Chapter 4th Revised File
Chapter 4th Revised File
Measurement Models
4.1 Measures
4.1.1 Operationalization and Measurement Items of Construct
4.2 Reliability and Validity of Questionnaire
4.2.1 Pilot Study
4.2.2 Reliability Test
4.2.3 Cronbach’s Alpha
4.3 Exploratory Factor Analysis
4.3.1 EFA on E-service Quality
4.3.2 EFA on Customer Satisfaction
4.3.3 EFA on Repurchase Intention
4.3.4 EFA on Perceived Value
4.4 Measurement Model Assessment (Confirmatory Factor Analysis)
4.4.1 Approaches of CFA in PLS-SEM
4.4.1.1 Extended Repeated Indicator
Approach 4.4.1.2: Two Stage Approach
(a) Embedded Two Stage Approach
(b) Disjoint Two Stage Approach
4.4.2 Stage One in Path Model (For Lower Order Constructs)
4.4.2.1 Factor Loadings
4.4.2.1 Indicator Multicollinearity
4.4.2.3 Reliability Analysis
4.4.2.4 Construct Validity
4.4.2.5 Convergent Validity
4.4.2.6 Discriminant Validity
4.4.2.6(a) Fornell and Larcker Criterion
4.4.2.6(b) HTMT ratio
4.4.3 Stage Two of the Path Model (For Higher Order Construct)
4.4.3.1 Validating Higher Order Construct
4.4.3.1(a) Collinearity
4.4.3.1(b) Significance and Relevance of Outer Weights
4.4.3.1(c) Significance and Relevance of Outer Loadings.
1
This section of the thesis is devoted to ensure that a valid and reliable instrument is used
to achieve the study objectives. The chapter discusses the process by which the
questionnaires were put for reliability and validity tests. The main goal of this chapter
was to create an instrument that could (a) translate the needed information into specific
questions/items that the respondent could easily answer; (b) inspire, motivate, and
encourage the respondent to participate and complete the questionnaire; and (c) reduce
response error.
4.1 Measures
With the fast growth of business to consumer (B2C) electronic commerce, electronic
retailers understand that irrespective of their business category and product offerings, they
have to provide superior service quality over the web termed as e-service quality. Delivering
superior service quality is regarded as an important policy for business growth and survival
(Zeithaml et al. 1996, Reichheld & Schefter 2000). Companies are creating attractive
websites to interact and converse with online customers.
The electronic delivery of retail services varies in many ways from traditional ‘Brick
and mortar’. Online services have distinctive characteristics that offline service do not have,
which can influence the perception of service quality. These features could include, for
example outages of backing up information, connectivity issues and server problems. With
the growing interest in service in online retailing context, an increasing number of research
studies have concentrated on understanding online service quality. Many different scales
measuring online service quality have been developed mainly based on consumer perception
2
or on evaluation of service quality such as WebQual, E-QUAL, SITEQUAL, e-SQ and eTailQ.
Online retailer cannot easily use these scale to assess their current service performance because
these scales do not contain all aspects of the buying process. Electronic retailing (in distinction to
traditional retailing) is not a one and only rather than consistent marketing activity (White & Francis
2004). Thus, web based service systems vary based on product type, service content and channels of
delivery (Voss 2003). There are many research instruments availiable to measure e-service quality,
present study, however,has chosen the research instrument developed by Parasuraman,et.al.,(2005) as it
has been widely reported in many research studies while as (sohn and Tadisina,2008;Gounaris
et.al.,2010;Surajadja et.al.,2012).
The study's questionnaire was broken into four sections. Part I examines the demographic
profile of respondents; Part II to assess e-service quality; Part III was to assess customer satisfaction;
Part IV to assess repurchase intention and Part V to assess perceived value. The scale items of E-service
quality (30 items) were adapted from Parasuraman et al., (2005), to measure customer satisfaction,
seven (07) scale items were adapted from Srinivasan et al.,(2002), to measure repurchase intention
four(04) scale items were adapted from Kim et al.,(2012) and to measure perceived value, three (03)
scale items were adopted from Tsai and Huang,(2007).
