See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/358280876

Workshop on Structural Equation Modeling Using AMOS

Presentation · February 2022

DOI: 10.13140/RG.2.2.25538.66245

CITATIONS READS

0 977

1 author:

Dalowar Hossan
Universiti Putra Malaysia
29 PUBLICATIONS 32 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Extrinsic Motivational Factors and Workers Engagement at the Readymade Garments Industry in Bangladesh with Transactional Leadership as Mediator
View project

All content following this page was uploaded by Dalowar Hossan on 02 February 2022.

The user has requested enhancement of the downloaded file.

Workshop on Structural Equation Modeling Using AMOS

By
Dalowar Hossan; BBA, MSc
PhD student, University Putra Malaysia

https://www.facebook.com/dhossan, dalowarhossan.bd@gmail.com
 Research Title
The Effect of Leadership Style on Work Engagement: Mediating Role of Employee
Motivation and Moderating Effect of Gender
 RESEARCH FRAMEWORK

Gender

Transformational
Employee
Leadership Style
Motivation

Transactional
Leadership Style

Laissez-faire Work
Leadership Style Engagement
 RESEARCH QUESTION/OBJECTIVE/HYPOTHESIS

RESEARCH QUESTION RESEARCH OBJECTIVE RESEARCH HYPOTHESIS

1. What is the effect of 1. To investigate the effect H1: There is a significant effect of transformational
leadership style on work of leadership style on work leadership style on work engagement.
engagement? engagement. H2: There is a significant effect of transactional
leadership style on work engagement.
H3: There is a significant effect of laissez-faire
leadership style on work engagement.
 RESEARCH QUESTION/OBJECTIVE/HYPOTHESIS

RESEARCH QUESTION RESEARCH OBJECTIVE RESEARCH HYPOTHESIS

2. What is the effect of 2. To examine the effect H4: There is a significant effect of transformational
leadership style on of leadership style on leadership style on employee motivation.
employee motivation? employee motivation? H5: There is a significant effect of transactional leadership
style on employee motivation.
H6: There is a significant effect of laissez-faire leadership
style on employee motivation.
 RESEARCH QUESTION/OBJECTIVE/HYPOTHESIS

RESEARCH QUESTION RESEARCH OBJECTIVE RESEARCH HYPOTHESIS

3. What is the effect of 3. To assess the effect of H7: There is a significant effect of employee
employee motivation on employee motivation on motivation on work engagement.
work engagement? work engagement? dalowarhossan.bd@gmail.com
 RESEARCH QUESTION/OBJECTIVE/HYPOTHESIS

RESEARCH QUESTION RESEARCH OBJECTIVE RESEARCH HYPOTHESIS

4. What is the 4. To explore the H8: There is mediating effect of employee motivation in the
mediating effect of mediating effect of relationship between transformational leadership style and
employee motivation employee motivation work engagement.
in the relationship in the relationship H9: There is mediating effect of employee motivation in the
between leadership between leadership relationship between transactional leadership style and work
style and work style and work engagement.
engagement? engagement. H10: There is mediating effect of employee motivation in the
relationship between laissez-faire leadership style and work
engagement.
RESEARCH QUESTION/OBJECTIVE/HYPOTHESIS

RESEARCH QUESTION RESEARCH OBJECTIVE RESEARCH HYPOTHESIS

5. What is the 5. To identify the H11: There is moderating effect of gender on the
moderating effect of moderating effect of gender relationship between transformational leadership
gender on relationship on relationship between style and work engagement.
between leadership leadership style and work H12: There is moderating effect of gender on the
style and work engagement. relationship between transactional leadership style
engagement? and work engagement.
H13: There is moderating effect of gender on the
relationship between laissez-faire leadership style
and work engagement.
PREPARATION FOR DATA ANALYSIS
ASSUMPTIONS
Test of Univariate Data Normality
Test of Univariate Data Normality
Test of Univariate Outliers

Test of Multivariate Outliers

Mahalanobis Distance (D Square) is available in AMOS

Introduction to Structural Equation Modeling
Introduction to Structural Equation Modeling
Introduction to Structural Equation Modeling
Item parceling Operation
dalowarhossan.bd@gmail.com
AMOS: 3 Levels

dalowarhossan.bd@gmail.com
AMOS: 3 Levels (Continued…)
AMOS: 3 Levels (Continued…)
AMOS: 3 Levels (Continued…)
AMOS: Requirement
dalowarhossan.bd@gmail.com
 Step-1

