Download as docx, pdf, or txt
Download as docx, pdf, or txt
You are on page 1of 9

CHAPTER 3: RESEARCH METHODOLOGY

3.1. Research design


This study used the descriptive research design. It is a theory-based design method that is
created by gathering, analyzing, and presenting collected data. This allows a researcher to
provide insights into the why and how of research. Descriptive design helps others better
understand the need for the research.
In this research, the quantitative data analysis approach was adopted, which is for cases
where statistical conclusions to collect actionable insights are essential. Numbers provide a better
perspective to make critical business decisions. Quantitative research methods are necessary for
the growth of any organization. Insights drawn from hard numerical data and analysis prove to
be highly effective when making decisions related to the future of the business. Especially, this
approach is chosen in this study to test the relationship among variables.
Primary data refers to information and data that are gathered directly for the typical
purposes of any research (C.R.Katharin, 2004). An advantage of using primary data is that
researchers are collecting information for the specific purposes of their study. In this study, the
questions the researchers ask are tailored to elicit the data that will be suitable for the online
shopping study context. Thus, the author will have data and responses via online questionnaires.
The primary one can be gathered through a variety of techniques, including focus groups,
interviews, telephone interviews, and online questionnaires. The author chose the online
questionnaire survey from among the available options since it can rapidly reach focus groups
and offer convenience and flexibility to the participants.
Additionally, there are three main research techniques: experimental, observation, and
questionnaire. In particular, the questionnaire is typically used to collect primary data (Kumar
Ranjit, 2019). It is a tool for gathering data from a large sample of the population rather than
concentrating on a specific individual. Furthermore, the survey is very simple, time-saving, and
generates a huge number of data in a short amount of time (Kelley et al., 2003). Also, it is
comparatively precise and dependable, making it appropriate for investigations on attitudes and
behaviors. The survey can then be used to collect data for this study based on these justifications.
3.2. Sampling method
The study's sample was chosen using non-probability judgemental sampling techniques.
In judgmental sampling, the researchers choose samples based on their experience to complete a

1
task in a way that makes sure all of the members have similar traits (Taherdoost, 2016). It is
applied when the respondent was asked if they have experience using a certain product or doing
some actions (Alchemer, 2018). Judgmental sampling is suitable for this study because the
respondents are required to be iGen (who were born between 1995 - 2010) and need to have
smartphones that allow them to access the Internet. Thus, non-probability sampling with
judgmental is chosen in this study.
3.3. Sample size
One of the typical sampling methods that can be mentioned is the 10 times rules (Rigdon
et al., 2017). According to this method, the sample size of PLS-SEM studies should “equal to
the larger often times the largest number of formative indicators used to measure a single
construct, or ten times the largest number of structural paths directed at a particular construct in
the structural model” (Rigdon et al., 2017). In the research model of this article, there are 4
hypotheses indicating 4 paths directed at online shopping intention. Thus, the minimum size
required is 40. In this study, the sample size of 222 has clearly exceeded 40 (ten times the
number of paths directed at online shopping intention).
Another popular sampling method can be Daniel Soper. Daniel Soper (Westland, 2010)
was also referred to in the calculation of the sample size for the present research. This method is
an easy-to-use online tool that has academic support. Daniel Soper will calculate the sample size
required for this study that uses a structural equation model (SEM), given the number of
observed and latent variables in the model are 14 and 4 respectively. The anticipated effect size
is defaulted as 0.30, and the desired probability and statistical power levels are 0.05 and 0.8
respectively. Therefore, the total sample usable for data analysis comprised 222 responses, which
exceeds the minimum sample size required as 137.
3.4. Questionnaire design
The questionnaire was separated into two portions, including demographic questions and
key questions relating to e-review, e-satisfaction, and online shopping intention. To increase the
reliability and validity of the questionnaire, a pilot test with 30 individuals who had experience
buying online was undertaken prior to the main study. Before it was officially distributed, a few
minor adjustments were made to the original questionnaire. The questionnaire was then
translated into Vietnamese so that responders could easily and appropriately complete it. The
questionnaire is adapted from previous studies. The survey questionnaire was prepared based on