To meet the current research objectives, all of the items chosen for the questionnaire were
updated and reworded in terms of both phrasing and contextual uses. A preliminary pool of 70
items was created based on the literature review. After conducting in-depth interviews with
consumers, one more item was added, and each item was double-checked to ensure that it reflected the
study's needs. All of the items in the questionnaire were then ordered according to their ease of
understanding, with the easiest items being placed first, and so on, in order to encourage respondents to
complete the lengthy questionnaire.
Five-point Likert scale was used to measure e-service quality, customer satisfaction, repurchase
intention and perceived value ranging from 1 (strongly disagree) to 5 (strongly
3
agree). The scale was purified in two steps; first a panel of experts assessing the content and face
validity of the scale, and the second, a field test, as suggested by Brady and Cronin (2001) was
conducted. Ten online buyers with three years of experience in online-shopping and three research
faculty members made up the panel. No item was dropped in the initial screening by this panel. The
final questionnaire contained 41 questions, including 30 on e-service quality, 7 on customer satisfaction,
4 on repurchase intention,3 on perceived value construct, and 9 on demographics, after the inclusion,
and paraphrasing of several questions.
The second stage was to undertake a pilot study, for which scale items of e-service quality,
customer satisfaction, repurchase intention and perceived value are given below: -
4
14 Company delivers ordered items within a reasonable time frame.
30 Company offers the option to speak to a live person in case there is a problem.
5
2 Online shopping is a satisfying experience as it offers customized product at my
Convenience.
3 I am satisfied with cash-on-delivery mode of payment offered by this
company.
4 My choice to purchase from this E-commerce company was right.
5 I have truly enjoyed purchasing from this E-commerce company.
6 I am satisfied with my most recent decision to purchase from this E-commerce
Company.
7 I think I did the right thing by buying the products from this E-commerce
company.
1. Products that I have purchased from this E-commerce company are considered of good quality.
2. Products that I have purchased from this E-commerce company are value for money.
3. Products that I have purchased from this E-commerce are worth the money paid.
6
4.2.1 Pilot Study
The major goal of the pilot study was to become familiar with all the items and dimensions
associated with the understudy population and the underlying concept. A pilot study was done on a
small group of e-customers. The investigation's sample is balanced by including customers who
purchase goods and services from different e-retailer platforms which I taken in my study. The
need for this was felt because the sample respondents' needs, aspirations, and perceptions differed
significantly. The questions were thoroughly examined with the sample respondents and
interviews were done with them to get their views and ideas concerning the instrument in order to
ensure relevance and clarity of the items integrated in the measuring scale. A total of 120 people
took part in the pilot trial. Item-total correlation was used to examine the constructs' reliability
and validity.
The following were the final results of the pilot study after careful assessment of respondents'
opinions and conversations with experts in the relevant field: -
Items that were discovered to be too long, ambiguous, or confusing were rewritten to provide
more clarity and precision. As a result of the new version of the instrument being developed, and
finally the items which become relevant to my study are shown below: -
Scale items
s
EFF4 0.722
what I need.
Website of this E-commerce company is
EFF5 0.841
simple to use.
EFF6 Website enables me to get on to it quickly. 0.545
7
Website of this E-commerce company
EFF7 0.711
loads its pages quickly.
Website of this E-commerce company is
SYS1 0.533
always available for shopping.
Website of this E-commerce company does
SYS2 0.809
Availability
not crash.
Pages at this website do not freeze after I
System
SYS3 0.699
enter my order information.
Website of this E-commerce company
SYS4 0.706
launches and runs right away.
FUL1 Company has the items in-stock it claims. 0.477
E-Commerce company delivers orders as
FUL2 0.371
Fulfilment
promised.
Company delivers ordered items within a
FUL3 0.601
reasonable time frame.
The delivery of ordered products of this e-
FUL4 0.523
commerce company is very quick.
FUL5 Company is truthful about its offerings. 0.672
The company doesn’t share my personal
PRI1 0.702
information.
Privacy
creates.
I am compensated by the company for the
COM2 0.554
delay in delivery of ordered items.
8
From the above Table:4.1 it can be seen that Item-total correlation of all the items were above the
minimum threshold of 0.3 (Costello and Osborne, 2005).
Corrected item
Item label
Construct
Correlation
Scale Items
total
I am satisfied with the quality of products
0.700
CS1 offered by this E-commerce company.
Online shopping is a satisfying experience as it
0.594
CS2 offers customized product at my convenience.
