Chapter 3: Research Methodology
3.4 Analysis
3.4.4 Analysis of Mediation and Moderation
3.4.1.2 Procedures of Moderation
Moderation is to moderate the effect of predict and outcome. Moderation occurs when the effect of independent variable on dependent variable change because of moderator (Baron & Kenny, 1986; James & Brett, 1984).
Moderation is also known as interaction. It is used when the goal is to uncover an association between two variables. The association between dependent variable and independent variables would be moderated when the size or sign depends on third variable or set of mediation variable. Moderation is used to depict moderator variables
that could influence the magnitude of the effect of dependent and independent variables (Hayes, 2013).
Moderation is conducted with the purpose of testing an interaction between the moderator and independent variables in a model of dependent variable (Hayes, 2013). It is involved in differences on individual, situation or conditions that impact on relationship of predictor and outcome (Cohen & Edwards, 1989; Taylor &
Aspinwall, 1996). But, moderation can only explain the strength or direction of the relationship between dependent variable and independent variables. The following figures have been used to explain the moderation effect (Edwards & Lambert, 2007;
Hayes, 2013).
Figure 5: First Stage Moderation Model
Edwards and Lambert (2007) explains that in accordance with figure 5, nationality acts as a moderator of the path from work design to person-job fit, which is known as the first stage of the mediated effect of work design and work outcomes.
Figure 6: Second Stage Moderation Model
According to figure 6, nationality also acts as a moderator of the path from person-job fit to work outcomes, which is known as the second stage of the mediated effect of work design and work outcomes in Edwards and Lambert (2007) theory.
Figure 7: Direct Effect Moderation Model
However, figure 7 demonstrated that nationality is a moderator of the path from work design to work outcomes, which known as the direct effect of work design on work outcomes in Edwards and Lambert (2007) theory.
Figure 8: Total Effect Moderation Model or The Moderation of Nationality on Work Design, Person-job Fit and Work Outcomes
In figure 8, work design is an independent variable. Work outcomes are the dependent variable. Person-job fit is mediator, while nationality is moderator.
Edwards and Lambert (2007) called this figure as the total effect moderation model because it combined moderation of the direct effect with moderation of the first and second stages of the mediation effect, thereby moderating each path of the total effect of work design on work outcomes.
Hayes (2013) also explained that figure 8 includes both direct and indirect effects of independent variable moderated. As depicted, nationality moderates the indirect effect through its moderation of the effect of work design and person-job fit.
Nationality moderates the indirect effect through its moderation of the effect of person-job fit and work outcomes. At the same time, nationality moderates the direct effect of work design on work outcomes. By comparing between Thai and Taiwanese data, this study coded 0 for Thai data and 1 for Taiwanese data. The total amount of Thai data contains 204 samples, and the total amount of Taiwanese data contains 1509 samples.
In accordance with above explanation, the researcher expects that in this study nationality has moderate effect on work design, person-job fit and work outcomes.
Chapter 4 Results
This chapter mentions on the measurement, which is very good to predict the outcomes. Specially, this chapter also displays the results of this study.
4.1 Reliability and Validity Testing
Each item was carried out to analyze. The following section described about the results in details, which were chi-square, degree of freedom, CFI, RMSEA, SRMR and the Cronbach’s alpha.
4.1.1 Work Design
The resulted of CFA indicated that the 21 factor models of WDQ fit the data well (χ2 = 5760.704, df = 2639, p < 0.001, CFI = 0.928, RMSEA = 0.067, SRMR = 0.080). The result provided evidence of good structure of WDQ scale. The internal Cronbach’s alpha of work design was 0.947.
!45 Table 3: Reliabilities and Correlations among Variables (n=204)
Note: MPS is Motivating potential score; WSA is Work scheduling autonomy; DMA is Decision-making autonomy; WMA is Work methods autonomy; TV is Task variety; TS is Task significance; TI is Task identity; FFJ is Feedback from job; JC is Job complexity; IP is Information processing; PS is Problem solving; SV is Skill variety; Spec isSpecialization; SS is Social support; II is Initiated independence; RI is Received interdependence; IOO is Interaction outside organization; FFO is Feedback from others; Ergis Ergonomics; PD is Physical demands; WC is Work conditions; EU is Equipment use; Cronbach’s alpha appear are parentheses;* p < 0.05; ** p < 0.01
4.1.2 Job Satisfaction and Job Performance
The result of job satisfaction and job performance indicated that the two factor models fit the data well (χ2 = 23.3, df = 8, χ2/df =2.912, p < 0.001, CFI = 0.981, RMSEA = 0.098, SRMR = 0.485. The result provided evidence of good structure of job satisfaction and job performance. The Cronbach’s alpha of job satisfaction and job performance was 0.868 and 0.918, respectively.
