CHAPTER 4.
RESEARCH RESULT
4.1. The scale-testing result
4.1.1. Internal Consistency Reliability - Cronbach’s Alpha
According to Hoang & Chu (2008), the Cronbach's Alpha coefficient scale is
considered very good when it ranges from 0.8 to close to 1, good when it falls
between 0.7 and close to 0.8, and acceptable when it ranges from 0.6 to close to 0.7.
However, an excessively large Cronbach's Alpha coefficient, approximately 0.95 or
higher, may suggest an overlap in the observed variables in the scale, where they are
not sufficiently distinct from each other (Nguyen,2012). After reliability analysis, it
was found that the scales for social infuence, self – congruity, brand warmth, brand
activitism, information seeking, social media marketing, hedonic motivation, online
trust, social customer engagement, electronic word-of-mouth, brand loyalty, artificial
intelligence all in the acceptable value (above 0.6) and met the good scale standard
after reliability analysis.
CRONBACH’S ALPHA
SI 0.707
SC 0.719
BW 0.775
BA 0.701
IS 0.696
SMM 0.787
HM 0.764
OT 0.795
SCE 0.873
BL 0.801
EWOM 0.769
AI 0.808
Table 4.1. Cronbach’s Alpha
By Authors
4.1.2 Composite Reliability
The sensitivity of the number of observed variables in each scale is a
consideration when using Cronbach's Alpha coefficient for testing. To reduce this
sensitivity, researchers often utilize Composite Reliability, an alternative reliability
measurement method. Nguyen & Cao (2018) argue that the minimum validation
criterion for scales is 0.7. Therefore, each scale needs to be assessed for its content
value before considering elimination if it does not satisfy the aforementioned testing
requirement. The results of the Composite Reliability after further data validation
demonstrate that all scales meet the inherent value.
COMPOSITE RELIABILITY
SI 0.835
SC 0.841
BW 0.869
BA 0.834
IS 0.833
SMM 0.875
HM 0.864
OT 0.879
SCE 0.880
BL 0.882
EWOM 0.880
AI 0.866
Table 4.2. Composite Reliability
By Authors
4.1.3 Convergent Validity – Outer Loading and Average Variance Extracted
Researchers evaluated the concepts by examining their positive correlation
with other concepts in the same scale. The convergence values of these study concepts
were assessed using outer loading and average variance extracted (Nguyen & Cao,
2018).
The higher Outer Loading indicator reaches, the more it proves that the result
variable is conceptually correct. Morever, according to a Primer On Partial Least
Squares Structural Equatiom Modeling (2013), Hair and partner have recommended
that the outer loading value for the meaningful observation should be above 0.7. Hair
ei al (2013) suggest that observations with outer loadings below 0.4 should be
excluded, and when the coefficient falls from 0.4 to 0.7, the decision to exclude or
retain depends on the researcher’ assessment along with other indices such as
composite reliability (CR) and convergent validity (e.g., AVE) of that factor. In our
model research results we have the outer loadings of IS3 and SCE1 below 0.7. But
composite reliability and convergent validity of Information Seeking and Social
Customer Engagement is really good and qualified. So that we decide to retain 2
observations.
AI AIxS BA BL B E - H IS OT SC SC SI SM
CE W W M E M
OM
AI1 0.8
52
AI2 0.8
23
AI3 0.8
73
BA1 0.7
98
BA2 0.8
16
BA3 0.7
59
BL1 0.8
39
BL2 0.8
48
BL3 0.8
49
BW 0.8
1 04
BW 0.8
2 52
BW 0.8
3 32
HM 0.8
1 1
HM 0.8
2 42
HM 0.8
3 2
IS1 0.8
35
IS2 0.8
46
IS3 0.6
81
OT1 0.8
08
OT2 0.8
49
OT3 0.8
64
SC1 0.8
39
SC2 0.7
15
SC3 0.8
4
SCE 1.46
* AI 1
SCE 0.6
1 86
SCE 0.7
2 36
SCE 0.7
3 99
SCE 0.8
4 08
SCE 0.7
5 15
SCE 0.7
6 04
SI2 0.8
34
SI3 0.8
13
SM 0.8
M1 59
SM 0.8
M2 11
SM 0.8
M3 4
WO 0.8
M1 7
WO 0.8
M2 1
WO 0.8
M3 46
SI1 0.7
29
Table 4.3. Outer loading by Authors
The convergence value was also measured using Average Variance Extracted
(AVE). AVE is defined as the square of the averaged outer loading. When this value
surpasses 0.5, it suggests that the hypothesis can be elucidated by factors that account
for more than 50% of the variation in the observed variables. Consequently, the
findings suggest that all the specified variables, ranging from 0.626 in IS to 0.722 in
AI, contribute significantly to the explanation of the hypothesis.
AVE
SI 0.629
SC 0.640
BW 0.689
BA 0.627
IS 0.626
SMM 0.700
HM 0.679
OT 0.707
SCE 0.552
BL 0.714
EWOM 0.710
AI 0.722
Table 4.4. Average Variance Extracted by Authors
4.1.4. Discriminant Validity
If there actually is a difference between the concepts in the study, that
difference is said to have the discriminant validity. The Heterotrain-Monotrait Ratio of
correlations (HTMT), which is the average of all the correlations between the
observed variables for each concept and the other concept (Nguyen & Cao, 2018). We
use it to quantify. When HTMT is evaluated, the correlation between the concepts
represented by numbers 1, It indicating that the concepts don’t have discriminative
value. Henseler et al. (2015) suggest that if the value is smaller than 0.9, the
discriminant validity is qualified. In constrast, Clark and Watson (1995) and Kline
(2015) utilize a more stringent threshold of 0.85. And 0.9 is the most widely adopted
in practice. The table below shows that all HTMT levels are acceptable (below 0.9).
