4.2. Reliability and validity of the measurement scale
4.2.2. Exploratory factor analysis (EFA)
Since Cronbach alpha coefficients could only value the reliability of the measurement scales, it’s also required to assess the validity of the scale. The author uses EFA method to evaluate the validity of the scale, including convergent validity and discriminant validity.
EFA method belongs to the group of interdependence techniques, meaning that there are no dependent or independent variables; and it depends on the interrelationship between the variables. In addition to that, EFA is also a technique to downsize and summarize the data. The EFA method would help to identify the groups of variables necessary for the research problem and find out the interrelationship between variables. In the implementation of Exploratory Factor Analysis (EFA), researchers usually pay attention to some criteria as follows:
Sample size: Hair et al (2006) indicated that, to use the EFA method, the sample size should be at least 50, better 100 and the observation/items rate should be 5:1, meaning that 1 measurement variable would require at least 5 observations, at best 10:1. In this study, the sample size is 274 and 7.2 time bigger than the observed variables (38). So the sample size is eligible for the use of EFA method.
Kaiser – Meyer – Olkin (KMO) and Bertlett’s Test: KMO is an indicator used to evaluate the appropriateness of EFA method with the data. If the KMO indicator runs from .50 to 1, EFA would be applicable (Kaiser, 1974).
Furthermore, Bartlett's test of Sphericity is used to test the null hypothesis that the variables in the population correlation matrix are uncorrelated. If the observed significance level is < .05, it is small enough to reject the hypothesis and concluded that the strength of the relationship among variables is strong. It is a good idea to proceed a factor analysis for the data.
Kaiser criterion: this criterion would help to identify factors which are extracted from the scales. The less important factors would be eliminated.
The important factors would be maintained while we look at the Eigenvalues values, which represent the variations that could be explained by each factor.
Only factors whose Eigenvalues value ≥ 1 would be kept in the model (Anderson & Gerbing & 1988).
Variance explained criteria: The total variance explained should not be lower than 50% (Anderson & Gerbing, 1988).
To satisfy discriminant validity of the scale: the differences of factor loading values between variables in different factors should be ≥ .30 (Jabnoun et al., 2003).
Factor Loading: According to Hair et al. (1998), factor loading is an indicator ensuring practical significance of EFA method. If factor loading ≥ .30, it’s at minimum level; If factor loading ≥ .40, it is necessary and if it is ≥ .50, it’s considered practical significance. Furthermore, Hair et al. (1998) advised that if factor loading ≥ .30, the sample size should be at least 350. If the sample size is about 100, we should choose the factor loading ≥ .55. If the sample size is about 50, the factor loading should be ≥ .75. So, with 274 items of sample size in this study, if the observed variables having factor loading ≤ .50 would be dismissed.
For the unidimensionality scale like the measurement of shopping customer satisfaction in this study, the author would apply principal component analysis together with the Varimax rotation method to reduce the number of variables.
4.2.2.1. EFA implementation for independent variables
EFA results for independent variables showed in Table 4.4 indicated that KMO indicator is quite high: .924 > .50. Furthermore, Chi-Square of Barlett’s test reaches the value of 6733.641, at significant value p=0.000 (< .05), therefore the null hypothesis was rejected. It means that EFA method is very applicable in this study.
While evaluating the Kaiser criterion based on Eigenvalues and criterion of extracted variance, 6 factors have been extracted at the Eigenvalue 1.244 and the Total Variance Explained (TVE) value was 67.186%. The TVE value >50% also indicated that the EFA method is suitable.
Based on the results in Table of Rotated Component Matrix(a) (Appendix 3.1), 5 observation variables, namely IQ1, WD5, MA1, MA2, MA3 would be eliminated because the factor loading values are smaller than .50. Table 4.3 showed the details of 5 disqualified variables which are removed from the scales.