4.2 Description of Variables and Correlation Analysis
4.2.1 Description of Variables and Correlation Analysis of Thai Data 4.2.1.1 Descriptive Statistics of the Variables of Thai Data
The questionnaire item used 5-point Likert scale (1 for totally disagree and 5 for totally agree), which the average score of each constructs also range from 1 to 5.
Table 4 showed that mean of the motivating potential score was 58.666 and the standard deviation was 24.755. The mean and standard deviation of P-J fit were 3.838 and 0.740. Work outcomes consisted of job satisfaction and job performance. The mean and standard deviation of job satisfaction were 3.652 and 0.763. The mean and standard deviation of job performance were 3.868 and 0.618. The Cronbach’s alpha of MPS, P-J fit, job satisfaction and job performance were 0.952, 0.900, 0.868 and 0.918 respectively.
Table 4: Means, Standard Deviations, Reliabilities and Correlations among Variables (MPS as the representative of Work Design) (n=204)
Mean SD 1 2 3 4 5
1 Gender 0.799 0.402 -
2 MPS 58.666 24.755 -0.071 (0.952)
3 P-J Fit 3.838 0.740 0.006 0.472** (0.900)
4 JS 3.652 0.763 0.028 0.342** 0.656** (0.868)
5 JP 3.868 0.618 0.071 0.363** 0.552** 0.543** (0.918) Note: SD is Standard deviation; MPS is Motivating potential score; JS is job satisfaction; JP is job performance; Cronbach’s alpha appear are parentheses; ** p < 0.01
4.2.1.2 Correlation between the Constructs of Thai Data
Table 4 showed that motivating potential score and P-J fit were positively correlated (r = 0.472, p < 0.01), demonstrated that whenever there is higher motivating potential score, there is a better job satisfaction. Thus, the result supported that motivating potential score is positively related with P-J fit.
The table also showed that motivating potential score and job satisfaction were positively correlated (r = 0.342, p < 0.01). While motivating potential score and job performance were also positively correlated (r = 0.363, p < 0.01). This marked that whenever there is a higher motivating potential score, there is a better job satisfaction and job performance. Thus, this result supported that the motivating potential score and job satisfaction together with motivating potential score and job performance are positively related.
A significant correlation was found for the relationship between P-J fit and job satisfaction (r = 0.656, p < 0.01). This revealed that when P-J fit is high, job satisfaction also high. P-J fit and job performance were also correlated (r = 0.552, p < 0.01). This also demonstrated that there is a high P-J fit and high job performance.
Both results supported the hypothesis that P-J fit is positively related to work outcome, which are job satisfaction and job performance.
In agreement with table 4, the correlation coefficient between gender and MPS was -0.071(n.s.). While the correlation coefficient between gender and P-J fit was 0.006 (n.s.). And the correlation coefficient between gender and job satisfaction was 0.028 (n.s.). Finally, the correlation coefficient between gender and job performance was 0.071 (n.s.). From the number above, it indicated that there is no gender difference effect on MPS, P-J fit, job satisfaction and job performance.
4.2.2 Description of Variables and Correlation Analysis of Taiwanese Data
4.2.2.1 Descriptive Statistics of the Variables of Taiwanese Data
The questionnaire item used 5-point scale (1 for totally disagree and 5 for totally agree), which the average score of each constructs also range from 1 to 5.Table 5 showed that mean and standard deviation of the motivating potential score were 44.844 and 22.706. The mean and standard deviation of P-J fit were 3.399 and 0.854. The mean and standard deviation of job satisfaction were 3.437 and 0.852. And the mean and standard deviation of job performance were 3.737 and 0.605. The Cronbach’s alpha of MPS, P-J fit, job satisfaction and job performance were 0.819, 0.818, 0.897 and 0.850 respectively.
Table 5: Means, Standard Deviations, Reliabilities and Correlations among Variables of Taiwanese Data (n=1509)
Cronbach’s alpha appear are parentheses; ** p < 0.01; * p < 0.05
4.2.2.2 Correlation between the Constructs of Taiwanese Data
Taiwanese data showed that motivating potential score and P-J fit were positively correlated (r = 0.371, p < 0.01), demonstrated that whenever there is higher motivating potential score, there is a better job satisfaction. Thus, the result supported that motivating potential score is positively related with P-J fit.