Confirmatory Factor Analysis (CFA)

 CFA for Transactional Leadership Style (TAL)

Path Diagram

 Open CFA Template and save as CFA

TAL dalowarhossan.bd@gmail.com
 Open list variables in data set
 Drag and drop items TAL1 to TAL6
into indicator objects
 Label the latent object as TAL
 Click plugins and select name
unobserved variables to label
measurement errors (e1 to e6)
 Calculate estimates
 CFA for Transactional Leadership Style (TAL)

Analysis Results
Observed for:
 Fit indices
 Standardized Factor Loadings

Note:
Make sure these two criteria are fulfilled.

dalowarhossan.bd@gmail.com
 Type of Goodness fit indices

dalowarhossan.bd@gmail.com
 CFA for Transactional Leadership Style (TAL)

Test for Model Fit

 Requirements for RMSEA not

fulfilled as the value is more
than 0.08
 One factor loading (TAL4) is
negative and less than 0.50

Action:
 Delete indicator TAL4 (λ= -.04)
dalowarhossan.bd@gmail.com
 CFA for Transactional Leadership Style (TAL)

Results after deleting indicator TAL4:

 Requirements for fit indices are

fulfilled
 All factor loadings are more than
0.50

 The construct meets the model

fit
 Use these five (5) indicators for
TAL
Average Variance Extracted (AVE)
 CFA for Transactional Leadership Style (TAL)

Average Variance Extracted (AVE)

Based on the test of Model fit, TAL
comprises five (5) indicators.
Construct Reliability (CR)
 CFA for Transactional Leadership Style (TAL)

Construct reliability (CR) for TAL

 Confirmatory Factor Analysis (CFA)
1. Model Fit (At least 3 fit indices)
 Relative chi-square ≤ 5.0 (Must)
 RAMSEA ≤ 0.08 (Must)
 GFI, NFI, CFI, TLI ≥ 0.90 (At least one must)

2. Factor Loadings ( ≥ 0.5 and 1.0 < )

 High factor loadings ( ≥ 0.5 ) on a factor indicate high convergent validity
(Hair, et al, 2006)

3.Convergent Validity (AVE ≥ 0.5)

Convergent validity refers to a set of variables that presume to measure a
construct (Kline, 2005) - Average Variance Extracted (AVE)
 A high AVE (> .5) indicates a high convergent validity (Fornell& Larcker, 1981)

4. Construct Reliability (CR ≥ 0.70)

 7 Steps in CFA Model Test

1. Draw path diagram

2. Select data file
3. Display variable list
4. Drag variables & assign names
5. Set analysis properties (minimization history, standardized estimates, squared
multiple correlations and modification indices (4-thresholds))
6. Calculate estimates
7. View Text output
 CFA Model Test For

 Transactional Leadership Style (TAL)

 Transformational Leadership Style (TFL)
 Laissez-faire Leadership Style (LFL)
 Employee Motivation (EM)
 Work Engagement (WE)
 CFA for Transformational Leadership Style (TFL)

Path Diagram

 Open CFA Template and save as CFA TFL

 Open list variables in data set
 Drag and drop items TFL1 to TFL7 into
indicator objects
 Label the latent object as TFL
 Click plugins and select name unobserved
variables to label measurement errors (e1
to e7)
 Calculate estimates
CFA for Transformational Leadership Style (TFL)

Analysis Results

Observed for:
 Fit indices
 Standardized Factor Loadings

Note: Make sure these two criteria are

fulfilled.
CFA for Transformational Leadership Style (TFL)

Test for Model Fit

Observed for:
 Fit indices are not fully met
 However all Factor Loadings >0.5
 Refer to modification indices (MI) to
identify indicator for possible
deletion/correlate the two
corresponding measurement errors.
 CFA for Transformational Leadership Style (TFL)

Modification Indices

 Observe for pair of indicators associated

with height MI
 Indicators (e1 and e2) show the highest
MI (79.261)
 Try both options – either to delete
indicator or correlate the two
corresponding measurement errors
 Occurrence for e2 is the highest (4
times) & height MI
 Delete e2 (TFL6)
 CFA for Transformational Leadership Style (TFL)