Customer Satisfaction
From the Table:4.2 above the reliability analysis was examined to ensure that inter-item correlations
were substantial (>0.30). As it is evident that all the above statements were meeting the minimum
criteria.
9
Table 4.3 Reliability Analysis of Pilot Study (Repurchase Intention)
item total
Corrected
Item label
Construct
Correlatio
Scale Items
n
I intend to continue to purchase more
RI1 products from this E-commerce 0.802
company.
I am very likely to buy products
RI2 online recommended by my friends. 0.620
Repurchase
Intention I intend to recommend this E-
RI3 commerce company to other people. 0.571
Except for any unanticipated reason, I
RI4 intend to purchase products from this 0.467
E-commerce company.
The results of the above Table 4.3 shows reliability analysis from the data collected from the
pilot study. All statements were meeting the minimum threshold of 0.3. All statements were
carried for further data collection and analysis.
item total
Corrected
Item label
Construct
10
4.2.2 Reliability Test
A reliability test was performed to determine the degree of consistency among multiple variable
measurements. The Caramine and Zeller (1979) approach was used to ensure the scale's overall
dependability at each dimension. The method requires a relationship between an item's score and
the items specified, as well as item-to- total correlation and Cronbach's alpha. This was determined
using the correlation matrix shown in the Tables:4.5-4.8 as well as the application of Cronbach's
alpha, which is shown next to the correlation matrix table.
Inter-Item Correlation
EFF1 1.000
SYS1 1.000
System
SYS2 .621 1.000
Availability
SYS3 .431 .711 1.000
11
FUL3 .722 .544 1.000
PRI1 1.000
Privacy
PRI2 .624 1.000
RES1 1.000
COM1 1.000
Compensation
COM2 .547 1.000
CON1 1.000
Contact
CON2 .654 1.000
12
Table 4.6 Inter-item correlation of Customer Satisfaction items
Inter-Item Correlation
CS1 1.000
The above Table: 4.6 shows the inter-item correlation of the customer satisfaction falls between
the accepted value i.e., 0.3 to 0.9.
RI1 1.000
The inter-item correlation of the repurchase intention items displayed in the above Table(4.7) shows
that all lie within the acceptable range of 0.3 to 0.9.
13
Table 4.8 Inter-item correlation of Perceived Value Items
PV1 1.000
The inter-item correlation of the perceived value items displayed in the above Table (4.8) shows
that all lie within the acceptable range of 0.3 to 0.9.
Cronbach alpha is a reliability coefficient that evaluates the overall scale's consistency. Cronbach
alpha has a commonly accepted value of 0.70, while in exploratory studies the value of 0.60 is also
acceptable (Hair et al. 2007). According to Garson (2002) the higher α -score is, the more reliable
the measured construct is. Statistical Package for the Social Sciences (SPSS) is used to assess the
data for validity and reliability. The Cronbach alpha of the constructs for the independent variable
and their dimensions, moderating variable, and dependent variables are provided in Table 4.9.
Table 4.9 Cronbach’s alpha
Cronbach’s Alpha
S. No. Construct
(>0.6 or 0.7)
Efficiency 0.943
System
0.888
Availability
Dimensions
Fulfillment 0.973
E-service
1. Privacy 0.806 0.987
Quality
Responsiveness 0.928
Compensation 0.835
Contact 0.911
14
4.3 Exploratory Factor Analysis
Factor analysis reduces or summarizes a huge number of variables to depict them as separate
factors or components. It's a technique for determining a set of variables of interest which are
linked to a smaller set of unobservable variables. This is accomplished by categorizing variables
based on their inter-correlations. When little is known about the structure and number of
components, EFA can be used (Green et al., 2016).
For the e-service quality construct, an exploratory factor analysis (EFA) using the principal
component analysis method (PCA) was carried out to determine scale dimensionality. To analyze
the underlying variables of 30 items scale of e-service quality, Principal Component Analysis
(PCA) was performed using the statistical software program SPSS (20 Version). The construct
validity was tested by applying Bartlett’s Test of Sphericity and The Kaiser–Mayer–Olkin
Measure of sampling adequacy to analyze the strength of association among variables. The
findings of KMO test are 0.810, which exceeds satisfactory value of 0.50 (Kaiser, 1974).
