RESEARCH METHODOLOGY
3.5 Data Analysis
year = 15.8%, one to three years = 40.1%, three to five years = 18.9%, five to 10 years
= 10.1, and more than 10 years = 4.2%. The manager levels were as follows: 39.1%
executive or senior level, 31.3% middle management, and 29.6% first-line management. Organizations with codes of conduct comprised 85% of the total. Table 3.14 summarizes the demographic data of the participants’ supervisors and organizations.
Table 3.14 Demographic Features of Participants’ Supervisors (n = 862)
Variable Frequency Percentage
Sex
Male 519 60.2
Female 343 39.8
Position
First-line manager 255 29.6
Middle management 270 31.3
Senior management or executive level 337 39.1 Span of control
Less than five employees 309 35.8
Five to seven employees 162 18.1
Eight to 10 employees 103 11.9
More than 10 employees 288 33.4
Organization
Has code of conduct 724 84.0
Has no code of conduct 138 16.0
in the hypothesized model (Hayduk & Glaser, 2000). Second, confirmatory factor analysis (CFA) was employed to test the measurement model. If the fit of the CFA model is acceptable, the analysis can advance to testing the hypothesized structural regression (SR) model; otherwise, the measurement model must be revised. Following this, the hypothesized SR model was estimated with the same set of measurement models. The following section details each step of the process of testing the hypotheses and criteria to estimate the model fit.
3.5.1 Exploratory Factor Analysis
In this study, EFA was conducted to test common method bias and construct validity by using SPSS version 22. Testing common method bias is necessary before testing the study hypotheses (Sharma, Yetton, & Crawford, 2009), particularly for quantitative studies using self-report and a cross-sectional design (Podsakoff et al., 2003). Common method bias is a potential validity and generalizability threat for the research results and can lead to misinterpretations.
This research applied Harman’s single factor test on the data (Podsakoff et al., 2003) and examined the number of factors necessary to account for the variance in the variables. The basic assumption of this technique is that if a substantial amount of common method variance is present, a single factor will emerge from the factor analysis, or one general factor will account for the majority of covariance among the measures.
Apart from using EFA for statistical remedies to examine common method bias, EFA was also used to examine the construct validity for the study data. If six factors of the hypotheses SR model—ethical leadership, informational justice, interpersonal justice, affective organizational commitment, affective commitment to the supervisor, and job satisfaction—were extracted, the accuracy of the number of factors in the hypothesized model was specified (Hayduk & Glaser, 2000).
EFA was processed by loading all 38 items for the six variables in the study, using maximum likelihood method for factor extraction. Principal axis factoring was selected with an oblique rotation (direct oblimin), allowing for correlations among the factors, in order to explore number of factors. The Kaiser-Meyer-Olkin (KMO) value
needs to be greater than 0.6 (Kaiser, 1974), while the Bartlett’s test of sphericity needs to be statistically significant with a p-value of less than 0.05 (Bartlett, 1954).
3.5.2 Confirmatory Factor Analysis
CFA was conducted to test how well the actual data conformed to the measurement model. CFA indicates the relationship among the observed variables (the items of the questionnaire) underlying the latent variables. A correlation matrix of the items for each measurement was loaded through the program syntax for analysis using model estimation through maximum likelihood. A correlation matrix of all 38 items from six measurements was then loaded to confirm the factors in the model by using the same model estimation. Some items were identical across measurements; thus, the error terms for these items were likely to be correlated, so the estimated error terms for these items were allowed to freely co-vary, which suggested non-zero correlations between each individual error of all the factors (Kline, 2011). The details of the goodness-of-fit indices were as described in Section 3.5.4 and indicated in Table 3.15.
3.5.3 Hypothesized SR Motel Testing
This analysis began by drawing a hypothesized model using the CFA measurement model, as depicted in Figure 3.2. In this figure, ethical leadership, informational justice, interpersonal justice, affective organizational commitment, affective commitment to the supervisor, and job satisfaction are presented in an ellipse- shaped object that represents the latent variables. The items (indicators) of latent variables are represented in rectangles. The relationships between the latent variables and indicators are represented by a one-way arrow (). A line with a one-way arrow between two latent variables indicates the influence of one variable on the other: ethical leadership informational justice; ethical leadership interpersonal justice;
informational justice affective organizational commitment; informational justice affective commitment to the supervisor; informational justice job satisfaction;
interpersonal justice affective organizational commitment; interpersonal justice affective commitment to the supervisor; and interpersonal justice job satisfaction.
The effect size of the paths was determined by standardized path coefficient, which
measured the effect of one variable on the other variables. The significance of the standardized path coefficient was determined by a t-value when it was greater than 1.96 (Kline, 2011).
Figure 3.2 Hypothesized Model
3.5.4 Goodness-of-Fit Indices
To estimate a model fit, this study was selected well-established and widely used indices: a χ2 test of fit statistic and approximate fit indices. According to Kline (2011), four approximate fit indices have been widely reported in the SEM literature:
the Steiger–Lind root mean square error of approximation (RMSEA), Jöreskog–
Sörbom goodness-of-fit index (GFI) and absolute goodness-of-fit index (AGFI), Bentler comparative fit index (CFI), and standardized root mean square residual (SRMR).
To demonstrate model fit, the χ2 value should not be significant (p > 0.05);
RMSEA should be less than 0.05 (indicating a “good fit”); and GFI and AGFI should generally be 0 to 1.0, where 1.0 indicates the best fit. The combination of a CFI and SRMR threshold for concluding “acceptable fit” is based on CFI being ≥ 0.95 and SRMR ≤ 0.08. A factor loading value of any item of less than 0.4 is considered
unacceptable (Deng, 2010). Table 3.15 provides information on the fit indices and fit criteria used in the study.
Table 3.15 Goodness-of-Fit Indices
Fit Indices Fit Criteria References
χ2 p > 0.05 Kline (2011)
RMSEA < 0.50 good Browne and Cudeck (1992) 0.05 to 0.08 reasonable
0.08 to 0.1 tolerable
GFI ≥ 0.90 Joreskog and Sorbom (1982)
AGFI ≥ 0.90 Tanaka and Huba (1985)
CFI ≥ 0.95 Bentler (1990)
SRMR ≤ 0.08 Hu and Bentler (1999)
Loading value > 0.40 Deng (2010)