Results

The following parts analyze the results of the survey and test the hypotheses made earlier. Specifically, the dependent, independent, moderating, mediating, and controlling variables are analyzed and discussed using IBM's program SPSS (IBM, 2022).

4.1 Data Cleaning Check

4.1.1 Descriptive Statistics

A total of 327 people filled in the questionnaire. In order to check for missing data, a frequency test was performed in SPSS (cf. Appendix B), which examined all included variables. Unfortunately, 26 questionnaires were filled in incompletely which is why they were excluded from the sample to ensure that any systematic biases are not involved (Kitchenham et al., 2003). Furthermore, 147 respondents answered the so-called trap questions wrong (Liu & Wronski, 2018). Moreover, since the target group is consumers in the automotive industry, all participants who do not have a driver's license had to be filtered out at the beginning, which amounted to a total of six participants. Furthermore, a test for so-called outliers, known as observations or cases of a variable in the survey sample was run on the whole data set (Ghosh & Vogt, 2012). Three outliers could be found in the variable Age and one in Awareness of Sustainability. As outliers can bias other statistics, such as means or standard deviations (Cheng, 2022), the outliers found were excluded from the data set as well. As a result, 146 respondents filled out the questionnaire completely and correctly and thus, the analysis was based on a sample size of 146 participants. This size is still sufficient to analyze the data and arrive at valid results (Memon et al., 2020). In general, the sample size is composed of 70 women and 76 men, with an average age of 34 years, the youngest being 18 years old and the oldest 77 years old (cf. table 2).

4.1.2 Dummy Coding and Normality Check

Primarily, the variables Gender, Own Car and Location in the data set were recoded into categorical, binary variables, so-called dummy variables, in order to indicate the absence or presence of something and to include these variables into regression analyses (Cohen &

Cohen, 1983; Myers et al., 2013). Here, regarding the new variable of Gender, male displays

the value 1 as before but female indicates the value 0 instead of 2. The same was done for the variable Location and Own Car: whereas the terms city and yes still have the value 1, the new value for rural area and no is 0.

Lastly, a normality check was conducted for all variables included to test whether the data are symmetrically distributed around the center of all scores, known as normal distribution (Das

& Imon, 2016). Consequently, Age has a skewness outside the thresholds (1.07), however, an above-average number of participants are between the ages of 20 and 30 which is why the distribution is leptokurtic that is characterized by a more pointed than the normal distribution (cf. Appendix C) (Cheng, 2022). Furthermore, in order to check for normality for the binary variables Own Car, Gender and Location a graphical representation in the form of a histogram was used. As could be derived from the histograms for Gender and Own Car the datas are distributed normally (cf. Appendix C). Regarding the variable Location a platykurtic distributions which is flatter than the normal distribution (cf. Appendix C) (Cheng, 2022) could be investigated. However, this is due to the fact that above average participants are living in cities (cf. Appendix C).

4.1.3 New Variables

New variables were formed as a function of the existing variables to test the hypothesis and to calculate the correlation. Primarily, in order to analyze the independent variable, a categorical variable was created including all four presented models with value 1 implying model 1, value 2 implying model 2, and so on. Secondly, a mean variable was created regarding Acceptance of Circular Value Proposition which includes all items from the Likert scale. Moreover, a sum variable was developed which sum up how many of the 12 statements from the nominal scale the respective participant affirms. Therefore, it was possible to measure whether the respondent had a greater awareness of sustainability, which

affirmative statements. Lastly, a mean variable was created regarding the moderator Trust containing all 12 items of the Likert scale.

4.1.4 Reliability

Finally, the reliability of the scales used were measured which allowed us to investigate the internal consistency of the scales used (Cronbach, 1951). Here, reliability tests were applied for the variables Trust, Awareness of Sustainability and Acceptance of Circular Value Proposition. As can be seen in Table 1, all three variables feature a Cronbach's Alpha greater than 0.7. The Cronbach's Alpha measures the internal consistency which is the degree to which items of a scale are homogenous (Cronbach, 1951). If the relation is lower than 0.7, the scale or some of its questions are not acceptable (Adadan & Savasci, 2012).