2
validated and reliable measurement scales found in the literature. All items were measured on a
seven-point Likert scale ranging from (1) strongly disagree to (7) strongly agree. The following
table presents the measurement elements and sources, as shown in Table 4.1.
Table 3.1 Questionnaire structures
Constructs Items References
VA1. Please indicate the degree to which you would consider
a negative e-review about a product (containing some
negative information about its performance) to be relevant or
irrelevant for evaluating the online product.
VA2. Please rate the degree to which a negative e-review
about a product (containing some negative information about
(Stefanov, 2014)
its performance) is indicative of how good or bad the online
product is.
VA3. Please indicate the degree to which a negative e-review
about a product (containing some negative information about
its performance) would be useful to you for evaluating the
Valence online product.
(VA) VA4. Please indicate the degree to which you would consider
a positive e-review about a product (containing some positive
information about its performance) to be relevant or irrelevant
for evaluating the online product.
VA5. Please rate the degree to which a positive e-review
about a product (containing some positive information about
its performance) is indicative of how good or bad the online
product is.
VA6. Please indicate the degree to which a positive e-review
about a product (containing some positive information about
its performance) would be useful to you for evaluating the
online product.

3
3.5. Data analysis techniques
For data processing and analysis, the Partial Least Square (PLS) was used. The PLS
approach was appropriate because it made minimal demands concerning measurement scales,
sample size, and residual distributions (Monecke & Leisch, 2012). According to Hair et al.,
(2014), since the 2000s, the number of studies using PLS-SEM published increased by
increasing number of people. Especially in the fields of strategic management, information
system management, organizational behavior, and marketing research on satisfaction analysis
because PLS-SEM has advantages over CB-SEM in the following situations: ( 1) avoid problems
related to small sample size, non-standard non-delivery data; (2) it is possible to estimate
complex research models with many intermediate variables, underlying and observation
variables, especially structural models; (3) suitable for predictive orientation studies (Henseler et
al., 2014). This study applies PLS-SEM to identify the impact of e-review on customers’ online
shopping intention. The Likert scale with seven-point response options ranging from “(1)
Strongly Disagree” to “(7) Strongly Agree” was used to collect information from respondents on
how they perceive measurement items.
3.6. Assessing the outer measurement model
The evaluation of the outer model (measurement model) should be validated before
testing the hypotheses in the inner model (structural model) which examines the measurement
model's reliability (Cronbach's Alpha and composite reliability) and validity (convergent and
discriminant validity) is part of the evaluation process.
For the internal consistency reliability, a tool that measures the construct reliability,
Table 4.3 depicts that the Cronbach's Alpha values are all higher than the recommended
threshold value of 0.70 (Hair Jr et al., 2016). On the other hand, the results in the same table also
demonstrate that the composite reliability values are higher than the required value of 0.70 of
Hair Jr et al. (2016). Therefore, the internal consistency reliability in terms of Cronbach's Alpha
and composite reliability is verified and all constructs have significant reliability.
The term "convergent validity" refers to the assessment of many conceptually similar
items. In accordance with Hair Jr et al. (2016), the average variance extracted (AVE) is
recommended to assess the convergent validity, in which the value of AVE should be greater
than 0.50 before it is asserted to be validated. Indeed, Table 4.3 shows that all AVEs are

4
substantial and larger than 0.50. On the other aspect, it is also suggested by (Hair Jr et al., 2016)
that convergent validity could be confirmed by the value of outer loadings. Accordingly, if the
value of outer loadings is higher than 0.70, the convergent validity would be verified. Obviously,
the results in Table 4.4 indicate that all values are greater than 0.70. Consequently, the
convergent validity of this study has been confirmed.
Table 4.2 Overview of measurement model quality
Average Variance
Cronbach's Alpha Composite Reliability
Extracted (AVE)
VA 0.833 0.878 0.545
QN 0.659 0.801 0.512
QL 0.868 0.905 0.656
ES 0.875 0.923 0.800
OS 0.843 0.894 0.679

Table 4.3 Outer loadings of the measurement model


ES OS QL QN VA
ES1 0.900
ES2 0.902
ES3 0.881
OS1 0.836
OS2 0.760
OS3 0.827
OS4 0.870
QL1 0.804
QL2 0.856
QL3 0.760
QL4 0.854
QL5 0.771
QN1 0.762
QN2 0.824