The table 5 also showed that motivating potential score and job satisfaction were positively correlated (r = 0.449, p < 0.01). While motivating potential score and job performance were also positively correlated (r = 0.353, p < 0.01). This marked that whenever there is a higher motivating potential score, there is a better job satisfaction and job performance. Thus, this result supported that the motivating potential score and job satisfaction together with motivating potential score and job performance are positively related.
A significant correlation was found for the relationship between P-J fit and job satisfaction (r = 0.498, p < 0.01). This revealed that when P-J fit is high, job satisfaction also high. P-J fit and job performance were also correlated (r = 0.316, p < 0.01). This also demonstrated that there is a high P-J fit and high job performance.
Both results supported the hypothesis that P-J fit is positively related to work outcome, which are job satisfaction and job performance.
In accordance with table 5, the correlation coefficient between gender and MPS was 0.089 (p < 0.01). While the correlation coefficient between gender and P-J fit was 0.064 (p < 0.05). And the correlation coefficient between gender and job satisfaction was 0.129 (p < 0.01). Last but not least, the correlation coefficient between gender and job performance was -0.068 (n.s.). From the number above, it indicated that gender also has an impact on MPS, P-J fit, job satisfaction, but does not have an impact on job performance.
4.3 Hypothesis Testing
In this section, the hypotheses were tested with taking the gender, age and managerial position as a control variables. First of all, the predictability of work design on P-J fit was tested. Furthermore, the mediation and moderation effect on P-J fit and work outcomes. The results are showed on the table in this section.
4.3.1 Using Work Design and Person-Job Fit to Predict Work Outcomes
4.3.1.1 Work Design and Person-Job Fit
Work design includes four important dimensions, which also divided into more sub dimensions. However, in work design, the most important thing that can
explain overall of work design questionnaire is task characteristics, which can be calculated as “Motivating potential score or MPS”.
In table 4, the regression coefficient for MPS to predict P-J fit without control variable was 0.472 (p < 0.01). This showed that MPS has strong prediction towards P-J Fit. But in table 6, the three variables, which are gender, age and manager, were controlled. While motivating potential score was used as independent variable, P-J fit was used as dependent variable. In step 1, the effect of control variables were very small and not significant, which means that these control variables are not so important. However, the researcher kept these control variables in order to control the confounding because they may have some differences that could impact on the model.
Under three control variables, the regression coefficient for motivating potential score to predict P-J fit was 0.450 (p < 0.01).
Table 6: Parameter Estimation and Model Summary of MPS (n=204) P-J Fit (Dependent Variable)
Note: MPS is Motivating potential score; ** p < 0.01
Moreover, work design does not only have MPS, but also include job complexity, information processing, problem solving, skill variety, specialization, social support, initiated independence, received interdependence, interaction outside organization, feedback from others, ergonomics, physical demands, work conditions and equipment use. The correlation coefficient between job complexity and P-J fit was -0.104 (n.s.). The correlation coefficient between information processing and P-J fit was 0.373 (p < 0.01). The correlation coefficient between problem solving and P-J fit was 0.402 (p <0.01). The correlation coefficient between skill variety and P-J fit was 0.525 (p < 0.01). The correlation coefficient between specification and P-J fit was 0.479 (p < 0.01). The correlation coefficient between social support and P-J fit was 0.514 (p <0.01). The correlation coefficient between initiated interdependence and P-J fit was 0.197 (p < 0.01). The correlation coefficient between received interdependence and P-J fit was 0.130 (n.s.). The correlation coefficient between
interaction outside organization and P-J fit was 0.230 (p < 0.01). The correlation coefficient between feedback from others and P-J fit was 0.504 (p < 0.01). The correlation coefficient between ergonomics and P-J fit was 0.290 (p < 0.01). The correlation coefficient between physical demands and P-J fit was 0.334 (p <0.01). The correlation coefficient between work conditions and P-J fit was 0.280 (p < 0.01). The correlation coefficient between equipment use and P-J fit was 0.259 (p < 0.01).
From the number above, the result of work design showed that most of the sub dimensions of work design has strong prediction towards P-J fit.
4.3.1.2 Person-Job Fit and Work Outcomes
The table 7 showed that the result of P-J fit and work outcomes, which are job satisfaction and job performance. P-J fit was used as independent variable, while job satisfaction and job performance were used as dependent variable. In step 1, the effect of control variables was so small and insignificant, which indicated that these control variables are not so important. However, the researcher kept these control variables in order to avoid confounding effect. With three control variables, the regression coefficient for P-J fit to predict job satisfaction was 0.656 (p < 0.01). However, the regression coefficient for P-J fit to predict job performance was 0.546 (p < 0.01). In summary, P-J fit is stronger predict job satisfaction than job performance.