Table 4.3. Details of five unqualified observation variables Observation
variables (abbreviation)
Description
Factor loading
value IQ1 I believe the Website provides accurate
information to potential customers like me.
.478 WD5 The Website and all of its linked pages work well. .470 MA1 The general pricing of the website’s goods is
relatively low.
.491 MA2 This website has a bigger offering of lucky draw
and discount than similar websites.
.453
MA3 The product of other similar websites can be found at this site.
.420
After the elimination of unqualified variables, the author applied EFA method again with the remaining observation variables. The results (Table 4.5) show that KMO indicator = .917> .50 (qualified), Chi-square of Barlett’s Test reaches the value of 5960.108 with p = 0.000, meaning that the use of EFA analysis is very applicable.
The second-time EFA implementation also indicated that at the Eigenvalue value > 1, factor analysis also extracted 6 factors out of 29 remaining observation variables with the total variance extracted equal to 71.548 % (> 50%), thus reached qualification. It also indicated that 6 extracted factors including Information quality, Customer service, Website design, Security and Privacy, Deliver method, Merchandise attribute, can explain 71.548% variation of the data.
The author also applied the Cronbach alpha analysis again to check the reliability of the scale after eliminating the unqualified observation variables.
Results shown in Table 4.4 indicated that, after the elimination, the scale would be qualified in term of reliability.
Table 4.4. EFA results of independent variables
Items Components
1 2 3 4 5 6
WD1 .746
WD2 .767
WD3 .744
WD4 .684
WD6 .679
WD7 .748
WD8 .596
WD9 .741
WD10 .743
SAP1 .766
SAP2 .761
SAP3 .770
SAP4 .791
IQ2 .756
IQ3 .638
IQ4 .554
IQ5 .688
IQ6 .540
IQ7 .608
CS1 .811
CS2 .834
CS3 .805
DM1 .517
DM2 .690
DM3 .813
DM4 .752
MA4 .868
MA5 .842
MA6 .706
Extraction Method: Principal Component Analysis.
Rotation Method: Varimax with Kaiser Normalization.
a. Rotation converged in 6 iterations.
KMO .917
Bartlett's Test of Sphericity Approx. Chi-Square: 6733.641 Sig. 0.000
Eigenvalue 12.378 2.622 1.692 1.536 1.292 1.228
% of Variance (TVE: 71.548%)
19.192 12.459 12.099 9.716 9.200 8.882
Cronbach alpha coefficients
(After removing items) .922 .935 .851 .912 .854 .881
In short, based on the criteria of Exploratory Factor Analysis, results show that the scales of 6 independent variables of the study could reach reliability, discriminant validity and convergent validity. The final scale comprised of 29 observation variables and they were grouped into 6 factors equivalent with 6 independent variables proposed in the model: Website Design (WD), Security and Privacy (SAP), Information Quality (IQ), Customer Service (CS), Delivery method (DM), Merchandise attribute (MA).
4.2.2.2. EFA implementation for dependent variable
According to EFA results for dependent variable, namely Online Shopping Customer Satisfaction (OCS), showed in Table 4.5, EFA technique could only extract one factor with the total variance extracted of 79.748% > 50% at the Eigenvalue value of 3.190, together with KMO indicator of .820, Chi-square of Barlett’s Test got value of 791.335 at significant value p = 0.000 < .05, meaning that the use of EFA analysis is very applicable. And the scale of dependent variable, online shopping customer satisfaction, could satisfy the criteria of reliability, discriminant validity and convergent validity. It means that the scale is suitable for the study and could be used for the next steps of analysis.
Table 4.5. EFA results of dependent variable
Items Factor loading
OCS1 .917
OCS2 .895
OCS3 .885
OCS4 .874
Extraction Method: Principal Component Analysis.
KMO .820
Bartlett's Test of Sphericity Approx. Chi-Square: 791.335 Sig. 0.000
Total Variance Explained (%) 79.748
Eigenvalua 3.190
Cronbach alpha coefficient .914