4.2 Statistical Analysis 4.2.1 Correlation

Using SPSS, a so-called correlation matrix after Pearson was conducted which calculates correlation coefficients for all combinations of the thesis variables (Table 2).

Furthermore, means and standard deviations can also be seen for each thesis variable. The variable Driver's License shows no results since all participants who reported not having a driver's license were excluded from the data set, as explained above. Hence, the answer as well as the result for this question is the same for every respondent in the data set. As mentioned earlier, it can be inferred from the correlation matrix that the average age of the respondents is 34, that more of them live in cities and tend to have a greater awareness of

sustainability and the environment, and that almost exactly the same number of men and women participated. Furthermore, just under two thirds of the participants stated that they owned their own car. Lastly, the variables were tested for multicollinearity (cf. Appendix D) however, all VIF values were slightly above 1 and thus, nothing conspicuous was found (Daoud, 2017).

4.2.2 Hypotheses

Primarily, in order to test Hypothesis 1, Friedman's ANOVA was conducted because the question measuring the independent variable in the theoretical model is a ranking question asking participants to rank the models according to their preference. As mentioned earlier, the four models are distinguished by varying degrees of use-orientedness, with Model 1 having the highest, Model 2 the second highest, and so on. As can be seen in Table 3, Model 1 has the highest average ranking value, Model 2 the second highest, Model 3 the third highest and Model 4 the lowest average ranking indicating that the consumers are likely to positively evaluate use-oriented PSS as Model 1 has the highest degree of use-orientedness and Model 4 the lowest. Therefore, Hypothesis 1 claiming that Model 1 is most likely to be accepted compared to Model 2, Model 3 and Model 4 is confirmed.

In addition, a linear regression analysis was conducted in order to test for significant differences between the four models. As can be derived from Table 4, Model 1 has the most significant positive relation with Consumer Evaluation (p<0.001, b=0.98, t=43.10). Thus, with Model 1 increasing by one unit, Consumer Evaluation increases by 0.98 units.

Furthermore, Model 2 (Table 5) has a significant positive relation with Consumer Evaluation, however, the effect is smaller compared with Model 1 (p=0.049, b=0.16, t=1.98). Therefore, when Model 2 increases by one unit, Consumer Evaluation increases by 0.16. In contrast, Model 3 (p<0.001, b=-0.40, t=-5.89) and Model 4 (p<0.001, b=-0.74, t=-9.11), have even a significant negative relation with Consumer Evaluation (Table 6,7). This means that when Model 3 or Model 4 increase by one unit, Consumer Evaluation decreases by 0.40 and 0.74 respectively. As a result, Model 1 is the most positively evaluated compared to Model 2 which was evaluated as the second best and Model 3 and 4 were even evaluated negatively.

Hence, a further confirmation of Hypothesis 1 can be seen suggesting that Model 1 is the most likely to be accepted compared to Model 2, Model 3 and Model 4.

Secondly, in order to test the mediating effect (Consumer Evaluation) and thus the second hypothesis, a regression analysis was conducted using the plug-in system PROCESS (v4.1) (Hayes, 2022). As can be derived from Table 8, Use-oriented PSS has a significant positive relationship with the Consumer Evaluation of Model 1 and 2 and a significant negative relationship with the Consumer Evaluation of Model 3 and 4, as already explained above.

However, the relationship between Acceptance of Circular Value Proposition and Consumer Evaluation is not significant with respect to every model. As can be seen in Table 9, Consumer Evaluation of Model 2 (p=0.043, b=1.23, t=2.05), Model 3 (p=0.029, b=1.37, t=2.21) and Model 4 (p=0.046, b=1.22, t=2.02) have a significant positive relationship with

Acceptance of Circular Value Proposition. In contrast, Consumer Evaluation of Model 1 (p=0.228, b=0.90, t=1.21) does have a non-significant positive relationship with Acceptance of Circular Value Proposition. Moreover, there is no significant indirect ([-1.23, 2.38];[-0.17, 0.41]; [-0.88, 0.36]; [-1.42, 0.67]) as well as direct effect ([-0.31, 1.42]) between Use-oriented PSS and Acceptance of Circular Value Proposition as the confidence intervals include zero (Table 10) (Mallinckrodt et al., 2006). Therefore, the second hypothesis is rejected.