Table 7: Parameter Estimation and Model Summary of P-J Fit (n=204) Job Satisfaction
(Dependent Variable) .
Job Performance (Dependent Variable) .
Step 1 Step 2 Step 1 Step 2
Control Variable
Gender 0.022* 0.012** 0.079 0.071**
Age 0.159* 0.069** 0.077 0.003**
Manager 0.010* -0.074** 0.099 0.029**
Independent Variable
P-J Fit 0.656** 0.546**
R2 0.027* 0.438** 0.025 0.310**
F 1.851* 38.811** 1.719 22.323**
p 0.139* < 0.001** 0.164 < 0.001**
R2 0.411** 0.285**
F Change 145.673** 82.042**
p < 0.001** < 0.001**
Note: * p < 0.05; ** p < 0.01
4.3.2 Mediation of Person-Job Fit on Work Design and Work Outcomes
4.3.2.1 Mediation between MPS and Work Outcomes
According to table 4, without control variables, the correlation coefficient of motivating potential score on P-J fit, job satisfaction and job performance were 0.472 (p < 0.01), 0.342 (p < 0.01), 0.363 (p < 0.01) respectively. The result showed that motivating potential score has stronger prediction towards P-J fit, as compared to the others two work outcomes.
However, in table 8, the researcher had controlled on three variables, which are gender, age and managerial position. In step 1, the effect of control variables of both job satisfaction and job performance were too small and non-significant. This meant that the control variable is not important. However, the control variables might have some slightly impact on the model and also used to control the confounding.
In step 2, under the control of three variables, MPS effect on job satisfaction and job performance was very significant. The regression coefficient of motivating potential score to predict job satisfaction and job performance was 0.329 (p < 0.01) and 0.359 (p < 0.01) respectively. This showed that motivating potential score effect on the job performance is stronger than job satisfaction because the beta coefficient is higher.
In step 3, P-J fit was included into the model, job satisfaction had a full mediation because the direct effect of motivating potential score towards job satisfaction disappear, the beta coefficient decreased from 0.329 (p < 0.01) to 0.042 (n.s.), which meant that motivating potential score and job satisfaction are mediated by P-J fit. However, for job performance, it has both direct and indirect effect, which means that not only motivating potential score can effect on job performance, but motivating potential score and job performance can also be mediated by P-J fit. Therefore, researcher should be aware of job performance because it has both direct effect and indirect effect.
Table 8: Parameter Estimation and Model Summary of MPS and P-J Fit (n=204) Job Satisfaction
(Dependent Variable) .
Job Performance
(Dependent Variable) .
Step 1 Step2 Step3 Step 1 Step2 Step3
Control Variable
Gender 0.022 0.044** 0.015** 0.079 0.103** 0.081**
Age 0.159 0.096** 0.064** 0.077 0.008** -0.016**
Man. 0.010 -0.030** -0.077** 0.099 0.055** 0.020**
Independent Variable
MPS 0.329** 0.042** 0.359** 0.141**
P-J Fit 0.638** 0.484**
R2 0.027 0.128** 0.440** 0.025 0.145** 0.324**
F 1.851 7.280** 31.057** 1.719 8.420** 19.023**
p 0.139 < 0.001** < 0.001** 0.164 < 0.001** < 0.001**
R2 0.101** 0.312** 0.120** 0.180**
F change 22.955** 110.191** 27.828** 52.689**
p < 0.001** < 0.001** < 0.001** < 0.001**
Note: Man. is Managerial position; MPS is Motivating potential score;
* p < 0.05; ** p < 0.01
Figure 9: The Mediation of MPS and Job Satisfaction
Note: The number is the beta coefficient; (a) has only one predictor; (b) has two predictor; ** p < 0.01
Figure 10: The Mediation of MPS and Job Performance
Note: The number is the beta coefficient; (a) has only one predictor; (b) has two predictor; * p < 0.05; ** p < 0.01
4.3.2.2 Mediation between Work Design and Work Outcomes
In the previous tables, the researcher only mentioned about the motivating potential score. However, work design does not only have MPS, but also cover on job complexity, information processing, problem solving, skill variety, specialization, social support, initiated independence, received interdependence, interaction outside organization, feedback from others, ergonomics, physical demands, work conditions and equipment use. Some of them are similar to MPS, but some of them are different from MPS. Therefore, the researcher made a summarized of all sub dimensions of work design in table 9.
