4.1 KMO test
The sample size adequacy test was determined using Kaiser-Meyer-Olkin (KMO) and the value for this study was 0.9474 which is higher than the average cutoff points of 0.7 for sample adequacy for further analysis (Klein & Dabney, 2013). This result indicates the presence of a strong partial correlation among variables that demand factor analysis. Therefore, a KMO value of 0.9474 is sufficient and adequate to conduct factor analysis as presented in Table I below
As can be seen from Table I above, test statistics values of chi-square, and degree of freedom are 5635.00 and 197 respectively with a significance level of p-value 0.000. This infers the data is sufficient for carrying out factory analysis and it is conducted here below.
4.2 Exploratory Factor Analysis (EFA)
Exploratory factor analysis by principal component analysis was conducted to discover factor structure in the data using stata14 software. In total, 33 questions were used in the process and grouped under seven concepts as theoretically identified before. The components with more than 1 eigenvalue were retained and questions with correlation values higher than 0.3 were rotated and assigned for respective components. Table II below shows loadings as it assigned to each factor.
The result in Table II is an exploratory factor analysis for this research resulting in seven factors as empirically identified six for human resource management and one for employee engagement. From exploratory factor analysis, human resource management practices extracted were: Recruitment and selection as identified by Bratton and Gold (2017), training and development (Nadarajah, Kadiresan, Kumar, Kamil, & Yusoff, 2012), performance appraisal (Govender & Bussin, 2020), communication and supervisor support (Arimie, 2019), reward and recognition (Alfiyah¹ & Riyanto, 2019) and work-life balance (Jaharuddin & Zainol, 2019). Employee engagement construct was identified as one factor and all questions were loaded on it. After confirmation of item representativeness of the construct; further validation of the items was conducted in confirmatory factor analysis.
4.3 Measurement model for HRMP construct
4.3.1 Confirmatory factor analysis
For the assessment of human resource management practices, six sub-variables (Recruitment and selection, Training and development, Performance appraisal, Communication and supervisor support, Reward and recognition and Work-life balance) were used. Data collection instruments for human resource management practices were prepared based on those sub-variables. For each sub-variable representing human resource management practice; the number of questions (items) were used to collect data, but at the preliminary screening stage and based on test results; questions were filtered and only four questions (items) for each sub-variable were used for analysis. The test results summarized in Table III below reveal that instruments were well fit for all assessed goodness of fit statistics such as construct reliability (composite reliability and internal consistency), construct validity (convergent and discriminate validity) and model fitness indexes
It can be seen from Table III above that all items satisfy the minimum requirements for validity and reliability. For internal consistency, Cronbach’s alpha values are highly above the required minimum cut point of 0.7 as all items have alpha values of more than 0.93. Composite reliability also meets the minimum requirement since factor loadings of all items are more than or equal to 0.7. Almost all squared values of factor loadings are higher or equal to 0.5, which indicates the presence of composite reliability.
The average variance extracted indicates the presence of convergent validity since the values of the average variance extracted are far higher than the minimum cut points of 0.5. The values of AVE for this construct lie between the ranges of 0.56 to 0.71. The squared values of correlation between two variables representing construct are lower than the values of AVE and it indicates the presence of discriminant validity. The existence of discriminant validity for this data can be seen from the square root of AVE comparing with correlation values between latent construct. Even though correlation coefficients between some variables are higher than 0.60, it can be accepted since the AVE of the respective construct is greater than correlation values. The values in Figure 1 below confirm this fact
4.3.2 Assessments of model fitness for HRMP construct
Model fitness shows how closely the estimated covariance matrix matches the observed (sample) covariance matrix. The measure of model fitness is estimated for three models: the estimated model, the saturated model and the independence model under three measurements of fitness in SEM as measures of absolute fit, measures of incremental fit and measures of parsimonious fit (Gupta, 2015). Scholars put forward that no single statistical test best describes the strength of the model’s predictions and suggested employing different indices developed over time (Newsom, 2020; Yaşlioğlu & Yaşlioğlu, 2020). Therefore, indices under each measure of fitness in SEM as absolute fit and incremental fit were employed for this study. From absolute fit indices χ2, goodness-of-fit index (GFI), root mean square error of approximation (RMSEA), and standardized root mean square residual (SRMR) were used. TLI and NFI (CFI) were used from incremental (relative) fit indices.
