In this chapter, the researcher presents an outline of the research method used for the study, including research framework, hypotheses, research procedure, data collection, measurement, reliability and validity of instruments and data analysis. Before continuing to the data analysis process of the main study.
Research Framework
Based on literature review, the dimensions under each construct were defined in the previous chapter. The researcher developed the research framework for this study according to the relationships that have been discovered in other studies. Refer to Figure 3.1. for the framework of employee silence, employee well-being, work engagement and mentoring relationship, which is used to address the purpose of study. The framework shows the hypotheses being tested and the variables under study.
Figure 3. 1. Research framework.
23
Research Hypothesis
Hypothesis 1:Different types of employee silences affect employee well-being in different directions.
Hypothesis 1a:
Acquiescent Silence will have a negative effect on employee well-being in organization.
Hypothesis 1b:
Defensive Silence will have a negative effect on employee well-being in organization.
Hypothesis 1c:
Pro-social Silence will have a positive effect on employee well-being in organization.
Hypothesis 1d:
Opportunistic Silence will have a negative effect on employee well-being in organization.
Hypothesis 2:
Different types of employee silences affect work engagement in different directions.
Hypothesis 2a:
Acquiescent Silence will have a negative effect on work engagement in organization.
Hypothesis 2b:
Defensive Silence will have a negative effect on work engagement in organization.
Hypothesis 2c:
Pro-social Silence will have a positive effect on work engagement in organization.
Hypothesis 2d:
Opportunistic Silence will have a negative effect on work engagement in organization.
Hypothesis 3:
Employee well-being has a mediating effect on the relationship between different types of employee silences and work engagement.
Hypothesis 3a:
Employee well-being has a mediating effect on the relationship between Acquiescent Silence and work engagement.
Hypothesis 3b:
Employee well-being has a mediating effect on the relationship between Defensive Silence and work engagement.
24
Hypothesis 3c:
Employee well-being s has a mediating effect on the relationship between Pro-social Silence and work engagement.
Hypothesis 3d:
Employee well-being has a mediating effect on the relationship between Opportunistic Silence and work engagement.
Hypothesis 4:
The mentoring relationship will weaken the relationship between different types of employee silences and work engagement.
Hypothesis 4a:
The mentoring relationship will weaken the relationship between Acquiescent Silence and work engagement.
Hypothesis 4b:
The mentoring relationship will weaken the relationship between Defensive Silence and work engagement.
Hypothesis 4c:
The mentoring relationship will weaken the relationship between Pro-social Silence and work engagement.
Hypothesis 4d:
The mentoring relationship will weaken the relationship between Opportunistic Silence and work engagement.
25
Research Procedure
Based on literature review, the dimensions under each construct were defined in the previous chapter. To explain the procedure, the steps can be categorized into different stages. In research procedure, the steps are: explore background of study, select interested topic, review of literature, identify research questions, develop framework, research method and instrument, then conduct expert review and pilot test. And then, moving to data analysis process form collect data, analyze data, interpret data, and inform others.
Figure 3. 2. Research procedure.
Explore Background of Study
26
Research Design
This study will use a quantitative approach to test the hypothesized relationships by analyzing the collected data with IBM SPSS 23.0 and SPSS Amos 24 software. By a quantitative approach, the researcher can get sufficient amount of data to statistically analyze the relationship between employee silence, employee well-being and work engagement which are moderated by mentoring relationship.
Sampling and Data Collection
For this study, the target population was composed of employees working in organizations of over 30 people and with mentoring relationship in the organization, in order to investigate whether the employee silence behavior will be affected by mentoring relationship in organization. Convenience sampling approach was utilized. The final sample comprised of 435 employees from different industries in Taiwan to understand silence situation in general workplace and put focus on investigating employees’ behavior in organization.
Data collection process mostly used online distribution through the months of March and April 2018. The collection process were carried out in the following ways: (1) a post and direct messaging via Facebook and Line; (2) requests to friends from specific fields/companies. In addition, all the above e-mails and posts include a request to forward the links and QR codes to friends of the addressees. As for the online questionnaires, this research received a total of 435 valid respondents in the end.
