• No results found

5 RESULTS

5.4 MULTIPLE REGRESSION

The first model is statistically significant (F(3, 348) = 11.695, ρ <0.000), with an R2 of .092. Meaning that the demographic variables Gender (β = -0.214, ρ .000), Electric car (β = 0.191, ρ .001) and Level of automation (β = 0.012, ρ .842) in this model explain only 0,09% of the variance. Gender and Electric car show to have a significant relationship with the dependent variable, while Level of automation does not. However, these relationships were not in the aim of this research and therefore these results will not be further discussed.

The second model includes the independent variables Drivers (Social Responsibility, Quality of Life and Social Connection) and Barriers (Accountability and Safety) was also found to be statistically significant F(8,242) = 36.623, ρ <0.001, with an R2 of .461. It means that the independent variables reliably predict the dependent variable Usage intentions. This model explains 46,1% of the variance in the Usage intentions, which is a change of 36,9%.

The first hypothesis states that motivational drivers positively influence the Usage intentions of autonomous vehicles and has been split into three sub hypotheses: Social Responsibility, Social Connections, and Quality of Life. The regression analysis shows that Social Responsibility (β = 0.202, ρ .000) is significantly related to the Usage intentions, and therefore evidence is found to support hypothesis 1.1. This means that when Social Responsibility increases with one unit, Usage intentions will increase with 0.202 unit. When it comes to Social Connections (β = 0.122, ρ .014) evidence is found to support hypothesis 1.2. The result is that if Social Connections increase with one unit, Usage intentions will increase with 0.122. With Quality of Life (β = 0.249, ρ .000) evidence is found to support hypothesis 1.3. It means that if Quality of Life increases with one unit, Usage intentions will increase with 0.249.

The second hypothesis states that motivational barriers negatively influence Usage intentions of autonomous vehicles and has been split into two sub hypotheses: Safety and Accountability. The regression analysis shows that Safety (β = -0.218, ρ .000) is significantly related to the Usage intentions, and therefore evidence is found to support hypothesis 2.1. It means that if Safety goes up by one unit, Usage intentions will go down by 0.218 unit. When it comes to Accountability (β = -0.074, ρ .088) there is no significant relationship and therefore no evidence is found to support hypothesis 2.2, which is therefore rejected.

The third model with the moderating variable Ownership is found to be statistically significant (F(9,342) = 34.117, ρ <0.000), with an R2 of .473. This model explains 47,3% of the variance, with an R2 change of 0,01%. This model will be used to determine the moderating effects of Ownership on Usage intentions.

Hypothesis 3 states that the type of Ownership positively influences the relationship of motivational drivers and barriers on Usage intentions of autonomous vehicles. The beta coefficient (β = 0,117, ρ <.005) shows that there is a significant moderating effect at the 1% level.

Therefore, evidence is found to support hypothesis 3 indicating that if Ownership goes up with one unit, Usage intentions increase with 0,117. In conclusion, it also means that if private ownership increases, so does the usage of autonomous vehicles. While this does not apply for shared autonomous vehicles.

The fourth model, with dependent variables, control variables and interaction variables, is found to be statistically significant (F(14,337) = 22.341, ρ <0.000). It means that the group of independent variables when used together reliably predict the dependent variable Usage intentions. With an R2 of .481 this interaction model explains 48,1% of the variance, which is 0,008% more than a model that contains only main effects. However, when examining the interaction terms individually, one can only conclude that Ownership has a significant impact on Social responsibility’s relationship with Usage intentions. The beta coefficient (β = 0.189, ρ < .001) shows that there is a significant moderating effect at the 1% level. It means that the relationship between Social responsibility and Usage intentions is increased by Ownership with 0.189.

5.4.1 ASSUMPTIONS OF MULTIPLE LINEAR REGRESSION

There are four assumptions when it comes to multiple linear regression. The first assumption is the linear relationship between the independent and dependent variables. The second assumption is a normal distribution of the residuals. The third assumption states that there should not be any multicollinearity present among the independent variables. The final assumption is that the data should be homoscedastic.

The results will be discussed in the subchapters below.

5.4.2 LINEAR RELATIONSHIP BETWEEN INDEPENDENT AND DEPENDENT VARIABLES

To determine the linear relationship between the dependent and independent variables a scatterplot is used (Appendix: Linear relationship between dependent and independent variables). All scatterplots include a best fitting line to provide a clear picture of the relationship. It can be concluded that for every independent variable there is a linear relationship with the dependent variable Usage intentions since they show a linear fit with the data points. This means that the assumption of linearity is met.