As a results, this study showed the good wdiscriminant validity.
E-
AI AI*SCE BA BL BW WOM HM IS OT SC SCE SI SMM
AI
AI*SCE 0.508
BA 0.75 0.383
BL 0.857 0.398 0.752
BW 0.63 0.362 0.847 0.498
EWOM 0.898 0.297 0.596 0.626 0.509
HM 0.702 0.351 0.723 0.715 0.592 0.603
IS 0.625 0.266 0.882 0.668 0.519 0.447 0.751
OT 0.667 0.301 0.709 0.597 0.522 0.578 0.874 0.777
SC 0.717 0.473 0.88 0.596 0.881 0.488 0.681 0.69 0.602
SCE 0.873 0.345 0.785 0.812 0.6 0.833 0.731 0.705 0.729 0.664
SI 0.283 0.206 0.517 0.329 0.301 0.321 0.392 0.297 0.356 0.383 0.439
SMM 0.821 0.424 0.783 0.773 0.629 0.595 0.762 0.697 0.646 0.699 0.674 0.254
Table 4.5. Discriminant Validity - Heterotrait-monotrait criteria (HTMT)
Source: Authors
4.2. Research model after testing by Smart - PLS 4
Figure 4.2. Structural model and test results of PLS-SEM by Authors
The results of our structural model testing are displayed are presented above
after testing the scale. The variable generally ensure the validity and reliability of the
model. The next section in the following chapter will offer an explanation of the path
coefficients observed in the figure above.
4.3. Multivariate regression results - Bootstrap 5000 samples
Hair et al. (2016) proposed a sample bootstrapping approach for the PLS-SEM
analysis technique that may replace up to 5000 samples to estimate the standard error
of the observed path coefficient. This helps determine the p-value of the hypothesis. In
marketing and management studies, it is common to use the 95% confidence interval (
equal to the 5% ) for the two-tailed test (Hair et al. , 2016). The outcomes of our
suggest hypothesis evaluate in this study are evaluated using the 5% significance
level.
Original Sample Standard P
Hypothesis Relationships Sample Mean Deviation Values
H1 SI -> SCE 0.111 0.113 0.047 0.018
H2 SC -> SCE 0.056 0.052 0.077 0.460
H3 BW -> SCE 0.077 0.081 0.052 0.142
H4 BA -> SCE 0.161 0.164 0.077 0.043
H5 IS -> SCE 0.083 0.089 0.076 0.275
SMM ->
H6 SCE 0.131 0.131 0.062 0.036
H7 HM -> SCE 0.135 0.137 0.066 0.044
H8 OT -> SCE 0.222 0.216 0.066 0.001
H9 SCE -> BL 0.378 0.38 0.059 0.000
SCE -> E -
H10 WOM 0.688 0.689 0.039 0.000
AIxSCE ->
H11 BL -0.044 -0.042 0.061 0.000
Table 4.6. Hypothesis test Source by Authors
With the results in the table 4.6, we have 11 proposed hypothesis and 9 are
supported, which involving H1, H4, H6, H7, H8, H9, H10, H11. The hypothesis H2,
H3, H5, which suggest a relationship between self – congruity, brand warmth,
information seeking and social customer engagement, is found to be statistically
insignificant, unlike the hypotheses mentioned earlier.
According to the results, online trust (β=0.222) has the highest impact on social
customer engagement. Social influence, brand activitism, social media marketing,
hedonic motivation all have impact on social customer engagement. And the brand
loyalty (β=0.378) and EWOM (β=0.688) was shown to be positively impacted by
social customer engagement. The relationship between social engagement and brand
loyalty moderated by artificial intelligence. But this suggests brand-customer
interaction on social media has been impacted by AI. AI would decrease social
customer engagement, which would help slow down brand loyalty.
Figure 4.3: Structural model 1
Figure 4.4: Structural model 2
R Square R Square Adjusted
BL 0.543 0.538
EWOM 0.473 0.472
SCE 0.533 0.521
Table 4.7: R square
Our study utilizes the coefficient R2 to examinate the fit of the model. Figure
4.3 (Model 1), five independent variables collectively explain 52.1% of the variation
in social customer engagement on social media platforms. In Figure 4.4 (Model 2),
factors such as social engagement and artificial intelligence account for 53.8% of the
variation in brand loyalty. Lastly, social customer engagement contributes to 47.2% of
the variation in electronic word-of-mouth behavior.
5.1.2. Research results
The whole scale is reliable because the Cronbach’s alpha coefficient of the
variables is greater than 0.7, with all values surpassing 0.8, indicating a very good
scale.
When using the Composite Reliability coefficient, all scales demonstrate
reliability, as all coefficients are greater than 0.7.
All observed variables of the independent variables are qualified because the
majority of the outer loading for all observed variables exceeds 0.7.
All HTMT values are less than 0.9, which means that this study demonstrated
good discriminant validity.
After using SmartPLS3 to test the scale, the variable ensure the reliability and
validity of the model. Regarding the factors influencing the relationship between
social customer engagement and brand loyalty, as well as electronic word-of-mouth
(E-WOM), our research indicates that there are five components affecting the
relationship with the following order from the highest to lowest impact: the most
substantial impact is the positive effect of online trust on social customer engagement
(β = 0.222), followed by the positive impact of brand activism on social customer
engagement (β = 0.161). the next one is the positive impact of hedonic motivation on
social customer engagement (β = 0.135). Lastly, the positive impact of social
influence on social customer engagement follows (β = 0.111).