Modification Indices

 All factor loadings >0.5

 However, All fit indices are adequate except
RMSEA (0.105)
 See the MI again
CFA for Transformational Leadership Style (TFL)

Modification Indices

 Observe for pair of indicators

associated with height MI
 Indicators (e3 & e4) and (e5 & e6)
show the highest MI 13.35 and 12.65
respectively
 Make correlation between these two
relationships
 CFA for Transformational Leadership Style (TFL)

Final Results

Results after deleting and correlation between

the items:
 All fit indices are met including RMSEA
 All factor loadings are more than 0.5

The construct meets the model fit

 Use these six (6) items for TFL
 CFA for Transformational Leadership Style (TFL)

Average Variance Extracted (AVE)

Based on the test of Model fit, TFL comprises

six (6) indicators.
 CFA for Transformational Leadership Style (TFL)

Construct reliability (CR) for TFL

 CFA for Laissez-faire Leadership Style (LFL)

Path Diagram

 Open CFA Template and save as CFA LFL

 Open list variables in data set
 Drag and drop items LFL1 to LFL3 into indicator
objects
 Label the latent object as LFL
 Click plugins and select name unobserved variables to
label measurement errors (e1 to e3)
 Calculate estimates
 CFA for Laissez-faire Leadership Style (LFL)

Analysis Results

 All Fit indices are met

 All Factor Loadings are more than 0.50

The constructs meets the model fit

 Use these three (3) indicators for FLF
 CFA for Laissez-faire Leadership Style (LFL)

Average Variance Extracted (AVE)

Based on the test of Model fit, LFL comprises three (3)

indicators.
 CFA for Laissez-faire Leadership Style (LFL)

Construct reliability (CR) for LFL

 CFA for Employee Motivation (EM)

Path Diagram

 Open CFA Template and save as CFA EM

 Open list variables in data set
 Drag and drop items EM1 to EM7 into indicator
objects
 - Label the latent object as EM
 Click plugins and select name unobserved
variables to label measurement errors (e1 to e7)
 Calculate estimates
 CFA for Employee Motivation (EM)

Analysis Results

 Fit indices are not met

 Two Factor Loadings are < 0.50
(EM6 and EM7)
 Delete the indicator with the
lower factor loading
 Delete EM6 (λ = 0.12)
 CFA for Employee Motivation (EM)

Results Observation

 Fit indices are still not met

 Factor Loading for EM7 is low (λ = 0.42)
 Delete the indicator EM7
 CFA for Employee Motivation (EM)

Results Observation

 Fit indices are still not met

 All Factor Loadings > 0.50
 Refer to Modification Indices (MI) to identify the
potential indicator to be deleted or correlated
 CFA for Employee Motivation (EM)

Modification Indices

 Observe for indicators associated with height

MI
 Indicators (e3 & e4) show the highest MI
75.58

Two options for action

1. Delete one of the related item, or

2. Correlate the above two measurement errors
For deletion, delete the one with the height
occurrence

 Delete e4 (EM4)
 CFA for Employee Motivation (EM)

Results after deleting EM4:

1. Fit indices are met

2. All Factor Loadings > 0.50

The construct meets the Model Fit

 Use these four (4) indicators for
EM
 CFA for Employee Motivation (EM)

Average Variance Extracted (AVE)

Based on the test of Model fit, EM

comprises four (4) indicators.
 CFA for Employee Motivation (EM)

Construct reliability (CR) for EM

 CFA for Work Engagement (WE)

Path Diagram

 Open CFA Template and save as

CFA WE
 Open list variables in data set
 Drag and drop items WE1 to WE7
into indicator objects
 Label the latent object as WE
 Click plugins and select name
unobserved variables to label
measurement errors (e1 to e7)
 Calculate estimates
 CFA for Work Engagement (WE)

Test for Model Fit

 Requirements for fit indices are fulfilled

 However, one factor loading (WE2) is
negative (-ve) and another (WE7) <0.5
 Delete indicator WE2 (λ = -.24)
 CFA for Work Engagement (WE)

Results Observation

 Requirements for fit indices are

fulfilled except RMSEA
 Factor loading for WE7 is less
than 0.50
 - Delete indicator WE7 (λ =
0.35)
 CFA for Work Engagement (WE)