Lastly, to test the three moderating effects (Awareness of Sustainability, Age, Trust) of the theoretical model (Hypotheses 2-5) a MANOVA in SPSS was conducted. In the process of testing the moderating effects Consumer Evaluation of each model built the dependent variables, Use-oriented PSS the predictor variable and Awareness of Sustainability, Trust and Age the moderators, respectively. Furthermore, the assumptions of a MANOVA were checked and consequently, the prerequisites of multivariate normality, homoscedasticity, linearity, and independence and randomness were met (Salkind, 2010).

Primarily, as can be derived from Table 11, Awareness of Sustainability has a non-significant effect on the relationship between Use-oriented PSS and Consumer Evaluation (Model 1:

p=0.671, ηp²=0.15, Model 2: p=0.155, ηp²=0.22, Model 3: p=0.069, ηp²=0.25, Model 4:

p=0.238, ηp²=0.20). Although Wilks' Lambda is significant (F(88,409)=3.24, p<0.001, ηp²=0.41, Wilk’s Λ=0.12) and hence implying that the model is efficient and that there is statistical difference between the four presented models (Patel & Bhavsar, 2013), Hypothesis 2 is rejected since the interaction effect is non-significant. In addition, in order to test for homogeneity of variance, Levene's test of equality of error variances was conducted, however, all p-values were above 0.05 and thus the model is efficient (Tovohery, 2020).

Secondly, Table 12 shows that Age has a non-significant effect on the relationship between Use-oriented PSS and Consumer Evaluation (Model 1: p=0.045, ηp2=0.41, Model 2:

p=0.336, ηp²=0.33, Model 3: p=0.016, ηp²=0.45, Model 4: p=0.790, ηp²=0.25) as well. The prerequisite of homogeneity of variance is met and Wilks' Lambda is significant

(F(124,277)=1.43, p=0.008, ηp²=0.39, Wilk’s Λ=0.14) which implies that the model is efficient and that there is statistical difference between the four presented models on the dependent variables (Patel & Bhavsar, 2013). However, Hypothesis 3 is rejected since the interaction effect is non-significant.

Lastly, Trust has a significant effect on the relationship between Use-oriented PSS and Consumer Evaluation (Table 13) with an interaction variable characterized by all p<0.001 except for Model 4 (p=0.004). Furthermore, the partial eta squared values are all above 0.14 (Model 1: ηp²=0.89, Model 2: ηp²=0.55, Model 3: ηp²=058, Model 4: ηp²=0.51) implying a great effect (Cohen, 2013). The prerequisite of homogeneity of variance is met and Wilks' Lambda is significant (F(108,207)=4.86, p<0.001, ηp2=0.72, Wilk’s Λ=0.02). Consequently, Hypothesis 5 is confirmed. As a result, Trust moderates the positive relationship between Use-oriented PSS and Consumer Evaluation such that this relationship is stronger for higher levels of Trust.

In sum, while hypotheses 2,3 and 4 were rejected, hypotheses 1 and 5 could be confirmed (Figure 2).

Figure 2: Theoretical Framework including Hypotheses Results, own work

Finally, a cross table in SPSS was conducted as the control variables Gender, Location and Income may have an impact on the variable Consumer Evaluation.

Primarily, Gender was introduced as a control variable. According to the data in Table 14, there is no big difference between females and males. Model 1 was chosen most in 1st place, Model 2 in 2nd place, Model 3 in 3rd place, and Model 4 most in 4th place. In the female's group, greater differences between the rankings can be seen in some cases with somewhat smaller gaps in the men's group. Men, in particular, rated models 3 (17%) and 4 (22%), and thus private car and ride sharing, slightly more positively than women (7%, 13%). In contrast,

generally more likely to perceive less risk in sharing services (Dittmar et al., 2004) and hence, gender differences exist in the perception of trust as well (Kyriakidis et al., 2015) as explained above. As a result, Gender does not have a large influence on Consumer Evaluation as the evaluation is pretty similar, however, a difference in the distances between the rankings can be investigated and thus also whether the evaluation is more evenly distributed or declining.

