You are on page 1of 13

Example 5: Steps include EFA, CFA, and Causal Model using AMOS Software

(* Note that in this example, the EFA is first step performed through SPSS, and CFA and Causal
Model are second and third step respectively performed using AMOS.)

Person Job Fit


H1
(Independent Variable 1)

H3

Career Commitment Employee Creativity


(Mediator) H5 (Dependent Variable)
(H6, H7)
H4

Person Organization Fit


(Independent Variable 2) H2 H8, H9

Gender
Psychological Safety
(Moderator Variable) (Moderator)
H10, H11

Theoretical Model.

Hypotheses
H1: Person job fit has positive and significant effects on employee creativity
H2: Person organization fit has positive and significant effects on employee creativity
H3: Person job fit has positive and significant effects on career commitment
H4: Person organization fit has positive and significant effects on career commitment
H5: Career commitment has positive and significant effects on employee creativity
H6: Career commitment mediates the relationship between person job fit and employee creativity
H7: Career commitment mediates the relationship between person organization fit and employee
creativity

Page 1 of 13
H8: Psychological safety moderate the relationship between person job fit and employee
creativity
H9: Psychological safety moderate the relationship between person organization fit and
employee creativity
H10: The relationship between person job fit and employee creativity is significantly different
between male and female
H11: The relationship between person organization fit and employee creativity is significantly
different between male and female

Data Analysis:
Data is analyzed in three steps: Step 1 we performed a preliminary analysis of the scale by
Exploratory Factor Analysis using the Maximum likelihood and Varimax rotation through SPSS.
Step 2 consisted of further validating the factor structure which was the output of EFA sent to
CFA conducted through AMOS. Step 3 was to test the hypotheses by assessing the structural
model using AMOS.

Exploratory Factor Analysis


Exploratory Factor analysis using the maximum likelihood method with Varimax rotation is used
for analyzing the factor structure and correlation between items included in the scale. The results
of rotated factor matrix are provided in the following tables

Table 1: KMO and Bartlett's Test


Kaiser-Meyer-Olkin Measure of Sampling Adequacy. .920
Approx. Chi-Square 8971.048
Bartlett's Test of Sphericity df 465
Sig. .000

The KMO value is above 0.50 so it indicates that criteria of sampling adequacy is met. The
Bartlett test of sphericity is statistically significant (P<.05) so it shows that our correlation matrix
is statistically different from an identity matrix as desired.
. Table 2: Rotated Factor Matrixa
Factor
1 2 3 4 5

Page 2 of 13
PJ1 .595
PJ2 .815
PJ3 .733
PJ4 .701
PJ5 .822
PJ6 .849
PO1 .703
PO2 .730
PO3 .653
CC1 .754
CC2 .777 .320
CC3 .833
CC4 .847
CC5 .831
CC6 .722
CC7 .664
CC8 .726
EC1 .659
EC2 .814
EC3 .873
EC4 .739
EC5 .749
EC6 .376 .545
EC7 .663
PS1 .665
PS2 .824
PS3 .778
PS4 .813
PS5 .799
PS6 .795
PS7 .834
Extraction Method: Maximum Likelihood.
Rotation Method: Varimax with Kaiser Normalization.
a. Rotation converged in 6 iterations.

The results of the exploratory factor analysis show that the solution is based on 5 factors as
expected and all items are loading on their own factors with the exception of only two items
having cross-loadings (CC2 & EC6). The five-factor solution is explaining 65.3% variance of the
total variance. The results of the exploratory factor analysis show that our factors have good
level of validity. For further validation, we used the confirmatory factor analysis (CFA) which is
discussed next.

Confirmatory Factor Analysis


The AMOS version 18 is used for performing the Confirmatory Factor Analysis (Arbuckel,
2009). The model is assessed for testing the reliability, convergent validity, and discriminant

Page 3 of 13
validity. Following is the graphical representation of CFA initial model and the final calculated
model followed by results in the table 3.