The summarized values in Table IV below show the fitness of indices used to test the model goodness of fit
As can be seen from figures in the Table IV above, all comparative fit indexes show a model good fit. Both CFI and TLI have values above critical cut points. Among the absolute fit indexes SRMR, RMSEA and CMIN/df show model goodness of fit. Only chi-Square (χ2 statistic) shows a model bad fit, but it wasn’t considered for model acceptance for the data because of the sensitivity it has with sample size
4.3.3 Measurement model of Employee engagement
For the assessment of employee engagement as an endogenous variable for this study; three sub-variables (Vigor, Dedication and Absorption) were employed. Vigor describes the willingness to invest all of one’s self into work and refers to high levels of conscientiousness, persistence, energy, and mental toughness. Vigor is used in this study to measure the cognitive involvement of employees which is the central life interests of employees (Al-dalahmeh, Khalaf, & Obeidat, 2018), whereas, dedication refers to being strongly connected to one’s work while experiencing a sense of significance, pride, enthusiasm, and challenge. Dedication is used in this study to measure emotional involvement which is the conscious desire and choice to participate in work and consider work to be central to their self-concept (Muldoon, Keough, & Liguori, 2017). Absorption implies being involved deeply in one’s work, such that time passes quickly and disconnecting from work becomes difficult. It is used for this study to measure behavioral involvements such as timeliness, attendance, effort, say, stay and strive (Iddagoda & Opatha, 2020)
Validity and reliability tests are fit for all items as can be seen in Table V below. The figures in Table 5 show that all items have Cronbach’s alpha values greater than 0.83 which is an indication of items' internal consistency.
It can be predictable from Table V above that factor loadings of all items were greater than or equal to the minimum requirement of 0.7 with average variance extracted values greater than 0.60 which are the indication of the existence of composite reliability and convergent validity. The correlation values between each variable are far lower than the squared root values of AVE of each construct and/or squared values of correlation between construct are also far below AVE values of each construct. Therefore, the discriminant validity of the construct is fit. The graph 2 below confirms the values in the table V described above
4.3.4 Assessments of model fitness for employee engagement construct
For this model all absolute fit indexes SRMR, RMSEA and CMIN/df show model goodness of fit except chi-Square (χ2 statistic) which shows model bad fit, but it wasn’t considered for model acceptance for the data because of the sensitivity it has with sample size
The values in Table V show that each fit index has a value of the above cut points and this confirms our model goodness of fit. It can be observed from figures in the Table 6 above, that all comparative fit indexes show a model good fit. Both CFI and TLI have values above critical cut points. Among the absolute fit indexes SRMR, RMSEA and CMIN/df show model goodness of fit. Only chi-Square (χ2 statistic) shows a model bad fit, but it wasn’t considered for model acceptance for the data because of the sensitivity it has with sample size
4.3 Path analysis
Path analysis is preferred by many researchers because of the ability it decompose causal relationships between direct and indirect effects by employing all possible causal relationships across all the variables in the model. The path diagram links the network of relationships of several variables placed in sequence as independent, intervening and dependent variables to be studied in the research. Path analysis used a standardized coefficient called 'beta weight (β) which indicates the direct influence of an exogenous variable on an endogenous variable in a particular path model. Residual errors in path analysis unlike error variance in CFA do not correlate with all the variables in the model under investigation and are assumed to be normally distributed and have a mean of zero.
The standardized path coefficient has been converted to a standard Z value that allows the researcher to compare the relative strength of the effects of all the different exogenous variables in a particular path model. In addition to standardized path coefficients, there are residual errors reflecting unexplained variants or the influence of all the non-measurable exogenous variables directly plus the measurement errors which reflect the cause of the unknown variability in the analysis results. Unlike in the regression model; the path model reflects three (direct, indirect and total) effects. The direct effect can be seen from the path coefficient of one exogenous variable to an endogenous one as the same as regression beta (β) coefficients. Whereas, the indirect effect is unique to the path model which represents a sequence of paths through one or more intervening variables. To get the value indirect effect, it can done by multiplying the path coefficient from the exogenous variable to the intervening one with the path coefficient from that intervening variable to the endogenous variable. The total effect is the sequence of paths through one exogenous variable to the intervening one plus from that intervening variable to the endogenous one. Getting the total value of the effect can be done by adding the direct effect with the indirect effect
The goodness of the model fit with the theoretical one in path analysis is determined by the simultaneous effect contributed from all exogenous variables onto the endogenous variables whose value is called R2. The R2 value in path analysis is known by the coefficient of determination which is also referred to as the association index. This value is used as a value scale to express the magnitude of the effect of all exogenous variables on endogenous variables simultaneously or referred to as the combined effect
Under this model, the relationship between human resource management practices and employee engagement was examined. All parameters used to investigate the effects of human resource management practice are statistically significant except reward and recognition which have an insignificant value of (P>0.05). All five study variables (recruitment and selection, training and development, performance appraisal, communication and supervisor support and work-life balance have a statistically significant effect on employee engagement at (P<0.001). From the results in Table VII below, it can be seen that the coefficient of determination for the model is 0.75 which indicates 75% of explaining power of independent variables' joint effect together on the variation in the dependent variable of the study
From the values in Table VII above, we can summarize that an improvement in recruitment and selection by one standard deviation leads to a 0.29 standard deviation improvement in employee engagement level assuming another variable has not changed. Like ways one standard deviation improvement in training and development would result in a 0.26 increase in employee engagement keeping all other variables unchanged. Keeping other variables constant, an increase in performance appraisal in one standard deviation would result in a 0.19 standard deviation increase in employee engagement. One standard deviation increase in communication and supervisor support led to a 0.21 standard deviation increase in employee engagement holding other variables constant. If other variables were not changed, one standard deviation increase in work-life balance would result in a 0.12 standard deviation increase in employee engagement.