Results Observation

1. Requirements for fit indices are not fulfilled due to

RMSEA
2. All factor loading are more than 0.50

 Refer to Modification Indices (MI) to identify the

potential indicator to be correlated
 CFA for Work Engagement (WE)

Modification Indices
dalowarhossan.bd@gmail.com
 Observe for pair of indicators
associated with height MI
 Indicators (e3 & e4) show the
height MI 4.84
 Correlate the above two
measurement errors (e3 & e4)
 CFA for Work Engagement (WE)

Results after correlation:

1. All fit indices are met

2. All Factor Loadings > 0.50

The construct meets the Model Fit

 Use these five (5) indicators for WE
 CFA for Work Engagement (WE)

Average Variance Extracted (AVE)

Based on the test of Model fit, WE comprises five (5)
indicators.
 CFA for Work Engagement (WE)

Construct reliability (CR) for WE

 Result Presentation for CFA
Table xx: Construct reliability and validity

Variable / Item F.L C.A C.R AVE

Work Engagement 0.85 0.84 0.52
WE1 0.74
WE3 0.69
WE4 0.69
WE5 0.71
WE6 0.78
Laissez-faire Leadership
0.81 0.82 0.59
Style
LFL1 0.76
LFL2 0.75
LFL3 0.81
dalowarhossan.bd@gmail.com
 STEP-2

MEASUREMENT MODEL
Data Set for Measurement Model

Variables:
 Work Engagement five (5) indicators
WE1, WE3, WE4, WE5, WE6

 Transformational Leadership Style six (6) indicators

TFL1, TFL2, TFL3, TFL4, TFL5, TFL7

 Employee Motivation four (4) items

EM1, EM2, EM3, EM5

 Transactional Leadership Style five (5) items

TAL1, TAL2, TAL3, TAL5, TAL6

 Laissez-faire Leadership Style three (3) indicators

LFL1, LFL2, LFL3
Purpose of Measurement Model

 Test for model fit

 Test for discriminant validity
 Test for normality
 Test for outliers
 Test for multicollinearity
Measurement Model – An Overview

1. Test for Model Fit

 Fit Indices (Relative chi-square <5.0, RMSEA < 0.08, GFI/AGFI/CFI/TLI > 0.90)
 Standard Factor Loading (all positive, >0.50, None >= 1.0)

2. Discriminant validity
 Square root of √ AVE > r

3. Normality
 Skewness -2 to +2 (Tabchnick & Fidell, 2007)
 Kurtosis -7 to +7 (Byren, 2010)

4. Outliers (Mahalanobis distance)

 Both P1 & P2 equal 0.000 and 0.000

5. Test for multicolliniearity

 r < 0.90
 Steps in Measurement Model

1. Draw path diagram

2. Run analysis
3. Test for model fit
4. Test for discriminant validity
 Square root of √ AVE > r

5. Test for normality

 Skewness -2 to +2 (Tabchnick & Fidell, 2007)
 Kurtosis -7 to +7 (Byren, 2010)

6. Test for outlier (Mahalanobis distance)

 Both P1 & P2 equal 0.000 and 0.000

7. Test for multicolliniearity

 r < 0.90
 Measurement Model

1. Draw Path Diagram

- Measurement Model represent

the second level in SEM analysis
after CFA
- Include All latent variables in the
analysis and placed as one level
(regardless exogenous or
endogenous)
- In this measurement model, TFL,
TAL, EM, WE and LFL are included
here
 Measurement Model

2. Run Analysis

 Set the analysis properties (minimization history,

standardized estimates, squared multiple
correlations, modification indices and test for
normality & outliers (10-thresholds).
 Click the calculate Estimate button to run the
analysis
 Measurement Model

3. Model Fits Results Observation

 Model fit indices are adequately

meet
 All factor loadings are more than
0.5
 Now test for discriminant validity
 Test for Discriminant Validity

 Discriminant validity refers to the extent in which a construct is truly distinct from other
constructs.
 Discriminant validity involves relationship between a particular latent construct and other
constructs of a similar nature.
 Construct that is truly distinct from other constructs (Hair et al. (2010).