The results also show the value of Bartlett ‘s test of Sphericity is 0.00, which meets the criteria of
value lower than 0.05. The significant value of Bartlett‘s test of sphericity stipulates that the
sampled data is normally multivariate and relevant for factor analysis and revealed a Chi-Square at
8216.317 (P≤0.000) which verified that correlation matrix was not an identity matrix, thus validating
the suitability of factor analysis (Table 4.8).
Sig. .000
After these preliminary steps, factor analysis with Principal Component Analysis as an extraction
method has been performed on 610 questionnaires. According to the Comrey and Lee (1992), for
conducting EFA a sample size of 50 is extremely poor, 100 is poor, 200 is acceptable, 300 is
good, 500 is very good, and a sample size of 1000 or more is exceptional. To explore the
dimensionality of the thirty-four (30) item scale, the study used R-mode Principle Component-
15
Analysis (PCA) with a Varimax Rotation and Eigen value equal to or more than 1, which extracted
thirty items (30) grouped into seven factors with explained variance of 71.333 percent in the data
(Table 4.9).
According to Hooper (2012) that in social sciences, if the total variance explained by the sample
data is more than 60%, then sample can be considered as good and can be used for further analysis.
However, most of the factor loadings were greater than 0.50 implying a reasonably high
correlation between extracted factors and the individual items. The communalities of thirty-two
(30) items ranged from 0.526 to 0.808 indicating that a large amount of variance has been
extracted by the factor solution.
The seven factors were labelled as F1- Efficiency F2- System Availability F3 -Fulfilment F4 –
Privacy F5 – Responsiveness F6 – Compensation F7 – Contact. The first (efficiency) factor
contained most of the elements(7 respectively) followed by third factor (fulfilment) and fifth
factor (responsiveness) containing (5 elements each) and the remaining fourth factor (privacy),
sixth factor (compensation) and seventh factor (Contact) contained (3 elements each).
Thus, from the results of factor analysis, it is clear that these seven factors can be used to evaluate
the quality of e-service as done in earlier studies by Parasuraman et al.(2005), Madu and Madu,
(2014) and Surjadaja et al., (2019).
Table 4.11 Summary of Results from Scale Purification: Factors, Communalities, Eigen
Value and Explained Variance
Explained Variance
Factor/Dimensions
Communalities
Eigen Value
Item
Elements
No
EFF2 0.746
anywhere. 8.782 16.623
Website enables me to complete the
EFF3 0.729
transaction/s quickly.
16
Company website makes me easy to find what
EFF4 0.734
I need.
17
Website of this E-commerce company is
EFF5 0.689
simple to use.
SYS2 0.712
not crash.
System
F2
3.584 28.238
Pages at this website do not freeze after I enter
SYS3 0.769
my order information.
credit/debit card/s.
policy.
18
RES4 Company takes care of problems promptly. 0.700
representatives/available online.
1 2 3 4 5 6 7
EFF1 0.801
EFF2 0.823
EFF3 .811
EFF4 .813
EFF5 .798
19
EFF6 .753
EFF7 .748
SYS1 .677
SYS2 .813
SYS3 .856
SYS4 .768
FUL1 .768
FUL2 .817
FUL3 .825
FUL4 .767
FUL5 .707
PRI1 .805
PRI2 .883
PRI3 .815
RES1 .667
RES2 .802
RES3 .825
RES4 .819
RES5 .756
20
COM1 .831
COM2 .879
COM3 .854
CON1 .818
CON2 .869
CON3 .797
ExtractionMethod:PrincipalComponentAnalysis.
Rotation Method: Varimax with Kaiser Normalization.
Exploratory factor analysis (EFA) with a principal component method was again conducted on
customer satisfaction construct in order to detect the scale’s dimensionality. In other words,
Principal Component Analysis (PCA) was carried out to explore the underlying factors of 7 items.
The construct validity was tested by applying Bartlett’s Test of Sphericity and The Kaiser–Mayer–
Olkin Measure of sampling adequacy to analyze the strength of association among variables. The
result of the Bartlett’s Test of Sphericity is 0.000, which is lower than 0.05 (or 5%) (Bartlett,
1950), which meets the criteria of value lower than 0.05 in order for the factor analysis to be
considered appropriate. Furthermore, the Kaiser-Meyer-Olkin (KMO) Measure Sampling
Adequacy is 0.875 which is between 0.5 and 1.0 (Cerny & Kaiser, 1977) revealed a Chi-Square at
2828.205, (P≤0.000) which verified that correlation matrix was not an identity matrix, thus
validating the suitability of factor analysis (Table 4.12).