Figure 1: Initial CFA Model

Page 4 of 13
Figure 2: Final CFA Model

Page 5 of 13
Table 3: Reliability and Convergent Validity
Variables/ Standardized Cronbach Composite Average Maximum
Constructs Items Factor Loadings Alpha Reliability Variance Shared Variance
Extracted
PJ1 .642
PJ2 .856
PJ3 .771
Person Job Fit .910 .911 .632 .168
PJ4 .768
PJ5 .844
PJ6 .865
PO1 .729
Person
PO2 .781 .772 .774 .534 .128
Organization Fit
PO3 .678
CC1 .815
CC2 .875
CC3 .893
Career CC4 .912
.947 .947 .693 .356
Commitment CC5 .866
CC6 .777
CC7 .721
CC8 .780
EC1 .712
EC2 .844
EC3 .888
Employee
EC4 .781 .915 .917 .615 .356
Creativity
EC5 .827
EC6 .671
EC7 .746
PS1 .740
PS2 .860
PS3 .829
Psychological
PS4 .859 .937 .938 .683 .168
Safety
PS5 .823
PS6 .824
PS7 .846
Model Fitness: X2=1270.52, df=424, X2/df= 2.99, RMSEA=.074, RMR=.034, GFI=.813, CFI=.904

The result of CFA shows that model had good fit statistics including x2/df=2.99, RMSEA of
0.074, RMR of 0.034, and CFI of .904. The recommended values are provided in the bracket
based on the guidelines of Hu and Bentler (1999) and Browne and Cudeck (1992) (RMSEA<.08,
RMR<.05, CFI>.90). All items standardized factor loading was above 0.60 and AVE is also
above 0.50 so it is an indication of good convergent validity (Hair, Sarstedt, Ringle, & Gudergan,
2017). Another evidence of convergent validity is that Maximum Shared Variance is less than
respective Average Variance Extracted for all variables. The Cronbach alpha and composite
reliability for all variables are above 0.70 so it shows that our variables had good reliability.

Page 6 of 13
Table 4: Discriminant Validity
Person Person Career Psychological Employee
Job Organization Commitment Safety Creativity
Fit Fit
PersonJobFit .795
PersonOrganizationFit .358 .731
CareerCommitment .340 .296 .832
PsychologicalSafety .410 .305 .400 .827
EmployeeCreativity .284 .146 .597 .319 .785

For establishing discriminant validity, we used the Fornell & Larcker (1981) criteria. The values
in the diagonal bold are square root of AVE and other values are inter-variable correlation. The
requirement is that the diagonal bold values should be higher than other values in its respective
rows and column which is met as can be seen in the table. Thus, we can say that our variables
have good discriminant validity.

Hypotheses Testing (Structural Model)


To examine the relationship between person-job fit, person-organization fit, and employee
creativity, we used the structural equation modelling using the AMOS path analysis by imputing
the Factor Score from CFA using AMOS. As part of hypotheses testing, we tested the career
commitment as a mediator and psychological safety as a moderator. Following is the graphical
representation of structural model followed by results.

Figure 3: Proposed Structural Model for Hypotheses Testing

Page 7 of 13
Figure 4: Measurement Model-Results

Table 5: Regression Weights


H. Paths Remarks
Estimate S.E. C.R. P
No.
H1 PersonJobFit>EmployeeCreativity .066 .034 1.962 .050* H1 Supported
H2 PersonOrganizationFit>EmployeeCreativity -.089 .042 -2.101 .036 H2 Not Supported
H3 PersonJobFit>CareerCommitment .196 .038 5.102 *** H3 Supported
H4 PersonOrganizationFit>CareerCommitment .219 .051 4.336 *** H4 Supported
H5 CareerCommitment>EmployeeCreativity .565 .042 13.456 *** H5 Supported
PsychologicalSafety>EmployeeCreativity .066 .036 1.822 .068
Model Fitness: X2=29.69, df=1, X2/df= 29.69, RMSEA=.281, RMR=.027, GFI=.970, CFI=.936
***<.05, **<.01, *<.001
Results indicated a good fit for the model presented including RMR of 0.027, GFI of .970, and
CFI of .936. The RMSEA failed to achieve the desired values as RMSEA should be less than
0.08 for model fitness to achieve.
Hypotheses resulting based on path analysis shows that person job fit is positively and
significantly associated with employee creativity (β=.066, P<.05). Person organization fit is
negatively and significantly associated with employee creativity (β=-.089, P<.05). Person job fit
is positively and significantly associated with career commitment (β=.196, P<.05). Person
organization fit is positively and significantly associated with career commitment (β=.219,
P<.05). Career commitment is positively and significantly associated with employee creativity
(β=.565, P<.05). Psychological safety is positively but insignificantly associated with employee
creativity (β=.066, P>.05). Based on these results, we accept the H1, H3, H4, and H5. We
rejected H2 since p-value is significant but the nature of relationship is negative which is
contrary to our hypothesized nature of relationship.