5.4.3 NORMALLY DISTRIBUTED RESIDUALS

To determine if the residuals are normally distributed, a scatterplot with the standard predicted value on the x-axis and standardized residuals on the y-axis is used to analyse if the errors are normally distributed.

If a normal distribution is the case, no patterns should be visible, and the dots should be scattered around the line. For the analysis a histogram and probability plot are presented. The histogram (figure 2.I) shows the data generally following the line shaped like a bell.

Figure 2 .I Histogram and II. Probability Plot of the Standardized Residuals

The probability plot (figure 2.II) shows the dots are broadly following the line as well. Therefore, based on figures 2 and 3 we can conclude the assumption of normality of the residuals to be met.

When checking for autocorrelations with the Durbin-Watson test, the value must be between 0 and 4. For the fourth model a value of 1,766 is detected which indicates a positive correlation.

Figure 3 Scatterplot of Standardized Residuals

5.4.4 NO MULTICOLLINEARITY AMONG INDEPENDENT VARIABLES

Two methods are being used to test for multicollinearity. Multicollinearity should not be present in the data, meaning that none of the independent variables can be strongly correlating (≥0.80) between themselves.

To test if the assumption is met, a Pearson Correlation matrix was previously presented in table 5. There are no values greater than 0.7, meaning that there is no multicollinearity among predictor valuables.

Secondly, the Variance Inflation Factors (VIF’s) are analysed and should not exceed 10 and the average should not be bigger than 1. Also, the tolerance values of each variable in the regression analysis are checked and should be greater than 0.2. All four regression models show that the VIFs are below 10 and an average of 1.217 for Model 1, 1.420 for model 2, 1.392 for model 3 and 1.556 for the final model. The tolerance level is greater than 0.2 for all models. It is concluded that multicollinearity will not be a problem

Table 7 The Variance Inflation Factors (VIF) and Tolerance Values

Variables Regression 1 Regression 2 Regression 3 Regression 4

VIF Tol. VIF Tol. VIF Tol. VIF Tol.

Gender 1,061 0,942 1,103 0,907 1,114 0,897 1,137 0,879

Electric car 1,260 0,793 1,296 0,771 1,298 0,771 1,306 0,765

Level of automation 1,328 0,753 1,349 0,741 1,350 0,741 1,359 0,736

D_SR_centered 1,646 0,607 1,650 0,606 1,667 0,600

D_QL_centered 1,666 0,600 1,682 0,595 1,702 0,588

D_SC_centered 1,554 0,644 1,555 0,643 1,579 0,633

B_S_centered 1,539 0,650 1,565 0,639 1,585 0,631

B_A_centered 1,204 0,831 1,214 0,823 1,229 0,814

O_centered 1,100 0,909 1,144 0,874

D_SRxO 2,090 0,478

D_QLxO 2,153 0,464

D_SCxO 2,027 0,493

B_SxO 1,469 0,681

B_AxO 1,341 0,746

Average VIF 1,217 1,420 1,392 1,556

N = 352

5.4.5 HOMOSCEDASTICITY

The fourth assumption is that the data should be homoscedastic, meaning that error terms are the same across every value of the independent values. When evaluating Cook’s Distance, providing an overall measure for the influence of individual cases, the maximum value is 0.065 which is well below the point of concern (1.0). When performing the Levene’s Test to test if the population variances of the two groups (male and female) are equal, that null hypothesis is rejected (p >0.05). This means that the equal variances are not assumed and there is a significant difference in the mean between males and females (t342.639 = 4,447, p < .001). The difference in Usage intentions for males was 0,35 more than for females (Appendix: Levene’s test and Independent samples T-test).

With the four assumptions of the hierarchical multiple linear regression were all found to be met, the conclusion is drawn regarding the results of the hypotheses being all valid.

5.4.6 SUMMARY OF HYPOTHESIS TESTING

Table 8 Overview of hypotheses and outcomes

Hypotheses Outcome

H1.1 Social responsibility positively influences Usage intentions of autonomous vehicles Accepted H1.2 Social connections positively influence Usage intentions of autonomous vehicles Accepted H1.3 Quality of life positively influences Usage intentions of autonomous vehicles Accepted H2.1 Safety negatively influences Usage intentions of autonomous vehicle Accepted H2.2 Accountability negatively influences Usage intentions of autonomous vehicles Rejected H3. Type of ownership positively influences the impact of motivational drivers and barriers

on the Usage intentions of autonomous vehicles

Accepted