Sig. .000
21
The study applied R-mode Principle Component Analysis (PCA) with a Varimax Rotation
and Eigen value equal to or more than 1, which extracted seven items (7) with an explained
variance of 65.220 percent.(Table 4.13) Communalities of seven (7) items values above 0.50
ranged from 0.594 to 0.725, indicating a moderately good correlation between the individual
items. No factor was extracted from the analysis
Communalitie
Eigen Value
Explained
Construct
Variance
Item
Elements
s
No
22
I think I did the right thing by buying the
CS7 0.601 0.204
products from this E-commerce company.
Exploratory factor analysis (EFA) with a principal component method was again conducted on
Repurchase Intention to detect the scale’s dimensionality using the Principal Component Analysis
(PCA). The construct validity was tested by applying Bartlett’s Test of Sphericity and The
Kaiser– Mayer–Olkin Measure of sampling adequacy to analyze the strength of association among
items. The Bartlett's Test of Sphercity yielded a result of 0.000, which is less than 0.05 (Bartlett,
1950), which satisfies the requirement of a value less than 0.05 for the factor analysis to be
accepted as adequate. The measure of Sampling Adequacy by Kaiser-Meyer- Olkin (KMO) is
0.807 which is above the minimum value of 0.6 (Hoque et al., 2016).
Sig. .000
The study applied R-mode Principle Component Analysis (PCA) with a Varimax Rotation and
Eigen value equal to or more than 1, which extracted four items (4) with an explained variance of
74.994 percent. (Table 4.15) Communalities of four (4) item were above 0.50 ranged from 0.680
to 0.807, indicating a moderately good correlation between the individual items. No factor was
extracted from the analysis.
Table 4.15 Summary of Results from Scale Purification: Communalities, Eigen Value
and Explained Variance
Communalities
Eigen Value
Explained
Construct
Variance
Item
Elements
No
23
I am very likely to buy products online
RI2 0.807 0.481
recommended by my friends.
Exploratory factor analysis (EFA) with a principal component method was again conducted on
Perceived value in order to detect the scale’s dimensionality using the Principal Component
Analysis (PCA). The construct validity was tested by applying Bartlett’s Test of Sphericity and
The Kaiser– Mayer–Olkin Measure of sampling adequacy to analyze the strength of association
among items. The Bartlett's Test of Sphercity yielded a result of 0.000, which is less than 0.05
(Bartlett, 1950), which satisfies the requirement of a value less than 0.05 for the factor analysis to
be accepted as adequate. The measure of Sampling Adequacy by Kaiser-Meyer- Olkin (KMO) is
0.737 is above the minimum value of 0.6 (Hoque et al., 2016).
Sig. .000
The study applied R-mode Principle Component Analysis (PCA) with a Varimax Rotation and
Eigen value equal to or more than 1, which extracted three items (3) with an explained variance of
82.461 percent.(Table 4.17)Communalities of three (3) items were above 0.50 ranged from 0.802
to 0.863, indicating a moderately good correlation between the individual items. No factor was
extracted from the analysis.
24
Table 4.17 Summary of Results from Scale Purification: Communalities, Eigen Value
and Explained Variance
Communalities
Eigen Value
Explained
Construct
Variance
Item
Elements
No
The measurement model is evaluated to determine the quality of the constructs in the study. The
evaluation of the quality criteria begins with the factor loadings, and then moves on to construct
reliability and construct validity.
Two most prominent approaches are recommended for specifying and estimating Higher Order
Constructs (HOC) in reflective-formative type higher order constructs (Ringle, et. al., 2012) and
are discussed below: -
In the repeated indicators approach, all indicators of the lower-order components are assigned to
the higher-order component (Lohmöller, 1989; Wold, 1982). When a higher-order construct
consists of lower-order components (dimensions), each measured with the indicators (items), the
higher-order component would be measured with the same indicators as the lower order
components. Becker, et. al., (2012) have evaluated (extended) repeated indicators for higher-order
reflective-formative constructs. The findings suggest that using (extended) repeated indicators
produces smaller biases in estimating the measurement model for higher- order constructs (i.e., the
relations between lower- and higher-order components).
