Chapter 4. Empirical Results
4.2. Reliability and Validity
4.2.1. Sample with participants which had used UBER before
The validity and reliability of the sample which had used UBER before is examined
firstly. This research adopted Exploratory factor analysis (EFA) to eliminate low loadings,
cross-loadings, or items loaded on the wrong factors. To test the suitability of the EFA
analysis, Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy was adopted for
assessment beforehand. While KMO index is considered to be greater than 0.50 for factor
analysis (Hair, Anderson, Tatham, & Black, 1998), the KMO index of this study is 0.893
and is significant (p<0.001), which is regarded to be suitable. The EFA result with
varimax rotation is listed below in table 4,
Table 4. EFA and Cumulative Percentage of Variance Explained before taking average on sample which participants had used UBER before
Component
UT10 .583 Extraction method: Principal component analysis; rotation method: Varimax with Kaiser normalization; IQ: Information Quality, UT: Trust on UBER, DT: Trust on Drivers, INT: Intention.
Based on the result of EFA, there are few issues to discuss. One item from the scale
of participants trust on UBER (UT11) was dropped due to wrong factor loading
(perceived trust on drivers: 0.608). Besides, the number of factors extracted should also
be discussed. The construct “trust” in this study contains 3 elements: benevolence,
integrity, and competence, and trust on UBER and drivers are measured respectively.
Hence, the number of factors should theoretically be 8 (Information Quality, perceived
benevolence, integrity, and competence respectively on UBER and drivers, and
participating intention). However, the factors of perceived benevolence and integrity on
benevolence and integrity on drivers. Therefore, there are finally 29 items only loaded on
6 factors, which is shown in table 4. Further, while benevolence, integrity, and
competence are formed to be trust, this study took averages respectively by the perception
of benevolence, integrity, and competence on UBER and drivers to became trust on
UBER and drivers, and did another EFA analysis below,
Table 5. EFA and Cumulative Percentage of Variance Explained after taking average on sample which participants had used UBER before
元件
% variance explained 23.751 19.543 18.544 12.239
Cumulative % variance explained 23/751 43.294 61.838 74.077 Extraction method: Principal component analysis; rotation method: Varimax with Kaiser normalization; IQ: Information Quality, UTB: Trust on UBER_Benevolence, UTI: Trust on UBER_Integrity, UTC: Trust on UBER_Competence, DTB: Trust on Drivers_Benevolence, DTI: Trust on Drivers _Integrity, UTC: Trust on Drivers _Competence, INT: Intention.
Unfortunately, the result of table 5 reveals few concerns toward the data’s validity.
Theoretically, the perception on UBER’s benevolence, integrity, and competence should
be categorized to the factor which represent people’s trust on UBER, and so is the
elements on people’s trust on drivers. However, through the EFA result, people’s
perception on UBER’s benevolence, integrity and people’s perception on drivers’
benevolence, integrity are regarded as one factor. On the other hand, the perception on UBER’s competence and the perception on drivers’ competence are regarded as another
factor. The result threatens this research’s construct validity, which includes convergent
validity and discriminant validity. As the perspective of convergent validity, the trust on
UBER do not contain benevolence, integrity, and competence as one construct, and the
trust on drivers have the same concern. And as the perspective of discriminant validity,
theoretically trust on UBER and trust on drivers should be separated. However, the
benevolence and integrity of UBER are related with the benevolence and integrity of
drivers, and UBER’s competence are related with drivers’ competence. Therefore, due to
the concerns on the convergent validity and discriminant validity, this research conducted
additional analysis for further examination on validity.
Confirmatory factor analysis (CFA) is performed for further evaluation. For
discussion of construct validity, table 6 shows the factor loadings, AVEs, CRs as
assessments of convergent validity, and displays Cronbach’s α as an index of reliability.
Besides, table 7 shows the square root the AVEs and the correlation of between each
constructs to discuss discriminant validity. The tables are shown below,
Table 6. Reliability and Validity: Standardized Factor Loadings for the Construct Indexes, Cronbach’s α, Average Variance Extracted, and Construct Reliability of sample
which participants had used UBER before Latent Construct Indicator Standardized
Loading AVE CR Cronbach’s UBER_Integrity, UTC: Trust on UBER_Competence, DTB: Trust on Drivers_Benevolence, DTI: Trust on Drivers _Integrity, UTC: Trust on Drivers _Competence, INT: Intention.
Table 7. Discriminant Validity: The Square Root of AVEs of sample which participants had used UBER before
1 2 3 4
1. Information Quality 0.729 2. Trust on the sharing economy
platform
0.436 0.713
3. Trust on the sharing peer 0.314 0.625 0.784
Note: The diagonal numbers are square root of AVE.
Factor loadings, AVEs, CRs in table 6 will be assessed to examine convergent
validity. Factor loadings are suggested to be greater than .40 (Hair et al., 1998), and the
result reveals that the standardized loadings of all items exceed the threshold. Besides,
the AVEs and CRs of all the items are above the recommended cut-off level respectively,
which AVE is suggested to be more than 0.5 and CR should be more than 0.7 (Fornell &
Larcker, 1981). Further, the Cronbach’s α value of the items also exceed the satisfactory level which is above 0.7, which means that the items have good reliability.
Therefore, convergent validity is considered to be acceptable when regarding factor
loadings, AVEs, and CRs as its reflections, that the indexes show the constructs, especially
trust, can still be formed by their original sub items, like benevolence, integrity, and
competence.
After that, discriminant validity is assessed by comparing the root square of AVEs
of each constructs and their correlation coefficient between other constructs. As
illustrated in table 7, it is found that the diagonal numbers which representing the root
square of AVEs of all the constructs are higher than off-diagonal values which means the
correlation coefficients. The result is considered reaching the satisfactory level of
discriminant validity (Fornell & Larcker, 1981).
According to the result revealed by EFA and CFA analysis above, this study would
like to discuss the construct validity again, especially focus on trust on UBER and trust
on drivers. Based on the analysis of EFA, this research admits that the convergent validity
and discriminant validity would be questioned because benevolence, integrity, and
competence cannot be formed as one factor that represent trust, and trust on UBER and
trust on drivers have components that are categorized as same factors. However, the
indexes through CFA not only suggest convergent validity of the measurements is
acceptable, but also indicated the discriminant validity of all construct are suitable.
Therefore, collecting the perspective from different analysis, this research suggests that
though the issues of construct validity existed, but is still tolerable.
Another issue is discovered through table 7 that the coefficient correlation between
trust on UBER and trust on drivers is 0.625. The number is high and reveals that
collinearity may existed between these two factors. Thus, this research conducted
Variance Inflation Factor (vif) examination below to test whether the factors have
collinearity effect,
Table 8. Vif Table of sample which participants had used UBER before
IQ_average: Information Quality, DT_average: Trust on Drivers, UT_average: Trust on
Construct Collinearity Statistics
Tolerance VIF
IQ_average .807 1.239
DT_average .608 1.646
UT_average .546 1.831
UBER
Constructs will be considered having collinearity effect when vif value is greater
than 10 (Cohen, West, & Aiken, 2014). Table 8 shows that all the vif value of the
constructs are much less than 10. Thus, even though trust on UBER and drivers have high
coefficient correlation in table 7, the vif result eliminate the concern of their collinearity
to a certain extent.