Page - 45 - in Critical Issues in Science, Technology and Society Studies - Conference Proceedings of the 17th STS Conference Graz 2018

relationships between variables that are assumed to be interrelated. These interrelations cannot be incorporated in standard regression models. A prerequisite for using such a model is a clear and theoretically founded idea about the interrelations of a certain matter (cf. Weiber and Mühlhaus 2014, p. 3). As described in the core model section, this clear idea was developed by using the established TAM and extending it based on previous works on trust. After modelling all possible interrelations, the model was gradually reduced by removing relations that displayed no significant effect. Figure 9 shows the remaining relations, color-coded according to significance levels. In the overall model we achieved good fit values. The model was tested by means of the Goodness-of-Fit-Index (GFI) and the Adjusted-Goodness-of-Fit- Index (AGFI). These indices are normalised to a [0;1] interval. Both values exceed the presumed benchmark of 0.9 for a good model fit (cf. Weiber and Mühlhaus 2014, 210ff). The GFI value for the model is 0.977. If we adjust for model complexity the AGFI value is 0.939. To further test the model fit we used the Comparative Fit Index (CFI), which has the same benchmark value of 0.9 (cf. ibid, p. 214f). The tested model achieves the perfect value of 1.0. Figure 9: Results of the SEM calculation Conclusion Summarizing the results of our study, we conclude that trust in app providers has the assumed significantly positive influence both on the TAM as well as on user’s users’ intention to change their behaviour. Therefore, it is an essential component of users’ decision making towards the acceptance of algorithmically generated recommendations. Concerning the constitution of mentioned trust, it turns out that two factors show a significant influence trust: Reputation shows a significantly positive influence on both forms of trust which 45