1. Embedded two stage approach: The first stage of the embedded two-stage approach
corresponds to the standard repeated indicators approach, which has an antecedent
construct in the structural model and produces a non-significant path coefficient estimate
from the antecedent construct to the higher-order component. Instead of interpreting the
model estimates, scores of all constructs are saved in the model and add these as new
variables to the dataset. In stage two, the construct scores are used as indicators in the
higher-order construct’s measurement model (Hair et. al., 2019).
2. Disjoint two stage approach: In stage one of the path model, the disjoint method uses
only the lower-order components of the higher-order construct (i.e., without the higher-
order component). All other constructs to which the higher-order construct is theoretically
related are directly linked to these. Researchers must then save the construct scores, but
just those of the lower-order components, to carry out the disjoint two-stage procedure.
These scores are then utilized to measure the higher-order construct in stage two (Hair et.
al., 2019).There is no compelling reason to choose one over the other (Cheah et al., 2019), because
both produce similar results. The current study used a two-stage disjoint strategy, in which latent
variable scores are preserved and used as formative indicators for higher order variables.
4.4.2 Stage one of the Path Model (For lower order constructs)
In stage one of the path model measurement model assessment is done through evaluating
Internal consistency (Factor loadings, VIF, Cronbach’s alpha, composite reliability),
Convergent validity (indicator reliability, average variance extracted) and Discriminant
validity (Hair et, al., 2019).
4.4.2.1 Factor Loadings
The extent to which each of the items in the correlation matrix correlates with the specified
principal component is referred to as factor loadings. Factor loadings vary from -1.0 to +1.0, with
larger absolute values suggesting a stronger relationship between the item and the underlying
factor (Pett et al., 2003). The factor loadings on all of the items in the study were greater than the
acceptable value.
26
Factor Loadings
EFF1 0.820
EFF2 0.854
EFF3 0.853
EFF4 0.863
EFF5 0.830
EFF6 0.812
EFF7 0.781
SYS1 0.644
SYS2 0.830
SYS3 0.882
SYS4 0.847
FUL1 0.818
FUL2 0.851
FUL3 0.868
FUL4 0.796
27
FUL5 0.774
PRI1 0.876
PRI2 0.927
PRI3 0.884
RES1 0.783
RES2 0.856
RES3 0.880
RES4 0.840
RES5 0.795
COM 1
0.874
COM 2
0.878
COM 3
0.893
CON1 0.875
CON2 0.917
CON3 0.881
CS1
CS2 0.759
CS3
28
0.809
CS4 0.821
CS5 0.846
CS6 0.808
CS7 0.807
RI1 0.827
RI2 0.895
RI3 0.891
RI4 0.849
PV1 0.904
PV2 0.930
PV3 0.889
Table 4.18 shows the factor loadings consisting of three sections. EFF1 to CON3 represents E-
service Quality items, CS1 to CS7 represents Customer Satisfaction items, the third section RI1 to
RI4 represents Repurchase Intention items and the fourth section PV1 to PV3 represents Perceived
Value. The values of the factor loadings are also commonly called indicator reliability. The factor
loadings of all indicators should be statistically significant. Common rule of thumb is that the
standardized loadings should be above 0.7.As it is evident from the above Table (4.18) that all the
items have loadings above the minimum threshold value. Blindly eliminating indicators when their
factor loading is below 0.70, researchers should carefully examine the effects of item removal on
the composite reliability, as well as on the content validity of the construct. Generally,
indicators with factor loadings below 0.70 should be considered for removal from the scale only
29
when deleting the indicator leads to an increase in the composite reliability or the average variance
extracted (Hair et al., 2011). The validity of the discriminant and convergent measures were
examined in the current study to ensure that the removal of these indicators with low factor
loadings did not affect them. Hence these items were retained.
4.4.1.1 Indicator Multicollinearity
Multi-collinearity is a statistical phenomenon where two or more variables in a regression model are
strongly correlated (Daoud 2017). To assess multicollinearity in indicators, the Variance Inflation
Factor (VIF) statistic is used (Fornell and Bookstein, 1982). According to Heir et al., (2016),
multicollinearity should be less than 3.