27
Instrumentation
This is a questionnaire-based or questionnaire-oriented research conducted in hope to figure out the factors that have effect on work engagement.The questionnaire for this study was designed with four parts including the demographic and background information. In avoidance of the bias due to respondents’ tendency when they fill out the questionnaire, the arrangement of the questionnaire was designed to respectively examine each construct in reversed causal order, such as work engagement items first, then employee well-being, employee silence and mentoring relationship. The instrument consists of measurement items of four variables, and 48 questions in total.
Data Analysis
This is a quantitative research, the data collection was obtained through questionnaires.
First of all, all collected data were entered into an SPSS data file. Secondly, the researcher used SPSS and AMOS SEM to run the analysis. During data analysis process, the following analyses were conducted, including descriptive statistics, confirmatory factor analysis in AMOS, Pearson correlation analysis, path analysis in AMOS and hierarchical regression analysis.
Measurement Employee Silence
The measurement of employee silence was adopted from a previous scale (Knoll &
Van Dick, 2013), by adjusting the particular word for each forms of employee silence. It comprises 12 statements to complete the item root “Sometimes I remain silent at work...”.
Item formulation was adapted by the exploratory research of Brinsfield (2009) and Milliken et al. (2003), and by Van Dyne et al.’s (2003) conceptual suggestions. All items use a 5-point Likert scale from 1 to 5, from strongly disagree to strongly agree. The questionnaire consists of three dimensions. Firstly, defensive silence reflects employees intentionally withholding the ideas, a sample item is “I remained silent at work because of fear of negative consequences.” Secondly, acquiescent silence describes employees’ fear of speaking up, such as “I remained silent at work because nothing will change.” Thirdly, refers to employees wanted to cooperate and protect others, a sample statement like “I remained silent at work because I do not want to embarrass others.” is the definition of pro-social silence. Lastly, opportunistic silence describes employees are more willing to protect their advantages, a sample item is “I remained silent at work because to not give away my
28
knowledge advantage”. According to previous research (Knoll & Van Dick, 2013) which already indicated the most significant items for measuring different types of employee silence. After tested by Amos, the researcher decided to adopt previous research that has already been well tested by EFA and CFA.
Employee Well-Being
The original data for the Work-Related Quality of Life Scale were gathered via a survey of UK (Van Laar, Edwards & Easton, 2007). The conceptual model of the quality of working life, which was measured through the Work-Related Quality of Life Scale incorporates a factor structure. In this research, the researcher only focused on two major dimensions, including General Well - Being (GWB), and Job and Career Satisfaction (JCS).
All items were scored on a 5-point Likert scale of 1 to 5, from strongly disagree to strongly agree.
Work Engagement
The research adopted the Utrecht Work Engagement Scale (UWES; Schaufeli &
Bakker, 2003) as a measurement of work engagement. The three subscales of the UWES—
vigor, dedication, and absorption were used in the study. All items were scored on a 7-point Likert scale from strongly disagree to strongly agree. The questionnaire consisted of three subscales. Firstly, an example item of vigor is “At my work, I feel bursting with energy.”;
Secondly, an example item of dedication is “I find the work that I do full of meaning and purpose.” Thirdly, an example item of absorption is “Time flies when I'm working.”.
Mentoring Relationship
Mentoring function measurement was adopted from a previous research (Scandura &
Ragins, 1993). This measurement included three dimensions. The first dimension is
“Psychosocial support”, reflects the friendship and psychological support of mentor (Kram
& Isabella, 1985), such as “I share personal problems with my mentor”. The second dimension of “Career development”, reflect protégés access the work-related support by mentor. The third dimension “Role modeling” reflects how the protégés behavior can be affected by mentors. In this research, the strength of mentoring relationship was assessed by aggregating all items in psychosocial support, career development and role modeling.
All items use a 7-point Likert scale from strongly disagree to strongly agree.