Secondly, regarding the control variable Location, both consumers living in a rural area as well as consumers living in a city ranked Model 1 mostly in 1st place, Model 2 in 2nd place and so on. However, as can be seen in Table 15 consumers living in a rural area rated Model 3 and Model 4 much more negatively than consumers living in a city and hence, appreciated

Model 1 (short-term car sharing) (67%) and Model 2 (long-term car sharing) (20%) much more. A reason for that might be that consumers living in rural areas who do not have a car are dependent on car sharing services to go on holiday or to travel short distances to the next big city as the connection to public transport is often rather worse than in cities (Murray, 1998). Consequently, Location does have an influence on Consumer Evaluation as although the rankings are similar the differences between the rankings is more even for consumers living in a city. In terms of consumers living in a rural area the ranking and hence the preferences of the consumers regarding each model presented is much clearer.

Lastly, Income does have an impact on Consumer Evaluation. As can be derived from Table 16 consumers earning less than €25,000, between €50,000 and €100,000 and between

€100,000 and €200,000 evaluated Model 1 most positively. In contrast, consumers earning more than €200,000 and between €25,000 and €50,000 ranked Model 2 and Model 3 in the first places. However, regarding consumers earning between 25,000€ and €50,000 the difference between Model 1 and Model 2 and Model 3 is very small and thus, these consumers evaluated Model 1, Model 2 and Model 3 rather similarly. Consequently, consumers earning more than €200,000 per year differ most from the rest of the groups as they are the only ones who rate Model 3 (private car sharing) most positively and who ranked Model 1 (short-term car sharing) in third place. A reason for that might be that consumers earning more than €200,000 usually possess a private car and through private car sharing they are able to rent out their car and earn money with it. Though, only four participants indicated to earn more than €200,000 and thus, this group is rather small. In sum, Income does have an influence on Consumer Evaluation but since the groups concerned tend to be rather small, this result is not very conclusive.

4.2.3 Additional Findings

As argued earlier, it might be interesting to examine the difference between the participants possessing an own car and those who do not as it might have an influence on the consumer evaluation of an use-oriented PSS including sharing services in terms of car sharing. For this purpose, a MANOVA in SPSS was conducted including descriptive statistics (Table 17). The output shows that customers possessing an own car evaluate Model 1 and Model 2 more negatively than consumers who do not have a private car. In contrast, Model 3 and Model 4 are evaluated slightly more positively from consumers owning a private car than consumers who do not. The reason for that might be that Model 3 and Model 4 illustrate car sharing services where private people can rent out their car or offer private rides. Hence, they can earn money with this type of car sharing and thus it might be more attractive to consumers owning a car. In general, the difference between these two groups is significant (p=0.04) and has a partial eta squared value of ηp²=0.07 implying that the effect size is moderate and that owning a private car explains a moderate part of the difference in the valuation of a utility-based PSS and thus in the acceptance of a circular value proposition between consumers who own a private car and customers who do not.

Secondly, a frequency test was conducted in order to check which are the most frequent reasons why consumers would participate in car sharing and which are the most selected reasons why not. As can be seen in Table 18 most consumers would participate in car sharing to travel more sustainably (64.4%), out of convenience (54.1%) and to reduce responsibilities (43.2%) as well as expenses (51.4%). Contrary, acceptance by society or cars as status symbols does not play a big role. In contrast, as can be derived from Table 19, most consumers would not participate in car sharing because the territorial coverage of car sharing services is sometimes not widespread (63%), because they have their own car (52.1%) and because it is hard for them to trust the offer regarding safety and quality (31.5%). On the

other hand, lacking knowledge about how it works or how to use smartphone apps as well as time does not seem a big problem for the consumers.