30
PRI2 2.510 PV3 1.971
PRI3 1.742
COM1 1.206
COM2 1.533
As shown in the table the VIF values for the study's indicators are below the
required threshold below 3. So, there are no issues of multicollinearity in these indicators.
4.4.2.2 Reliability Analysis
Composite Reliability
Construct Dimensions
(>0.6 or 0.7)
0.940
Efficiency
0.880
System Availability
Fulfilment 0.912
E-service quality Privacy 0.924
Responsiveness 0.918
Compensation 0.913
Contact 0.920
31
composite reliability above the required threshold. As a result, construct reliability is
established.
The degree to which multiple attempts to measure the same idea is known as convergent
validity. The metrics used for evaluating constructs convergent validity is the average
variance extracted (AVE) for all indicators of each construct. when the AVE value is greater
than or equal to the recommended value of 0.50, It indicates that items converge to assess the
underlying construct and therefore convergent validity is demonstrated (Fornell and Larcker,
1981).
Convergent Validity of E-service quality, Customer
Satisfaction, Repurchase Intention and Perceived Value
Average variance
Construct Dimensions
Extracted (>0.5)
Efficiency 0.690
System Availability 0.649
Fulfilment 0.676
E-service Quality Privacy 0.802
Responsiveness 0.692
Compensation 0.777
Contact 0.794
The AVE is the mean value of the squared loadings of the indicators associated with the
construct (i.e., the sum of the squared loadings divided by the number of indicators).
Therefore, the AVE is equivalent to the communality of a construct. AVE indicates that
construct explains 50 percent or more of the indicators’ variance that make up the construct
32
(Hair et al., 2022). It is evident from the Table 4.21 above that the study's convergent validity
results based on the AVE statistics demonstrate that the AVE values are greater than 0.50. As
a result, convergent validity is established.
4.4.1.1 Discriminant Validity
The degree to which measures of different ideas are distinct is known as discriminant
validity. The idea is that if two or more concepts are distinct, appropriate assessments of each
should not correlate highly (Bagozzi et al., 1991). Discriminant validity is established
through the following statistical techniques.
Fornell and Larcker (1981) proposed the traditional metric and suggested that each
construct’s AVE (squared variance within) should be compared to the squared inter-construct
correlation (as a measure of shared variance between constructs) of that same construct and
all other reflectively measured constructs in the structural model – the shared variance
between all model constructs should not be larger than their AVEs. The square root of AVE
(in bold) for all the constructs was found to be stronger than its connection with all other
constructs in this study. As a result, there is substantial evidence that discriminant validity is
established.
Fornell and Larcker Criterion of e-service quality dimensions, customer satisfaction, repurchase intention and perceived
value
COM 0.882
33
RES 0.149 0.386 0.259 0.504 0.384 0.209 0.168
Note; COM: Compensation; CON: Contact; CS: Customer Satisfaction; EFF: Efficiency;
FUL: Fulfillment; PRI: Privacy; PV: perceived value; RES: Responsiveness; RI:
Repurchase intention; SYS: System Availability.
The HTMT is defined as the mean value of the indicator correlations across constructs (i.e., the
heterotrait–heteromethod correlations) relative to the (geometric) mean of the average
correlations for the indicators measuring the same construct. Discriminant validity problems
are there when HTMT values are high. Henseler et al. (2015) propose a threshold value of
0.90 for structural models with constructs that are conceptually very similar. In such a setting,
an HTMT value above 0.90 would suggest that discriminant validity is not present. But when
constructs are conceptually more distinct, a lower, more conservative, threshold value is
suggested, such as 0.85 (Henseler et al., 2015). The below Table(4.23) represents the
Heterotrait-Monotrait Ratio (HTMT) data of lower order constructs of the study. The results
displayed in Table indicate that the HTMT criterion was met as all the values of the lower
order constructs in the table are less than 0.85. All model evaluation criteria for the reflectively
measured constructs have been met, providing support for their discriminant validity.
COM
CON 0.262
CS 0.292 0.372
34
RES 0.171 0.439 0.286 0.555 0.435 0.236 0.187
E-service Quality is a higher-order construct in the present study, based on seven lower-order
35
constructs as: Efficiency, System Availability, Fulfillment, Privacy, Responsiveness,
Compensation, and Contact. Collinearity and outer weights are tested in PLS-SEM to measure
higher order construct validity.
