5. Results
5.1 Study 1
35
36
the alpha would remain above .87. The variable loyalty appeared to be reliable ( = .93) and robust because if any of the items would have been deleted, the alpha remained above .90. Lastly, the control variable attitude appeared to be reliable ( = .79) and also robust as no items would have drastically changed the outcomes of the reliability test if deleted. The constructed variables were also checked on convergent and divergent validity with the use of a correlation matrix, shown under Appendix V. The convergent validity test was passed as all items have a correlation coefficient with the score of their own dimension greater than .40. The divergent validity test was passed as all items have a correlation coefficient with the score of their own dimension greater than those computed with other scores.
The composite measures were created next. For most variables this was done with the summated scales method. The weights for each item are therefore assumed to be equal across the variable. The variable price was grouped in 10 different equally sized bins. Reason for this is the high standard deviation of the variable (SD = 311.71). The higher values are therefore taken together which makes the evaluation easier and paint a clear picture of the moderating effect of price.
5.1.3 Descriptive statistics
Table 6 presents the correlation matrix with Pearson correlation coefficients of the relevant variables with their mean and standard deviation. First of all, it can be observed that in this sample more people engaged in freeriding than people who have not (M = .63). Also, the mobile device is used more often for searching than fixed devices (M = .47). For practical reasons, the variable price shows the
continuous variable rather than the decile formed variable. Therefore, it can be observed that the average purchase price is relatively high (M = 221.90), indicating that people especially remember the customer journeys of more expensive products. However, the standard deviation is also rather large (SD = 311.71), as the smallest amount is 2 euros and the largest is 1650 euros. The mean value of the purchase date indicates that the average time that has passed since the purchase is around 1 to 6 months ago (M = 2.05).
The correlation statistics shows that the main independent variable device and the dependent variable freeriding are positively correlated but not at a significant level (r = .16, p > .10). It provides the first indication that cross-channel free riding and fixed device usage are positively related. Loyalty shows a relatively large negative correlation with freeriding (r = -.29, p < .01) indicating loyal
customers free ride less on average. Price is positively correlated with free riding behaviour (r = .29, p
< .01), while interestingly it shows that women used mobile devices more often for searching than men (r = -.25, p < .01). More surprising is that the attitude towards online searching and customer loyalty is positively related (r = .46, p < .01). It may indicate that loyal customers have more experience in online searching for that particular store and as a result are savvier when it comes to online search in general. Additionally, loyal customers bought more utilitarian goods (r = -.22, p <
.01) and were more often female (r = .25, p < .01).
37
Table 6. Pairwise Pearson correlation matrix of the relevant variables of study 1.
5.1.4 Preliminary results
As preliminary analysis to review the possible relation between the type of device and cross-channel free riding intentions stated in hypothesis 1, the mean values between the different devices are compared. The mean value for cross-channel free riding for the group that uses a mobile device is .56 while it is .72 for the group that uses a fixed device. This difference in means is visualized by figure 2 and it can be observed that this confirms prior assumptions that cross-channel free riding intentions are stronger for the group that uses a fixed device for search purposes. To check whether this difference is statistically significant a Mann-Whitney test was performed as the dependent variable freeriding is binary. The test returned a statistically significant difference in the mean value between the different devices (U = 3388.50, z = -2.211, p < .05).
Figure 2. Difference in means of cross-channel free riding intentions between devices in study 1.
Variable Mean SD Freeriding Device Loyalty Type Price Attitude Age Gender Education Date Category Freeriding .63 .48 -
Device .47 .50 .14 -
Loyalty 3.63 .95 -.29** -.08 (.93)
Type .34 .48 .35** .05 -.33** (.90)
Price 221.91 311.71 .23** .14* .03 -.16** -
Attitude 3.79 .80 -.09 -.19* .46** -.34** 0.00 (.79)
Age 2.36 1.40 .13 -.05 -.05 .02 .07 -.14 -
Gender 0.50 0.50 -.01 -.25** .25** -.05 -.04 .19* .02 -
Education 3.11 1.21 .08 .13 .02 .09 .06 .13 -.22** .01 -
Date 2.05 .77 .23** .11 -.04 .10 .36** -.02 -.05 .05 .03 -
Category 2.72 1.78 .10 .19* -.08 .11 -.05 -.04 .21** -.09 .02 -.06 -
**p < .01, * p < .05
N = 183 and Cronbach alphas in parentheses and bold.
38 5.1.5 Regression analysis
A regression analysis was used to further examine the proposed hypotheses. Because of the nature of the dependent variable freeriding in this study (binary variable) a logistic regression model was used.
Table 7 presents the output of the hierarchical regression estimation. Specification 1 presents the regression model of the dependent variable freeriding on the control variables. The model is statistically significant (2 (6) = 16.92, p < .01) and highlights that some demographic and product specific characteristics affect free riding intentions as age and date significantly influenced the dependent variable freeriding. The R2 is not obtained by logistic regression as the variance is fixed due to the variance of the standard logistic distribution. The pseudo R2 is provided for logistic regressions and measures the proportion of change in terms of likelihood, but can be interpreted largely in a similar way as the regular R2. In model 1 the pseudo R2 equals 8%.