 Discriminant validity can be assessed by two (2) methods:

1. Correlation coefficient
 r ≥ 0.90 indicates high correlation between two constructs and thus violates the discriminant validity
(Fornell and Larcker, 1981; Hair et al.,2010)
2. √AVE and r
 AVE for two constructs against their squired correlation
 AVE for the two factors must be greater than their r2 (valid if √AVE > r ) (Bryne, 2010)
 Result Presentation for Discriminant Validity

 Calculation of Discriminant validity by James Gaskin

1. Open the view text and copy correlations value
2. Open Gaskin’s excel file and paste it on A2
3. Again go back to view text and copy standardized regression weights and paste it on F2
4. Click on red button and observe the results
5. If the √AVE is higher than r values then the constructs shows adequate discriminant validity

CR AVE EM WE LFL TAL TFL

EM 0.843 0.582 0.763
WE 0.842 0.517 0.673 0.719
LFL 0.813 0.592 0.568 0.611 0.769
TAL 0.858 0.549 0.478 0.478 0.408 0.741
TFL 0.874 0.537 0.586 0.734 0.573 0.354 0.733
 CR AVE EM WE LFL TAL TFL
EM 0.843 0.581 0.763
WE 0.848 0.527 0.669 0.726
LFL 0.792 0.561 0.568 0.619 0.749
TAL 0.857 0.548 0.478 0.462 0.428 0.740
TFL 0.885 0.565 0.522 0.674 0.557 0.330 0.752
 Test for Data Normality
 Refer to the “Assessment of normality” in the text view output.
 Data is considered to be normal if
 Skewness is between -2 to + 2 (Tabachnick & Fidell, 2007)
 Kurtosis is between -7 to +7 (Byren, 2010)

dalowarhossan.bd@gmail.com
 Test for Data Normality

If Critical Ratio value is : -1.96 to

+1.96, then data is normally
Distributed at 95% Confidence
Level
 Check for Outliers
High Mahalanobis D2
 Open view text and check Mahalanobis distance result
 High Mahalanobis d-squared with both p1 and p2 equal 0.000 and 0.000
 Delete the cases with p1 & p2 values are 0.000.
 Delete the cases from height to lowest number or, Numbering the respondents (observations) by sequence
before deleting
 Delete the
Observation
Number
which P<0.05
 STEP-3

STRUCTURAL MODEL HYPOTHESIS TESTING

 Structural Model

 Represent the theory with a set of structural equations and is usually depicted with a visual diagram.

 Structural model represents set of one or more dependence relationship linking the hypothesized
model’s constructs.

 This model is most useful in representing the relationships between exogenous and endogenous
variables.

 The focus here is NOT to test the construct validity of the latent variables as in CFA and measurement
model but to examine and test relationship between exogenous and endogenous constructs.

 Can also be used to test direct and indirect effects.

 Structural Model
Draw Path Diagram

 Save the measurement model as structural model

 Delete the correlation path from endogenous
variables (employee motivation and work
engagement)
 Create path between IVs and DV, IVs and Mediator,
and Mediator and DV
 Add the residuals for endogenous variables and
name unobserved variables from plugins
 Select minimization history, standardized estimates,
squired multiple correlation, indirect direct & total
effect, covariance estimates, correlations of
estimates, critical ratios for differences,
modification indices and threshold values 10 from
the analysis properties.
 Calculate estimates
 Structural Model

Set Analysis properties

 Select minimization history,

standardized estimates, squired
multiple correlation, indirect
direct & total effect, covariance of
estimates, correlations of
estimates, critical ratios for
differences modification indices
and threshold values 10 from the
analysis properties.
 Calculate estimates
 Structural Model

Observe the Model Fit Indices

 All the fit indices are adequate

 All factor loadings are more than 0.50
 There is no multi-collinearity

 Therefore, go to the view text to observe

and extract the hypotheses results
 Structural Model

Results extraction

1. Regression weight (unstandardized)

Unstandardized Estimate = b
C.R. = Critical Ratio / t-values (t>1.96)
P = Significant values (p<.05)

2. Standardized Regression Weight

Standardized Estimate = Beta
 Structural Model

Hypotheses Results Presentation

Results extract from view text; dalowarhossan.bd@gmail.com

 See unstandardized regression weight for
values b (unstandardized regression
weight), C.R. (critical ratio / t- values), and
P (significant Value)
 Observe standardized regression weight
for the value of Beta (standardized
regression weight)
 STEP-4

