CHAPTER 5 DATA ANALYSIS AND RESULTS
5.4. Testing hypotheses by applying PLS-SEM
Three equations were formed to describe the main structural coefficients for the research model:
BI = β1a CFCI + β1b CFCF + β2a PR + β2b PS + β7a CFCI × PR + β7b CFCI × PS + β8a
CFCF × PS + β8b CFCF × PS + ɛ1 (1)
PR = β3a CFCI + β4b CFCF + ɛ2 (2)
PS = β3b CFCI + β4a CFCF + ɛ3 (3)
Note: BI: Continuance intention to use mobile commerce; PR: Perceived risk;
PS: Perceived security; CFCI: CFC - Immediate; CFCF: CFC - Future.
5.4.1. Research model quality
We first assessed the proposed research model quality by using the coeefficient of determinant R2 values and Stone-Geisser Indicator (Q2) values. The continuance intention had a R2 value of 0.169, thus indicating the independent variables explained 16.9% variance of continuance intention. The continuance intention variable had a Q² value of 0.109 > 0, thus indicating the predictive relevance of independent variables.
Also, we conducted a bootstrapping procedure of 5000 sub-samples to further evaluate the investigated effects, and used Cohen’s Indicator (f2) (values of 0.35, 0.15, and 0.02 to signify large, medium, and small effects) to assess the effect sizes of the studied effects.
5.4.2. The direct effects
Table 5-15: The direct effect testing results
Paths β Hypotheses VIF Std. β Bootstrap t-values Conclusion CFCI UMC β1a H1a 1.47 0.16 [0.06; 0.25] 2.66** Support CFCF UMC β1b H1b 1.34 0.14 [0.04; 0.23] 2.41* Support PR UMC β2a H2a 1.49 -0.21 [-0.32; -0.10] 3.19** Support PS UMC β2b H2b 1.34 0.19 [0.06; 0.29] 2.65** Support CFCI PR β3a H3a 1.12 0.44 [0.34; 0.52] 8.04*** Support CFCI PS β3b H3b 1.12 0.17 [0.08; 0.26] 3.19** Support CFCF PS β4a H4a 1.12 0.38 [0.27; 0.47] 6.00*** Support CFCF PR β4b H4b 1.12 0.09 [-0.01; 0.19] 1.40ns Not support Note: CFCI: CFC - Immediate; CFCF: CFC – Future; PR: Perceived risk; PS:
Perceived security; UMC: Continuance intention to use mobile commerce
(Source: author’s calculation) The testing results supported the hypotheses of positive impacts of CFC- Immediate (β1a = 0.16, t = 2.66, p < 0.01) and CFC-Future (β1b = 0.14, t = 2.41, p <
0.05) on continuance intention to use mobile commerce. The negative impact of perceived risk (β2a = -0.21, t = 3.19, p < 0.01) and positive impact of perceived security (β2b = 0.19, t = 2.65, p < 0.01) on continuance intention to use mobile commerce were also confirmed. These results were necessary conditions for testing further moderating effects of CFC-Future and CFC-Immediate on these relationships. While the influences of CFC-Immediate on perceived risk (β3a = 0.44, t = 8.04, p < 0.001), the effect of CFC-Future on perceived security (β4a = 0.38, t = 6.00, p < 0.001) and the impact of CFC-Immediate on perceived security (β3b = 0.17, t = 3.19, p < 0.01) were significant, the effect of CFC-Future on perceived risk was not supported by data (β4b
= 0.09, t = 1.40, p > 0.05).
5.4.3. The moderating effects
The testing results showed support for two over four moderating hypotheses. In particular, CFC-Immediate weakens the relationship between perceived security and continuance intention to use mobile commerce (β6b = -0.16, t = 3.05, p < 0.01), CFC- Future strengthens the relationship between perceived security and continuance intention to use mobile commerce (β7a = 0.10, t = 2.10, p < 0.05) were two hypotheses supported by data. The hypothesis that CFC-Immediate strengthens the relationship between risk and continuance intention to use mobile commerce (β6a = 0.02, t = 0.44, p
> 0.05) and the hypothesis that CFC-Future weakens the relationship between perceived risk and continuance intention to use mobile commerce (β7b = -0.02, t = 0.32, p > 0.05) were not supported. The f2 values of the three significant relationships ranged from 0.02 to 0.03, pointing out that the strength of the studied relationships had small effect sizes.