In model 2, the main independent variable and moderator variables without interactions were added. The model is statistically significant (2 (10) = 44.36, p < .01) and the pseudo R2 now equals 27%, which is a significant increase over model 1 (2 (4) = 45.06, p < .01). The main independent variable device is statistically significant and provides further evidence in favour of hypothesis 1 ( = .82, p < .05). The coefficient can be interpreted as the odds of engaging in cross-channel free riding behaviour when you are using a fixed device is 2.27 times higher than when you use a mobile device on average. However, the standard deviation of the coefficient is still considerably large as the potential odds ratios in the 95% confidence interval range from 1.06 to 4.85 and should therefore be reviewed with caution. The coefficients of the moderators show relations with the dependent variable that were expected by the prior literature and intuition. Customer loyalty seems to significantly decrease the level of cross-channel free riding intentions ( = -1.01, p < .01), while hedonic ( = .90, p < .01) and more expensive products ( = .25, p < .01) increase such intentions.
In model 3, the interaction effects of the moderator variables and the main independent variable device were included. First the continuous independent variables loyalty, type and price were standardized to prevent multicollinearity, which can cause large problems in logistic regression models (Senaviratna & Cooray, 2019). Consequently, the interpretation of the variables is different compared to the previous models. The pseudo R2 equals 32% and this is a statistically significant increase from model 2 (2 (3) = 12.88, p < .01). The model shows strong evidence in favour of hypothesis 1 as device is significantly different from 0 at a 1% level ( = 1.36, p < .01). The
interaction between device and loyalty resulted in a significantly negative coefficient ( = -1.98, p <
.01). It provides evidence in favour of hypothesis 2, as the positive relation between device and freeriding disappears if the customer is more loyal to a specific brand or store. The interaction of device with type did not result in a statistically significant coefficient ( = .10, p > .10) and neither the interaction of device with price ( = .29, p > .10). Consequently, there is no evidence in favour of the relations drawn in hypotheses 3 and 4.
39
Table 7. Hierarchical logistic regression with main independent variable freeriding.
Dep. Variable:
Freeriding (1) (2) (3)
Device .82** 1.36***
(2.1) (2.62)
Device * Loyalty -1.98***
(-3.16)
Device * Type .10
(0.20)
Device * Price .29
(.65)
Loyalty -1.01*** -.33
(-3.68) (-1.14)
Type .90*** .93***
(3.83) (2.64)
Price .25*** .64**
(3.1) (2.14)
Attitude -.22 .61** .89***
(-1.13) (2.17) (2.82)
Age .25** .23 .24
(1.99) (1.64) (1.56)
Gender -.04 .51 .22
(-.11) (1.3) (.53)
Education .26* .12 .05
(1.79) (.71) (.28)
Date .69*** .25 .16
(3.25) (.96) (.56)
Category .07 .06 .08
(.71) (.52) (.69)
Constant -1.52 -3.84** -4.44***
(-1.32) (-2.23) (-2.79)
Observations 183 183 183
Pseudo R2 .08 .27 .32
Wald χ² 16.92*** 44.36*** 46.75***
t-values are in parentheses
*** p < .01, ** p < .05, * p < .10
Loyalty, type and price are standardized in model 3
5.1.6 Assumptions and robustness checks
Before any conclusion could be drawn, the constructed model was first examined for its assumptions while also several robustness checks were executed. The main assumption in logistic regression is the nonexistence of multicollinearity. The continuous independent variables of model 3 in table 7 were already standardized to account for this. Nevertheless, all models were checked for multicollinearity by computing the variance inflation factor (VIF) of each variable. The table under appendix VI shows the results of this analysis and no multicollinearity was detected as the boundary level of 5 was not exceeded (Bowerman & O'Connell, 1990). The fit of the main model 3 was measured by a Pearson chi-squared goodness of fit test and showed that the observed and expected proportions did not differ
40
significantly which indicated a good fit (2 (168) = 176.89, p < .10). The performance of the model to correctly classify outcomes was also estimated by examining the value under the Receiver Operating Characteristic (ROC) curve. This returned value was 0.8470 which indicated acceptable
discrimination for the model (Mandrekar, 2010).
The robustness of the model was strengthened by altering two independent variables which could have been constructed differently to see if the results remained similar. The variable type was constructed as continuous variable where the utilitarian dimension was reverse coded. Another possibility would have been to create a binary variable equal to 0 if the utilitarian scale was larger and 1 if the hedonic scale was larger. Model 4 under Appendix VII shows the same regression as model 3 of table 7 but now with the variable type_2 being this binary variable. The main independent variable device lost some significance but the same conclusions with respect to hypothesis 1 and 2 could still be drawn. Interestingly, the new variable type_2 had no significant direct effect on cross-channel free riding intentions ( = .69, p > .10), while in model 3 hedonic products significantly increased cross-channel free riding intentions ( = .93, p < .01). The pseudo R2 dropped from 32% to 27%, but the model is still highly significant (2 (13) = 63.71, p < .01).
Model 5 shows the specification where the variable for the product price is not split in 10 equally sized bins, but the continuous value in euros is used. The same relations with regard to the hypotheses are drawn compared to model 3, indicating that the main model is robust to different constructions of the variable price. Model 6 combines the two differently created variables in one model and again the same conclusions are drawn with respect to the hypotheses but the pseudo R2 drops to 24%. This confirms that the main model used in table 7 is the most accurate and also robust against different variations of the same variable.