Page 8 of 13
Mediation Testing
The mediation analysis is conducted by treating person-job fit and person-organization fit as
independent variables, employee creativity as dependent variable, and career commitment as
mediator. The mediation analysis is based on the analysis of indirect effects based on the
guideline by Baron and Kenny (1986) classical approach We performed mediation analysis by
using the direct and indirect effects based on bootstrap procedures (500 samples) and bias-
corrected bootstrap confidence interval (90%). The results are provided in the following table.
Table 6: Mediation Analysis
H. Path Total Direct Indirect Remarks
No. Effects Effects Effects
H6 PJF>CC>EC .177** .066 .111** Hypothesis supported since indirect effects
are statistically significant
H7 POF>CC>E .035 -.089* .124** Hypothesis supported since indirect effects
C are statistically significant
*<.05, **<.01, ***<.001
Result shows that career commitment is partially mediating the relationship between person job
fit and employee creativity as indirect effects are statistically significant (β=.111, P<.05).
Furthermore, career commitment is also partially mediating the relationship between person
organization fit and employee creativity (β=.124, P<.05). Based on these results, we accept the
H6 and H7.

Moderation Testing
The moderation analysis is conducted by treating person job fit and person organization fit as
independent variables, employee creativity as dependent variable, and psychological safety as
moderator variable. The results are calculated by creating interaction terms from standardized
score of variables using SPSS.
Table 7: Moderation Testing
H. Remarks
Estimate S.E. C.R. P
No.
H8 Interaction PersonJobFit*PsycholgoicalSafety> Not
.024 .054 .450 .653
EmployeeCreativity Supported
H9 Interaction PersonOrganizationFit*PsycholgoicalSafety> - Not
-.068 .053 .201
EmployeeCreativity 1.28 Supported
***<.05, **<.01, *<.001

Page 9 of 13
5

4.5

3.5 Moderator
EmployeeCreativity

3 f(x) = 0.176 x + 2.83 Low Psychologi-


f(x) = 0.184 x + 2.63 calSafety
2.5 Linear (Low Psycholog-
icalSafety)
2

1.5

1
Low PJF High PJF
Figure 5: Psychological Safety as Moderator between Person Job Fit and Employee Creativity

4.5

3.5 Moderator
EmployeeCreativity

3 Low Psychologi-
f(x)
f(x) == −− 0.0819999999999999
0.302 x + 3.359 x + 3.217
calSafety
2.5 Linear (Low Psycholog-
icalSafety)
2

1.5

1
Low POF High POF
Figure 6: Psychological Safety as Moderator between Person Organization Fit and Employee Creativity

Page 10 of 13
We tested the psychological safety as a moderator. Result indicate that interaction term of
person-job fit and psychological safety exerts negative and insignificant influence on employee
creativity (β=-.002, P>.05) so we reject the H8. Additionally, the interaction term of person-
organization fit and career commitment influence on employee creativity is positive but
insignificant (β=.055, P>.05) so we reject the H9. The result shows that there is no statistical
support for the moderating role of career commitment in our data which is contrary to
hypothesized nature of relationship.