Table 9 showed that the relationship between the sub dimensions of work design together with job satisfaction are fully mediated by P-J fit. However, decision-making autonomy 0.124 (p < 0.05), work methods autonomy 0.171 (p < 0.01), information processing 0.163 (p < 0.01), specialization 0.148 (p < 0.05), social support 0.192 (p < 0.01), ergonomics 0.205 (p < 0.01) and work conditions 0.132 (p<0.05) are partially mediated because they have a direct effect towards job satisfaction. Therefore, researcher should be aware of these dimensions because it does not only effect on P-J fit but also effect on job satisfaction.
For job performance, the relationship between the sub dimensions of work design together with job performance are fully mediated by P-J Fit; except task variety 0.142 (p < 0.05), task identity 0.271 (p < 0.01), information processing 0.182 (p < 0.01), social support 0.386 (p < 0.01) and feedback from other 0.201 (p < 0.01).
As task variety, task identity, information processing, social support and feedback from other do not only have an indirect effect, but also have direct effect towards job performance. Therefore, researcher should pay more attention to these sub dimensions.
In conclusion, work design is very important to the outcome variables, but they always mediated by P-J fit.
Table 9: Parameter Estimation and Model Summary of Work Design (n=204)
Note: DV is Dependent variable; MPS is Motivating potential score; WSA is Work scheduling autonomy; DMA is Decision-making autonomy; WMA is Work methods autonomy; TV is Task variety; TS is Task significance; TI is Task identity; FFJ is Feedback from job; JC is Job complexity; IP is Information processing; PS is Problem solving; SV is Skill variety; Spec is Specialization; SS is Social support; II is Initiated independence; RI is Received interdependence; IOO is Interaction outside organization; FFO is Feedback from others; Erg is Ergonomics; PD is Physical demands; WC is Work conditions; EU is Equipment use; * p < 0.05; ** p < 0.01
4.3.3 Moderation between Work Design and Work Outcomes
In order to test the moderation effect of nationality on the path model, interaction terms had been created by, first, multiplying the work design with nationality and, second, multiplying P-J fit with nationality. Due to many variables of work design, so in the moderation part, the researcher used the MPS as a substitute of work design because the results of sub dimensions of work design variables are quite similar to each other.
4.3.3.1 Moderation of Nationality on MPS (Work Design) and Person-Job Fit
In step 1, gender, age and MPS were set. Table 10 showed that after including
nationality in step 2, there was an increase of square multiple correlations ( R2 = 0.009, F change = 15.872, p < 0.001). When nationality was included, MPS
has a more significant predictability. However, the interaction term of MPS and nationality was not significant as showed in step 3 ( R2 = 0.000, F change = 0.056, n.s). From figure 11, step 3 is equivalent to the first stage moderation model of Edwards and Lambert (2007). The result showed that the interaction of MPS and nationality does not have a significant predictability on the model, the b coefficient was -0.001 (n.s.). Summing it up, there is no difference between Thai and Taiwanese in this point.
Figure 11: First Stage Moderation Model Note: The number in figure 11 is the b coefficient
!57 Table 10: Parameter Estimation and Model Summary of Moderation of MPS and Nationality (Thai n=204, Taiwanese n=1509)P-J fit (Dependent Variable) Step 1 . Step 2 . Step 3 . b betat b betat b betat Constant2.66529.81**3.02423.89**2.99817.74**Control Variable Gender 0.0020.0010.05**0.0420.0251.00**0.0430.0251.01** Age0.0040.0401.64**0.0000.0031.00**0.0000.0020.08**Independent Variable MPS0.0140.39716.17**0.0140.38115.44**0.0140.3956.34** Nationality -0.262-0.107-3.98**-0.231-0.094-1.55**Interaction MPS*Nat. -0.001-0.018-0.24**R 20.165**0.174**0.174**F 94.700**75.727**60.553**p < 0.001**< 0.001**< 0.001**R 20.009**0.000**F change 15.872**0.056**p< 0.001**0.813**Note: MPS is Motivating potential score; Nat. = Nationality; Code 0 = Thai data; Code 1 = Taiwanese data; ** p < 0.01
4.3.3.2 Moderation of Nationality on Person-Job Fit and Work Outcomes
4.3.3.2.1 Moderation of Nationality on Person-Job Fit and Job Satisfaction
From table 11, gender, age, P-J fit were included into step 1. The result showed that P-J fit was significant predicted job satisfaction (t = 22.64, p < 0.01).
However, after included nationality into step 2, there was no change in square multiple correlation as shown on table 11 ( R2 = 0.000, F change = 0.298, n.s.). P-J fit’s b coefficient was 0.529 (p < 0.01), but nationality’s b coefficient was 0.035 (n.s.).
In step 3, the interaction term of P-J fit and nationality was included, which is similar
In step 3, the interaction term of P-J fit and nationality was included, which is similar