4.4.3.1(a) Collinearity
High correlations between indicators in formative measurement models are unlikely since the
indicators are not essentially interchangeable. Collinearity is defined as a high correlation
between two formative indicators (Hair et al. 2014). The variance inflation factor (VIF) is used
to determine the level of collinearity in PLS-SEM..VIF ≥ 3 suggests the possibility of a
collinearity problem (Hair et al., 2011).
Efficiency 1.118
Responsiveness 1.269
Compensation 1.247
Contact 1.425
The above Table 4.24 demonstrates the variance inflation factor VIF values of e- service quality
which is the higher order construct in the present study. As it is evident that all the VIF values
are below the value of 3 so there are no collinearity issues in the higher order construct of the
study.
It's a crucial factor for assessing a formative indicator's impact. Outer weight is the result of
multiple regression with the latent variable scores as the dependent variable and the formative
indicators as the independent variables (Hair et al., 2017). Weights should be significant (as
demonstrated by p values less than 0.05). When the outer weight of an indicator is insignificant
but the outer loading is greater than 0.50, the indicator is generally retained. (Hair et al., 2017).
36
Original Sample Standard
T Statistics
Sample Mean Deviation P Values
(|O/STDEV|)
(O) (M) (STDEV)
The data on Table (4.25) shows all the dimensions of e-service quality have p values less than
0.05 except the outer weight of compensation dimension. The next step is to check outer
loadings of the compensation indicator. When the outer loadings of this indicator have
significant p value, the indicator is retained (Hair et al., 2017).
Outer Loadings
When an indicator weight is not significant, it is not always assumed that the measurement
model is of poor quality. Here, the researcher is advised to take into account a formative
indicator's absolute contribution to the construct, which is determined by its outer loading
(Cenfetelli & Bassellier, 2009). A formative indicator's loading should, at the very least, be
statistically significant. Even if an indicator doesn't provide a large relative contribution to the
construct, loadings of 0.5 and higher indicate that the indicator makes a sufficient absolute
contribution to the construct (Hair et al., 2021).
37
Table 4.26 Outer Loadings of Higher Order Construct (HOC)
Compensation ->E-service
0.510 0.502 0.067 6.046 0.000
Quality
As it is evident from the above Table (4.26) that the compensation dimension has loadings
above 0.50 and has a significant p value which is the required criteria for the validity of the
formative construct. Hence, all the criteria were met, the validity Higher Order Construct (e-
service quality) was established.
As in the fig 4.2 the circles (blue) represent the constructs of present study i.e., e- service
38
quality, customer satisfaction, repurchase intention and perceived value model and the
indicator variables are measured by the rectangles reflect factors. The indicators of the e-
service quality i.e., dimensions are combined to form it without any assumptions about the
intercorrelation patterns among them (Garson, 2016). The direction of causality is from the
indicators to the construct, and the weights of formative indicators represent the importance of
each indicator in explaining the variance of the higher order construct (Hreats et al., 2013). The
values in the circles of customer satisfaction and repurchase intention are the coefficient of
determination (R2). The R2 represents the variance explained in each of the endogenous
constructs and is a measure of the model’s explanatory power (Shmueli & Koppius, 2011), also
referred to as in-sample predictive power (Rigdon, 2012).
Summary
Assuring the validity and reliability of the construct measures is the aim of reflective
measurement model assessment, which supports their acceptability for inclusion in the path
model. The important requirements include discriminant validity, convergent validity, internal
consistency reliability (Cronbach's alpha, reliability, and composite reliability), and indicator
reliability. A construct is said to be convergently valid if it explains for more than half of the
variance of the indicator under evaluation. Establishing discriminant validity, ensures that each
construct is empirically distinct and captures a phenomenon not reflected by other constructs in a
statistical model and is a crucial component of validity assessment. Recent study has shown that
the HTMT criterion should be the preferred option, even though the Fornell-Larcker criterion has
long been the main criterion for discriminant validity assessment. When generating confidence
intervals for the HTMT, researchers should use bootstrapping to determine whether the data
significantly vary from a given threshold. If they satisfy all of these criteria, reflective
measurement models are acceptable for additional PLS-SEM studies.
39
40
.
41
42
43