Gender as Moderator Variable- Multigroup Analysis


Our last two hypotheses are related to Gender as moderator variable. We hypothesized that
person-job fit and person-organization fit influence on employee creativity is influenced by
gender as control variable (H10, H11). Before testing the multigroup analysis, we need to ensure
measurement model invariance (need to asses that measurement of constructs is same for both
groups). The three types of measurement model invariance are configural, metric, and scalar.
The configural invariance is tested by creating two separate models (male & female) and then
comparing the model fit based on estimating two groups freely i.e. without constraints. We
found satisfactory model fit after revising the model based on two separate models
(X2=1385.975, df=836, X2/df=, RMSEA=.043, RMR=.044, GFI=.815, CFI=.939) so it shows
that we have achieved the configural invariance. Next, we moved to test the Metric invariance.
We performed this by chi-square difference test by comparing the chi-square of fully
unconstrained model versus constrained model where the regression weights were constrained.
The result shows that our model achieved the Metric invariance test (X2 for Unconstrained
model=1385.975, df=836, X2 for fully constrained model-Measurement weights=1403.56,
df=862, ΔX2 =17.585, P>.05). The Scalar invariance is tested by comparing the intercepts and
structural covariances. We did this by comparing fully unconstrained model with the fully
constrained intercept model and fully constrained structural covariance model. The result shows
that there is no significant difference between the fully unconstrained model and fully
constrained intercept model (X2 for Unconstrained model=1385.975, df=836, X2 for fully
constrained model-Measurement intercepts=1448.04, df=893, ΔX2=62.06, P>.05). The result
also shows that there is no significant difference between fully unconstrained model and fully
constrained structural covariance model (X2 for Unconstrained model=1385.975, df=836, X2 for

Page 11 of 13
fully constrained model-Structural Covariance intercepts=1474.96, df=908, ΔX2=88.99, P>.05).
Based on insignificant difference between these models, we can say that our model achieved the
scalar invariance. Thus, overall, we achieved the configural, metric, and scalar invariance so we
can move to test the Multigroup analysis and compare the coefficients between male and female
groups.
Table 8: Comparison of Coefficients and Chi-Square Difference between Male and Female
for Multigroup Analysis
H.No. Beta for Beta for Chi-Square Remarks
Male Female Difference
(ΔX2)
H10 Person-Job Fit >Employee Creativity .021 .186** 5.467* Supported
H11 Person-Organization Fit > Employee -.095 -.045 .260 Not
Creativity Supported
***<.05, **<.01, *<.001
The beta coefficients were obtained by constraining the measurement weights for the
hypothesized relationship based on gender one by one. The result for person-job fit on employee
creativity showed that regression weights were .021 for male and .186 for female. The chi-square
difference indicates that the difference between the both groups are statistically significant so we
accept the H10. The results for the person-organization fit and employee creativity shows that
regression weights for male were -.095 and for female, it was -.045. The chi-square result shows
that the results are statistically insignificant so we reject the H11.

Table 9: Summary of Hypotheses


H.No. Statement Status
H1 Person job fit has positive and significant effects on employee creativity Supported
H2 Person organization fit has positive and significant effects on employee creativity Not Supported
H3 Person job fit has positive and significant effects on career commitment Supported
H4 Person organization fit has positive and significant effects on career commitment Supported
H5 Career commitment has positive and significant effects on employee creativity Supported
H6 Career commitment mediates the relationship between person job fit and employee Supported
creativity
H7 Career commitment mediates the relationship between person organization fit and Supported
employee creativity
H8 Psychological safety moderates the relationship between person job fit and employee Not Supported
creativity
H9 Psychological safety moderates the relationship between person organization fit and Not Supported
employee creativity
H10 The relationship between person job fit and employee creativity is significantly Supported
different between male and female
H11 The relationship between person organization fit and employee creativity is Not supported
significantly different between male and female

Page 12 of 13
References
Arbuckle, J. L. (2009). Amos (Version 18.0). Crawfordville, FL: Amos Development
Corporation.
Baron, R. M., & Kenny, D. A. (1986). The moderator–mediator variable distinction in social
psychological research: Conceptual, strategic, and statistical considerations. Journal of
personality and social psychology, 51(6), 1173.
Browne, M. W., & Cudeck, R. (1992). Alternative ways of assessing model fit. Sociological
methods & research, 21(2), 230-258.
Fornell, C., & Larcker, D. F. (1981). Structural equation models with unobservable variables and
measurement error: Algebra and statistics.
Hair Jr, J. F., Sarstedt, M., Ringle, C. M., & Gudergan, S. P. (2017). Advanced issues in partial
least squares structural equation modeling. saGe publications.
Hu, L. T., & Bentler, P. M. (1999). Cutoff criteria for fit indexes in covariance structure analysis:
Conventional criteria versus new alternatives. Structural equation modeling: a
multidisciplinary journal, 6(1), 1-55.

Page 13 of 13

You might also like