Table 5-16: The moderating effect testing results
Paths β Hypotheses VIF Std. β Bootstrap t-values Conclusion CFCI*PR UMC β6a H6a 1.27 0.02 [-0.05; 0.1] 0.44ns Not support CFCI*PS UMC β6b H6b 1.40 -0.16 [-0.24; -0.07] 3.05** Support CFCF*PS UMC β7a H7a 1.23 0.10 [0.02; 0.18] 2.10* Support CFCF*PR UMC β7b H7b 1.32 -0.02 [-0.1; 0.07] 0.32ns Not support Note: CFCI: CFC - Immediate; CFCF: CFC – Future; PR: Perceived risk; PS:
Perceived security; UMC: Continuance intention to use mobile commerce.
(Source: author’s calculation) 5.4.4. Testing for asymmetric impact
We followed a guideline of five steps on testing a parameter difference in PLS proposed by Rodríguez-Entrena et al. (2018). Accordingly, the parameters of interest needed to be obtained (Step 1) and the difference between the parameter estimates of interest were calculated (Step 2). In step 3, the bootstrap estimates of the parameters needed to be obtained and calculated the parameter difference for every bootstrap sample. In step 4, we calculated the estimation of the variance of the estimated parameter difference. Finally, step 5 computed the determination of the empirical
quantiles of the bootstrapped parameter difference or CIs. Three CIs were calculated including standard CI, percentile CI, and basic CI. If CIs include zero, then a statistical difference between the two estimated parameters cannot be confirmed.
Table 5-17: Testing the relative importance of direct effects
Relationships Standard
CI*
Percentile CI*
Basic
CI* Conclusion CFCI PR > CFCF PR H5a [0.15;
0.55]
[0.14;
0.55]
[0.15;
0.55] Support CFCF PS > CFCI PS H5b [0.10;
0.41]
[0.10;
0.41]
[0.10;
0.41] Support Note: CFCI: CFC - Immediate; CFCF: CFC – Future; PR: Perceived risk; PS:
Perceived security; CI: confident interval;* p < 0.05.
(Source: author’s calculation) The testing results showed that the 95% CIs derived from the bootstrap procedure with 5000 sub-samples do not contain the zero with regard to the estimation, we conclude that all asymmetric impact hypotheses are supported by data. More specifically, as presented in Table 5-10, the testing results of hypothesis H5a indicated that standard CI, percentile CI and basic CI range from 0.15 to 0.55, 0.14 to 0.55 and 0.15 to 0.55, respectively (do not contain the value of 0). Thus, the hypothesis H5a was supported by the collected data. Similarly, the testing results of hypothesis H5b indicated that values of standard CI, percentile CI and basic CI range from 0.10 to 0.41 (also do not contain the value of 0). Therefore, the hypothesis H5b was also supported by data. In other words, CFC-Immediate better predicts risk perception than security perception while CFC-Future better predicts security perception than risk perception.
Also, CFC-Immediate better predicts risk perception than CFC-Future does while CFC-Future better predicts security perception than CFC-Immediate does.
Also, the testing results showed that four control variables (gender, job, income and age group) did not have significant effects on continuance intention to use mobile commerce. Finally, the we adopted a two-tailed BCA (bias-corrected and accelerated) bootstrapping procedure with 5000 sub-samples with significance level of 0.05 to
consolidate the significance of the direct, indirect and moderating effects. The results (see Table 5-11) showed that the significant relationships had the bootstrap confidence intervals not including the value of zero, indicated that the estimated results were solid.
5.4.5. Post-hoc analysis
Table 5-18: ANOVA analysis results Fact
ors Variables Sum of
Squares d
f Mean
Square F S
ig.
Job
Perceived risk
0.32 4 0.08 0
.17
0 .96 Perceived
Security 1.94 4 0.48 0
.48 0
.75 Continuance
intention 1.42 4 0.36 0
.14
0 .97
Inco me
Perceived risk 0.95 3 0.32 0
.65
0 .58 Perceived
Security 0.28 3 0.09 0
.09 0
.96 Continuance
intention 4.92 3 1.64 0
.66
0 .58
Age group
Perceived risk 0.14 3 0.05 0
.10 0
.96 Perceived
Security 2.64 3 0.88 0
.88
0 .45 Continuance
intention 2.19 3 0.73 0
.29
0 .83 (Source: author’s calculation) In order to bring more practical insights of whether or not we should apply differetiated marketing strategies for different segments of demographic variables such as job, income and age group, we adopted ANOVA precedure to test the differences of the perception about perceived risk, perceived security and continuance intention to use mobile commerce. The results, as summarized in Table 5-11, indicated that all the Sig. values were greater than 0.05, implying that we could not reject the null hypotheses. Thus, we concluded that there were no differences between job, income
and age group categories regarding perceived risk, perceived security and continuance intention to use mobile commerce.
