filter the best items that can represent the constructs in the study.

Construct Validity Analysis
To analyse the validity of the construct, the researcher has used the factor loading to test whether all items are loaded to the appropriate factors or not. According to Greeno et al., (2007) there are various suggestions to assess the acceptable level of factor loading based on the literature. For this study, the researcher set the factor loading value for each item that must exceed the value of 0.50 as suggested by (Hair et al., 1998). According to Hair, et al (1998) sample size of more than 120 is practical to use cut-off loading factor > 0.50. Therefore, researchers have adopted this criterion in testing the results of factors loading. The validity analysis of the construct based on the factor loading for each study variable is as follows.

Construct Validity Analysis for Safety Climate Instruments
In this study, safety climate variables are measured using Safety Climate Scale (SCS) by Wu et al., (2007) through five dimensions namely CEO safety commitments and actions, manager safety commitments, employee safety commitments, perceived risk, and emergency response. Researchers have tested the factor loading for each item in the five dimensions. The results of the analysis found that all items in the five dimensions of safety climate have been loaded to factors that match the value of the acceptable factor loading which is between 0.76-0.89 for the dimensions of CEO safety commitment and action, 0.79-0.89 for the dimensions of manager safety commitment, 0.79-0.89 for the employee safety commitment, 0.72-0.90 for the perceived risk dimension and 0.71-0.84 for the emergency response dimension. Table 1 below shows the factor loading values for all dimensions in the safety climate above the value of 0.50. Therefore, the researcher chose to retain all items in the safety climate dimension for further analysis.

Perceived risk
While working I will not fall 0.82 While working I will not be electrocuted 0.89 While working I will not be stuck by the machine 0.88 While working I will not be exposed to the extreme heat of the work environment 0.84 While working I will not be touched by harmful substances 0.90 While working I will not be exposed to infectious substances (such as bacteria and viruses) I know where the fire extinguisher is placed 0.80 I know where the first aid facility is located 0.82 I know the emergency route clearly 0.71 I know the procedure that needs to be done to deal with the problem of electric shock 0.83 I know the procedure to take in the event of a fire 0.84

Construct Validity Analysis for big Five Personality Instruments
Personality in this study was measure using the questionnaire Mini-International Personality Item Pool (IPIP) by Donnellan et al., (2006). The Mini-IPIP is divided into five traits namely openness to experience, conscientiousness, extrovertness, agreeableness, and neurotics. The researcher performed the factor loading analysis on the five personality traits as shown in Table 2. The results of the analysis found that all items in the five personality traits have a factor loading value exceeding 0.50. The factor loading values between 0.80-0.87 for openness to experience, 0.79-0.85 for conscientiousness, 0.75-0.84 for extrovertness, 0.75-0.86 for agreeableness, and 0.79-0.84 for neurotics. Therefore, all items in the personality trait have good construct validity for further analysis.

Construct Validity Analysis for Safety Performance Instruments
Safety performance in this study was measure using the Safety Performance Scale (SPS) by Wu et al (2008). Table 3 shows the factors loading for the safety organization and management dimensions of safety between 0.18-0.69, safety equipment and measures between 0.47-0.85, safety training practices between 0.86-0.91, safety training evaluation between 0.85-0.93, accident investigation between 0.88 -0.93 and accident statistics between 0.83-0.92. Researchers found that the factor loading value for all items exceeded the value of 0.50 except for two items, namely the item is from the safety organization and management dimension that is "my management states in writing their safety policy" and the item from the dimensions of safety equipment and measures that are "the pathway is always neat and orderly" which shows the factor loading values of 0.18 and 0.47 respectively. The value of factor loading less than 0.50 indicates a weak construct validity. This is in line with the recommendations by Hair et al (1998); Kline (2015), the value of factor loading must be > 0.50. Therefore, researchers have decided to remove these two items for further analysis. The factor loading value for safety performance after removing these two items is 0.53 to 0.93. The factor loading value of 0.53-0.93 proves that the items in safety performance exceed the factor loading conditions proposed by Hair et al (1998); Kline (2015), and can be used for further analysis. The factory provides appropriate safety regulations 0.58 The factory has procedures for managing workplace safety 0.53

Safety equipment and measures
The pathway is always neat and orderly 0.47 (remove) The

Construct Reliability Analysis
Based on previous studies involving PLS-SEM analysis, construct reliability measurements are usually tested through internal consistency methods (Straub et al., 2004). Internal consistency analysis was determined by Cronbach Alpha (CA) and Composite Reliability (CR) values (Hair et al., 2012). According to Hair et al (2011), CA and CR values must be equal to or greater than 0.70 to achieve internal consistency reliability. After analysis, the researcher found that the value of CA (0.79 -0.96) and the value of CR (0.84 -0.96) in this study exceeded the value of 0.70 as suggested by (Hair et al., 2011). Therefore, the researcher confirmed the existence of internal consistency in this study as in Table 4.

Convergent Validity Analysis
Convergent validity refers to the extent to which a group of items represents a construct (Gefen & Straub, 2005). In PLS analysis, convergent validity is measured through the value of Average Variance Extracted (AVE) (Urbach & Ahlemann, 2010). Hair et al., (1998) suggested that the AVE value for each item in the construct must be > 0.50. In this study, AVE values for safety climate (0.65 -0.74), personality (0.64 -0.71) and safety performance (0.69). In this study, the AVE value is above 0.50 as in Table 5. Based on the value of this AVE, this study proves the existence of the convergent validity of the instruments in this study.

Discrimination Validity Analysis
Discriminatory validity refers to the extent to which each construct differs from the other construct. In PLS analysis, discriminant validity was measured using a cross-loading method (Henseler et al., 2015). The cross-loading method is made between the construct and the items in the model. The assessment made is that the cross-loading value of all items measured with the matched constructs should have a higher value than the matching values with other constructs (Grégoire & Fisher, 2006;Hair et al., 2011;Henseler et al., 2015). In this study, the cross-loading value of each item matched to a particular construct was found to be higher than the unmatched construct as shown in Table 6. Thus, it proves the existence of discriminatory validity in this study.