29
Control Variables
Found by the previous scholars that work engagement might be influenced by some demographic variables. Hence, the demographic variables will be incorporated into the questionnaire as well. In this study, position level and tenure are adopted as the control variables. According to previous research, work engagement was found to have close relationship with employee’s tenure (Glavas & Piderit, 2009 ) and position level (Kahn, 1990). During analysis process, these two main control variables were controlled in hierarchical regression. Tenure is coded: (1) 0~1 (2) 2~3 (3) 4~5 (4) 6~7 (5) 8~9 (6) 10 and above years. Position level is coded: (1) general staff, (2) line manager, (3) middle level manager, (4) high level manager. Control variables were tested during data analysis, and the research used SPSS (Statistics Package for Social Science) hierarchical regression technique to examine the effect of control variables on work engagement.
Validity and Reliability
Confirmatory factor analysis (CFA) was utilized to ensure the instruments were appropriately designed to attain the validity of measurement. The model fit and item loadings were confirmed by CFA to assess the significance of measurement for this study (Hinkin, 1998). Besides, a Harmon’s single factor test was also conducted to verify common method variance (CMV) because of the self-reported approach for data collection.
To ensure the internal consistency, an alpha coefficient test was performed. The results are presented in this section to describe the model assessment processes for the five variables:
Acquiescent Silence, Defensive Silence, Pro-social Silence, Opportunistic Silence, Employee Well-Being, Mentoring Relationship and Work Engagement.
30
Confirmatory Factor Analysis (CFA)
CFA was a measurement to determine how well the survey model generalize through groups and time (Brown, 2014). CFA was used to examine whether the existing measurement is appropriate for the current population. Hair, Black, Babin, & Anderson (2010) indicated that there were three categories of fit indices of CFA model such as absolute fit indices, incremental fit indices and parsimony fit indices. At least one absolute index and one incremental index as well as chi-square (x2) results should be reported to evaluate the model fit. The incremental fit index which also known as Bollen’s (1992) IFI.
For IFI, the values exceeding .90 are considered as acceptable, and also can exceed 1 as well. The following are preferred criteria for goodness of fit of the model: Goodness of Fit Index (GFI), Normed Fit Index (NFI), Comparative Fit Index (CFI), Root Mean Square error of approximation (RMSEA), Non-Normed Fit Index (TLI), and Incremental Fit Index (IFI). Refer to the summary of goodness of fit criteria in Table 3.1.
Furthermore, if a research measurement failed the good fit criteria, the researcher would modify the measurement model and apply multi-group comparison between two randomly split samples to cross validate the modified model in the following steps. First, this research would use SPSS to randomly split (50 %) the full sample into two groups (Sample A= 228/435; Sample B= 207/435). Second, using multi-group comparison analysis in AMOS SEM to make sure the two samples show no significant difference when fitting the data to the same model. The above steps are applied for each modified measurement.
Table 3. 1.
Summary of Goodness of Fit Criteria
Fit Indices Criteria References
x2/df 2.00~5.00 Tabachnick and Fidell, 2007
NFI >0.90 Byrne, 1994
GFI >.080 Byrne, 1994
CFI >0.90 Byrne, 1994
RMSEA 0.08 ~ 0.10 MacCallum, Browne, & Sugawara, 1996 NNFI(TLI) >0.90 Bagozzi and Yi, 1988
IFI >0.90 Hu and Bentler, 1999
31
CFA result of Employee Silence.
For the independent variable, 12 items were run through the CFA in AMOS in this procedure. Refer to Figure 3.3. that this model provides an acceptable model-to-data fit [X2 (N = 435) = 228.695; df = 48; CFI = .931; NFI = .915; IFI = .932;TLI = .905 RMSEA
= .093]. Refer to Table 3.2. for Confirmatory Factor Analysis of different types of employee silence which indicated all items were significant and AVE, CR were all above acceptable value (Fornell & Larcker, 1981).
Table 3. 2.