5
QN3 0.756
QN4 0.465
VA1 0.714
VA2 0.745
VA3 0.718
VA4 0.729
VA5 0.754
VA6 0.770

Discriminant validity guarantees that the relevant items load highly on the construct in
consideration while loading poorly on other constructs. As a result, these objects are easily
distinguished from those of other constructions. To assess the discriminant validity, several
approaches were used, namely the Fornell-Larcker’s criterion (Fornell & Larcker, 1981), the
cross-loadings, and the "Heterotrait-Monotrait" ratio (HTMT) (Henseler et al., 2015).
To begin, according to the results in Table 4.5, the conventional Fornell-Larcker's
criterion was deployed and found that the square root of AVE is greater than the correlation
coefficients. Additionally, the cross-loadings are also examined and the results in Table 4.6
demonstrate that all loadings have a strong load to the respective constructs, whereas having a
weak load to irrelevant ones. Ultimately, the recently introduced HTMT criterion is also
checked, in which the HTMT ratio is suggested to be less than the threshold of 0.90 (Henseler et
al., 2015). In fact, Table 4.7 shows that all of the values met the criteria for being lower than
0.90. Given these results, the discriminant validity is ascertained.
Table 4.4 Fornell-Lacker’s criterion
ES OS QL QN VA
ES 0.894
OS 0.546 0.824
QL 0.562 0.569 0.810
QN 0.406 0.397 0.526 0.812
VA 0.486 0.566 0.519 0.388 0.739

6
Table 4.5 Cross-loadings
  ES OS QL QN VA
ES1 0.900 0.523 0.587 0.358 0.436
ES2 0.902 0.442 0.491 0.385 0.408
ES3 0.881 0.494 0.419 0.348 0.459
OS1 0.473 0.836 0.520 0.359 0.519
OS2 0.492 0.760 0.370 0.256 0.364
OS3 0.364 0.827 0.447 0.318 0.464
OS4 0.442 0.870 0.536 0.376 0.520
QL1 0.443 0.467 0.804 0.425 0.416
QL2 0.466 0.470 0.856 0.427 0.444
QL3 0.487 0.464 0.760 0.459 0.417
QL4 0.461 0.474 0.854 0.395 0.427
QL5 0.413 0.424 0.770 0.423 0.393
QN1 0.311 0.268 0.468 0.800 0.327
QN2 0.383 0.296 0.430 0.871 0.280
QN3 0.298 0.429 0.404 0.791 0.363
VA1 0.384 0.432 0.402 0.240 0.714
VA2 0.303 0.363 0.310 0.258 0.745
VA3 0.351 0.357 0.286 0.176 0.718
VA4 0.382 0.457 0.404 0.287 0.729
VA5 0.342 0.448 0.415 0.378 0.754
VA6 0.376 0.436 0.461 0.373 0.770

7
Table 4.6 HTMT ratio
  ES INT QL QN VA
ES          
INT 0.623        
QL 0.639 0.664      
QN 0.494 0.504 0.650    
VA 0.564 0.672 0.604 0.493  

The next step after the confirmation of the measurement model is the structural model.
This includes the measurement of the determination coefficient (R2) and the direction
coefficients of 1000 re-samples using a bootstrapping process. As set out in Table 4.8, the
evaluation of the structural model gives an indication of the testing of hypotheses. The results
showed that e-review valence has a positive effect on e-satisfaction. Thus, H1 is supported (β 1⁄4
0.070, p 1⁄4 0.000). Furthermore, the results pointed out that the quantity of e-review has an
insignificant negative effect on e-satisfaction. Hence, H2 is unsupported (β 1⁄4 0.074, p 1⁄4
0.125). In contrast, e-satisfaction is significantly influenced by the quality of e-reviews (β 1⁄4
0.083, p 1⁄4 0.000). Thus, H3 is supported. Moreover, the results revealed that e-satisfaction has
a significant positive effect on online shopping intention. Therefore, H4 is supported (β 1⁄4
0.057, p 1⁄4 0.000).

8
Table 4.7 Hypothesis testing results
Hypothesis Relationship Path Standard P Values Decision
Coefficients Deviation
H1 Valence (VA)  E- 0.248 0.070 0.000 Supported
satisfaction (ES)
H2 Quantity of e-review 0.113 0.074 0.125 Not
(QN)  E-satisfaction supported
(ES)
H3 Quality of e-review 0.374 0.083 0.000 Supported
(QL) E-satisfaction
(ES)
H4 E-satisfaction (ES)  0.546 0.057 0.000 Supported
Online shopping
intention (OS)

You might also like