MEDIATION ANALYSIS
(INDIRECT EFFECT TEST)
 Mediation Analysis (Bootstrap Test)

1. Open the existing structural model (save as and rename the structural model as Bootstrap
Mediation test)
2. Assign path labels
3. Create models
4. Set analysis properties
5. Run analysis
6. Extract analysis results
7. Decision & interpretation
 Mediation Analysis (Bootstrap Test)

1. Mediation Test

 Open the existing

structural model
 Rename the file (structural
model) into Bootstrap
Mediation Test
 Mediation Analysis (Bootstrap Test)

2. Assign the 7 path labels

 Mediation Analysis (Bootstrap Test)
2. Procedure for assigning the path labels

 Assign label for LFL ->EM

1. Double click of the arrow head for path LFL
EM
2. At the object properties dialog box, click on
parameters tab and type LFL-EM in the
regression weight box

 Assign label for EM ->WE

1. Click the arrow head for EM WE
2. Type EM-WE in the regression weight box

 Follow the same procedures to label the

rest of the paths
 Mediation Analysis (Bootstrap Test)

3. Create Models

 Currently there is only one

model- “Default Model”
 Rename this Default Model
into Mediation Model
 Create an additional model
and label the new model as
Direct Mode
 Mediation Analysis (Bootstrap Test)

3. Create Models (procedure)

 For the Direct Model, assign

the following 4 parameter
constraints:
LFL-EM=0
TFL-EM=0
TAL-EM=0
EM-WE=0

Note: All path related to EM are

set as 0
 Mediation Analysis (Bootstrap Test)

4. Analysis Properties

Open the Analysis Properties and

click Bootstrap tab and set the
following options;
 Perform bootstrap and set at
5000
 Bias-Corrected (BC) confidence
intervals (CI) and set at 95 (95%
CI)
 Bootstrap ML (Maximum
Likelihood)
 Mediation Analysis (Bootstrap Test)

5. Run Analysis

 Click the calculate estimates

tool to run the analysis
 If red arrow in the output
path diagram is highlighted,
the analysis runs
 See the differences between
Direct Model results and
Mediation Model results
 Mediation Analysis (Bootstrap Test)

6. Results Extraction

 Direct model Beta & P - Find beta value from standardized regression weight and p-value from regression
weight but ensure that you have selected Direct Model.
 Mediation model Beta & P – Find beta value from standardized regression weight and p-value from
regression weight but ensure that you have selected mediation model is selected.
 Mediation Model – SIE & P – Select Mediation model and go to estimates then matrix and select
standardized indirect effect for taking Beta value. After that go to Two tailed significant BC at Estimates /
Bootstrap for p-value.
 Mediation Model – 95% BC CI – Go to Lower Bound BC and collect LB value and then go to upper bound BC
and collect UB value from there.

Put those values to the result presentation file in excel (for result explanation go to thesis AMOS)
 Mediation Analysis (Bootstrap Test)
Mediating effect of EM between LFL and WE
95% Bootstrap
Beta P
Lower Bound Upper Bound
Direct WE <--- LFL 0.253 0.00
Indirect WE <--- LFL 0.175 0.005
Std indirect effect 0.078 0.00 0.037 0.143

Mediating effect of EM between TFL and WE

95% Bootstrap
Beta P
Lower Bound Upper Bound
Direct WE <--- TFL 0.514 0.00
Indirect WE <--- TFL 0.431 0.00
Std indirect effect 0.085 0.00 0.044 0.146

Mediating effect of EM between TAL and WE

95% Bootstrap
Beta P
Lower Bound Upper Bound
Direct WE <--- TAL 0.19 0.00
Indirect WE <--- TAL 0.118 0.025
Std indirect effect 0.072 0.00 0.036 0.126
 Mediation Analysis

6. Decision Criterion

 If in direct model is significant and standardized indirect effect is also significant and the UB and LB both either
positive or negative which means “0” not in between then there is mediation effect.

 If the direct model, mediation model and standardized indirect effect all are significant and the UB and LB both are
either positive or negative which means “0” not in between then there is partial mediation.

 If the direct model, the standardized indirect effect is significant and the UB and LB both are either positive or negative
which means “0” not in between but the mediation model not significant then full mediation.