Confirmatory Factor Analysis of Employee Silence (N = 435)
Variable Dimensions NO Factor
loading
AVE CR
Employee Silence
ESAS 10 .68
Acquiescent Silence ESAS 11 .84 0.59 0.81
ESAS 13 .77
ESDS 4 .76
Defensive Silence ESDS 5 .69 0.56 0.79
ESDS 19 .80
ESPS 6 .83
Pro-social Silence ESPS 7 .80 0.61 0.82
ESPS 16 .69
ESOS 15 .65
Opportunistic Silence ESOS 17 .73 0.52 0.76
ESOS 20 .78
32
Figure 3. 3. Measurement model of different types of employee silences
33
CFA result of Employee Well-Being.
For the mediator, 11 items were run through the CFA in AMOS in this procedure.
Refer to Table 3.3. which shows some fit indices of the initial model (11 items) were not acceptable [X2 (N = 435) = 285.989; df = 43; CFI = .908; NFI = .894; IFI = .909;TLI=.883 RMSEA = .111]. In order to improve the fit, two items were deleted from this measurement.
One item (EWWB 1) “I feel well at the moment.” was deleted because of heavy cross loading in two dimensions (Hair, Black, Babin, Anderson, & Tatham, 2006, pp.149-151) which means the respondents had difficulty in recognizing this item for the designated construct in this measurement. The other item (EWJS 6) “I have a clear set of goals and aims to enable me to do my job.” Comparing with other items, it has the lowest mean in this dimension. Since the sample are Taiwanese who are known to lack confidence in defining clear goals in their work, this item may not represent well this sub-dimension, and therefore was deleted. After model respecification, the modified model (9 items) showed good fit measurement as [X2 (N = 435) = 129.189; df = 26; CFI = .947; NFI = .934; IFI
= .947;TLI = .926 RMSEA = .093].
Refer to Figure 3.4 and Table 3.4 for CFA of Employee Well-Being which indicated all factor loadings were significant, while AVE and CR were all above acceptable value (Fornell & Larcker, 1981). Also, Table 3.5. showed no significant difference in the multi-group comparison which indicated the modified model is valid for the two randomly split samples.
Table 3. 3.
Summary of Goodness of Fit for Employee Well-Being Measurement Model of Scale.
Model X2 df x2/df NFI CFI RMSEA TLI IFI
Initial CFA
(11 items) 285.989 43 6.651 .894 .908 .111 .883 .909 Modified CFA
(9 items) 129.189 26 4.469 .934 .947 .093 .926 .947
34
Table 3. 4.
Confirmatory Factor Analysis of Employee Well-Being (N = 435)
Variables Dimension NO Factor
loading
Result of Multi-group Comparison of Employee Well-Being
Model DF CMIN P NFI
Figure 3. 4. Measurement model of employee well-being
35
CFA result of Mentoring Relationship.
For the moderator, 15 items were run through the CFA in AMOS in this procedure.
Refer to Table 3.6. which shows the initial model (15 items) had less than satisfactory fit as [X2 (N = 435) = 554.473; df = 87; CFI = .918; NFI = .904; IFI = .918;TLI = .901 RMSEA
= .111]. In this measurement, one item was deleted, the item (ME 14) “Mentor has devoted special time and consideration to my career.” The item cross loaded (Hair et al., 2006, pp.149-151) between career support and psycho-social support dimensions which means to the respondents the meaning of this item highly overlaps between two constructs and therefore was deleted. After remodel re-specification, the modified model (14 items) showed improved model fi as [X2 (N = 435) = 365.160; df = 74; CFI = .943; NFI = .930;
IFI = .944;TLI=.930 RMSEA = .095].
Refer to Figure 3.5 and Table 3.7 for CFA of Mentoring Relationship which indicated all items’ factor loading were significant, while AVE and CR were all above acceptable value (Fornell & Larcker, 1981). Also, Table 3.8. showed no significant difference in the multi-group comparison which indicated the modified model is valid for the two randomly split samples.