5.4.6. The summarization of hypothesis testing results
The summarized hypothesis testing results are presented in Table 5-11. The results demonstrated that seven over eight direct effect hypotheses and two over four moderating hypotheses were supported by data.
More specifically, the impacts of CFC-Immediate (β1a = 0.16, t = 2.66, p < 0.01) and CFC-Future (β1b = 0.14, t = 2.41, p < 0.05) on continuance intention to use mobile commerce were supported. The negative impact of perceived risk (β2a = -0.21, t = 3.19, p < 0.01) and positive impact of perceived security (β2b = 0.19, t = 2.65, p < 0.01) on continuance intention to use mobile commerce were also confirmed. The influences of CFC-Immediate on perceived risk (β3a = 0.44, t = 8.04, p < 0.001), CFC-Future on perceived security (β4a = 0.38, t = 6.00, p < 0.001) and the impact of CFC-Immediate on perceived security (β3b = 0.17, t = 3.19, p < 0.01) were significant. The effect of CFC-Future on perceived risk was not supported by data (β4b = 0.09, t = 1.40, p >
0.05).
Regarding the proposed moderating effects, the testing results showed support for two over four moderating hypotheses. In particular, the moderating effect of CFC- Immediate on the relationship between perceived security and continuance intention to use mobile commerce (β6b = -0.16, t = 3.05, p < 0.01) and the moderating effect of CFC-Future on the relationship between perceived security and continuance intention to use mobile commerce (β7a = 0.10, t = 2.10, p < 0.05) were two hypotheses supported by data. The moderating of CFC-Immediate on the relationship between risk and continuance intention to use mobile commerce (β6a = 0.02, t = 0.44, p > 0.05) and the moderating effect of CFC-Future on the relationship between perceived risk and continuance intention to use mobile commerce (β7b = -0.02, t = 0.32, p > 0.05) were not supported
Table 5-19: Summary of path analysis testing results
Paths β Hypotheses VIF Std. β Bootstrap t-values Conclusion
Direct effects
CFCI UMC β1a H1a 1.47 0.16 [0.06; 0.25] 2.66** Support CFCF UMC β1b H1b 1.34 0.14 [0.04; 0.23] 2.41* Support PR UMC β2a H2a 1.49 -0.21 [-0.32; -0.1] 3.19** Support PS UMC β2b H2b 1.34 0.19 [0.06; 0.29] 2.65** Support CFCI PR β3a H3a 1.12 0.44 [0.34; 0.52] 8.04*** Support CFCI PS β3b H3b 1.12 0.17 [0.08; 0.26] 3.19** Support CFCF PS β4a H4a 1.12 0.38 [0.27; 0.47] 6.00*** Support CFCF PR β4b H4b 1.12 0.09 [-0.01; 0.19] 1.40ns Not support Moderating effects
CFCI*PR UMC β6a H7a 1.27 0.02 [-0.05; 0.1] 0.44ns Not support CFCI*PS UMC β6b H7b 1.40 -0.16 [-0.24; -0.07] 3.05** Support CFCF*PS UMC β7a H8a 1.23 0.10 [0.02; 0.18] 2.10* Support CFCF*PR UMC β7b H8b 1.32 -0.02 [-0.1; 0.07] 0.32ns Not support Control variables
Gender UMC 1.10 -0.03 [-0.1; 0.05] 0.60ns Not support
Job UMC 1.05 -0.01 [-0.08; 0.07] 0.13ns Not support
Income UMC 1.07 0.06 [-0.02; 0.12] 1.25ns Not support Age Group UMC 1.32 -0.01 [-0.09; 0.08] 0.17ns Not support
R2 (%) R2 BI = 16
Effect size (f2) f2CFCI BI = 0.02; f2CFCF BI = 0.02 f2PR BI = 0.03; f2PS BI = 0.03;
f2CFCI*PS BI = 0.03;
f2CFCF*PS BI = 0.02
Stone-Geisser’s Q² Q²BI = 10.5
Note: CFCI: CFC - Immediate; CFCF: CFC – Future; PR: Perceived risk; PS:
Perceived security; UMC: Continuance intention; *** p < 0.001; ** p < 0.01; * p <
0.05; ns: not significant