 If the direct model is not significant and the UB positive and LB negative which means “0” is in between then there is
no mediation.

 If the direct model is significant and mediation model is significant but the standardized indirect effect not significant
and the UB positive and LB negative which means “0” is in between then there is no mediation.
 Mediation Analysis
Mediating effect of EM between LFL and WE
95% Bootstrap
Beta P Decision
Lower Bound Upper Bound
Direct WE <--- LFL 0.253 0.00
Partial
Indirect WE <--- LFL 0.175 0.005
Mediation
Std indirect effect 0.078 0.00 0.037 0.143

Mediating effect of EM between TFL and WE

95% Bootstrap
Beta P Result
Lower Bound Upper Bound
Direct WE <--- TFL 0.514 0.00
Partial
Indirect WE <--- TFL 0.431 0.00
Mediation
Std indirect effect 0.085 0.00 0.044 0.146

Mediating effect of EM between TAL and WE

95% Bootstrap
Beta P Result
Lower Bound Upper Bound
Direct WE <--- TAL 0.19 0.00
Partial
Indirect WE <--- TAL 0.118 0.025
Mediation
Std indirect effect 0.072 0.00 0.036 0.126
 Mediation Analysis (for example)
Mediating effect of XXX between YYY and ZZZ
95% Bootstrap
Beta P Decision
Lower Bound Upper Bound
Direct ZZZ <--- YYY 0.253 0.00
Indirect ZZZ <--- YYY 0.055 0.725 Full Mediation
Std indirect effect 0.078 0.00 0.037 0.143

Mediating effect of XXX between YYY and ZZZ

95% Bootstrap
Beta P Result
Lower Bound Upper Bound
Direct ZZZ <--- YYY 0.114 0.184
Indirect ZZZ <--- YYY 0.431 0.00 No Mediation
Std indirect effect 0.085 0.00 -0.044 0.146

Mediating effect of XXX between YYY and ZZZ

95% Bootstrap
Beta P Result
Lower Bound Upper Bound
Direct ZZZ <--- YYY 0.19 0.00
Indirect ZZZ <--- YYY 0.118 0.025 No Mediation
Std indirect effect 0.072 0.621 -0.036 0.126
 STEP-5
MODERATION TEST (CONTINUOUS MODERATION ANALYSIS)

Continuous Moderator

Culture

Independent Variable Dependent Variable

Compensation Productivity

Figure: Research Framework

 Steps in Moderation Analysis

1. Open the SPSS data file and save the variables compensation, culture and productivity as
standardized values.
2. Create a new variable as Interaction and save it
3. Create the model of continuous moderation
4. Set the analysis properties
5. Run the model and observe the result from view text
 Steps in Moderation Analysis
1. Open the SPSS data file and save the variables compensation, culture and productivity as standardized values.

 Go to Analyze and select descriptive then descriptive

 Select culture, compensation, and productivity
 Save standardized values as variables
 Steps in Moderation Analysis
Open the SPSS data file and save the variables compensation, culture and productivity as standardized values
Steps in Moderation Analysis
Create a new variable as Interaction and save it
 Steps in Moderation Analysis
Compute the new variable as Interaction by multiplying standardized compensation and culture
Steps in Moderation Analysis
Create a new variable as Interaction and save it
Create the Model Of Continuous Moderation
Set the analysis properties and run the model
Observe the Results
 Decision

1. Compensation has a significant effect on productivity

2. Culture has a significant effect on productivity
3. Interaction has no significant relationship with productivity

Decision: This means Culture does not play a significant moderating role in the relationship between
compensation and productivity
 STEP-6
MODERATION TEST
(MULTI-GROUP ANALYSIS)
 Steps in Multi-Group Analysis

1. Open the existing structural model and rename as MGA Moderating Test
2. Create Groups as Male and Female
3. Select data file, grouping variable and group value
4. Create models
5. Set analysis properties and run analysis
6. Test the overall model
7. Test for individual paths
 Steps in Multi-Group Analysis

1. Open the Structural Model

 Open the existing Structural Model

 Rename the path diagram as MGA
Moderation Test

Note: The moderating variable (Gender) is

NOT displayed in the path diagram
 Steps in Multi-Group Analysis

2. Create Groups

 Click Analyze, Manage Groups

 At the Manage Groups dialog box,
rename Group number 1 as Male.
Click NEW button
 Rename Group number 2 as Female.
Click CLOSE button
 AMOS creates two groups as
displayed in the list
 Steps in Multi-Group Analysis