Table 3. 6.
Summary of Goodness of Fit for Mentoring Relationship Measurement Model of Scale.
Model X2 df x2/df NFI CFI RMSEA TLI IFI
Initial CFA
(15 items) 554.473 87 6.373 .904 .918 .111 .901 .918 Modified CFA
(14 items) 365.160 74 4.935 .930 .943 .095 .930 .944
36
Table 3. 7.
Confirmatory Factor Analysis of Mentoring Relationship (N = 435)
Table 3. 8.
Result of Multi-group Comparison of Mentoring Relationship
Model DF CMIN P NFI
Variables Dimension NO Factor
loading
37
Figure 3. 5. Measurement model of mentoring relationship
38
CFA result of Work Engagement.
For the dependent variable, 17 items were run through the CFA in AMOS in this procedure. Refer to Table 3.9. which shows the initial model (17 items) had less than satisfactory fit as [X 2(N = 435) = 777.183; df = 116; CFI = .889; NFI = .872; IFI = .882;
TLI=.870 RMSEA = .115]. In this measurement, four items were deleted. Firstly, item (WGAB14) “I get carried away when I’m working.” and item (WGAB16) “It is difficult to detach myself from my job.” These two items presented similar concept about employees who are very engaged at work in same dimension, thus had higher covariances between items, and were deleted. Secondly, item (WGDE10) “I am proud on the work that I do.”
which means to the respondents the meaning of this item highly overlaps with other two constructs and therefore was deleted. Lastly, the item (WGVI17) “At my work I always persevere, even when things do not go well.” had low correlation with subscales (vigor), which indicated poor consistency and was deleted. After remodel re-specification, the modified model (13 items) showed improved fit measurement as [X2 (N = 435) = 295.767;
df = 62; CFI = .944; NFI = .930; IFI = .944; TLI = .930 RMSEA = .095].
Refer to Figure 3.6 and Table 3.10 for CFA of Work Engagement which indicated all items’ factor loading were significant, while AVE and CR were all above acceptable value (Fornell & Larcker, 1981). Also, Table 3.11. showed no significant difference in the multi-group comparison which indicated the modified model is valid for the two randomly split samples.
Table 3. 9.
Summary of Goodness of Fit for Work Engagement Measurement Model of Scale
Model X2 df x2/df NFI CFI RMSEA TLI IFI
Initial CFA
(17 items) 777.183 116 6.700 .872 .889 .115 .870 .882 Modified CFA
(13 items) 295.767 62 4.770 .930 .944 .093 .929 .944
39
Table 3. 10.
Confirmatory Factor Analysis of Work Engagement (N = 435)
Variables Dimension NO Factor
loading
Result of Multi-group Comparison of Mentoring Relationship
Model DF CMIN P NFI
40
Figure 3. 6. Measurement model of work engagement
41
Harmon’s Single Factor Test
When the measurements for testing all variables are self-reported at the individual level, common method biases may occur. Harmon’s single factor Test was conducted to examine common method variances (CMV) in this study. The variances extracted for the largest un-rotated component was 35.50%, which falls below the 50 % suggested threshold (Podsakoff, MacKenzie, Podsakoff, & Lee, 2003).
Alpha Coefficient Test
Cronbach’s Alpha is adopted for reliability calculation of this study. To reach strong item covariance and capture the sampling domain, the coefficient alpha 0.70 is referenced for a satisfactory standard. Nunnally (1978) suggested that all variables should reach a Cronbach’s alpha of 0.70. All variables are within a good range, the reliability value of each variable reached the score around 0.75 to 0.95. Table 3.12 summarized the alpha values for each variable.
Table 3. 12.
Cronbach's Alpha of Measurements (N = 435)
Variables Cronbach’s Alpha
Acquiescent Silence 0.80
Defensive Silence 0.79
Pro-social Silence 0.81
Opportunistic Silence 0.75
Employee Well-Being 0.88
Mentoring Relationship 0.95
Work Engagement 0.93
42