3. Select Data File

 Select the exercise use data set

 In this step, (for each group) you will
perform three tasks
1. Select data file
2. Select grouping variable (Gender)
3. Select group value
(1) Male (2) Female
 Group 1 - Male

1. Click the Male group.

Note the data file has
been selected
2. Click Grouping
Variables button. At
the dialogue box,
select GENDER
3. Click Group Values
button. At the dialogue
box, select 1
4. Then OK
 Group 2 - Female

1. Click the Female group. Note

the data file has been
selected
2. Click Grouping Variables
button. At the dialogue box,
select GENDER
3. Click Group Values button. At
the dialogue box, select 2
4. Then OK
4. Create Models

1. Click Analyze, select Multiple Group

Analysis to create models
2. AMOS displays the following reminder.
Click OK
3. AMOS displays the next model options.
Click OK
 4. Create Models (con…)

AMOS creates six (6) models, however only

TWO models will be used, namely:
1. Unconstrained 2. Measurement residuals

 Unconstrained is the Variant-Group model

which assumes that models for male and
female are DIFFERENT.
Male ≠ Female
 Measurement Residuals is the Invariant-
Group model which assumes models for
male and female are NO DIFFERENCE
Male = Female

dalowarhossan.bd@gmail.com
 5. Run the Analysis

1. Set the following options in the

analysis properties
2. Click calculate estimates button to run
the analysis

Note: AMOS automatically assigned labels

to all paths for the two group
 6. Test the Overall Model

 The first step is to test the presence of

moderation effect on the overall model i.e. to
establish that there is moderation effect of
the moderator variable on the overall model
 This effect can be established by comparing
the Unconstrained (Variant-Group) against
Measurement Residuals (Invariant-Group)
models
 If the Unconstrained model is better than
Measurement Residuals model, then we can
conclude that there is presence of moderation
effect of Gender

Decision Criteria: Unconstrained < Measurement

Residuals = Presence of Moderation Effect
Results Observation
 Click on View Text to observe the results
 Click on the Model Fit and compare the X2 (CMIN) between Unconstrained and Measurement Residuals models
 Both the models are significant (p<0.05), however Unconstrained model is better (X2 is smaller) than
Measurement Residuals Model
 Therefore we can conclude that there is some form of moderation effect in the overall structural model
7. Test for Individual Paths

 Once we have established the presence of moderation effects in the overall

structural model, proceed to test moderation effect on the individual paths

 According to Test of Hair (2010) moderation effect is established for a two-

group moderator if:
 Beta for the one group is significant while beta for the other group is non-
significant
 Both Betas for both groups are significant, however beta for one group is
positive while beta for the other group is negative
 View Text Results

 Refer to the Estimates from

View Text output for both
Male and Female groups
Results to be Extracted

 Refer to both
unstandardized and
standardized
estimates for both
groups (male &
female) and compare
 Steps in Multi-Group Analysis
Results Presentation of Moderation Test
Gender Path coefficient b Beta C.R. P Decision
Male EM <--- LFL 0.367 0.331 3.245 0.001
Does not moderate
Female EM <--- LFL 0.275 0.26 2.98 0.003
Male EM <--- TFL 0.247 0.226 2.469 0.014
Does not moderate
Female EM <--- TFL 0.415 0.423 4.587 0.00
Male EM <--- TAL 0.271 0.312 3.665 0.00
Does not moderate
Female EM <--- TAL 0.213 0.239 3.131 0.002
Male WE <--- LFL 0.12 0.111 1.233 0.218
Moderate
Female WE <--- LFL 0.305 0.23 2.718 0.007
Male WE <--- TFL 0.466 0.437 4.772 0.00
Does not moderate
Female WE <--- TFL 0.585 0.475 4.972 0.00
Male WE <--- TAL 0.144 0.17 2.246 0.025
Moderate
Female WE <--- TAL 0.101 0.09 1.26 0.208
Male WE <--- EM 0.315 0.323 3.386 0.00
Moderate
Female WE <--- EM 0.21 0.167 1.711 0.087

View publication stats