Carl Westin

4-5 Discussion 81 the pedigree of the participants (see Wærn and Ramberg 164 for an exception). Most studies have relied on novices in relation to the task, the decision aid used, and con- text conditioned in the experiment. In contrast, this papers used a sample of experts (i.e., experienced controllers). As a novice user it is reasonable to assume that a decision aid is perceived as an expert. An expert user, however, may perceive the decision aid differently and less as an expert. For example, several ATC studies have highlighted acceptance issues in context of CD&R decision aids. 8, 12 This can possibly explain why controllers indicated a trust preference for the human source. Future research should increasingly investigate trust and acceptance in relation to expert users and their work environments. 4-5-3 Limitations Since trust was not measured before interaction with the decision aid, there were no data to compare post-simulation measurements with. In line with previous research, any dispositional attitudes should have been manifested early in simulations, and it is possible that measures in fact reflect such a priori attitudes. There were no indi- cations that acceptance or agreement changed during the simulation. Furthermore, the reliability (i.e., accuracy) of advisories were not varied. All advisories were safe in that they solved the conflict, albeit not necessarily in terms of participant pref- erences (i.e., the conformance manipulation). As such, participants may not have had enough reason to reject advisories and therefore accepted more or less all. This does, however, go against previous results, in which controllers did reject noncon- formal advisories to a larger extent than conformal advisories. 101 Finally, the small sample size was a limiting factor in that very large differences were required in order to detect a difference between conditions. The simulation was rather short, and although participants received training be- fore each source condition, it may have been insufficient and overly subtle in terms of generating an attitude of trust and acceptance towards the system. Participants may have equated both aids to be similar until proven otherwise. Descriptive infor- mation about each source was intentionally left out in an effort to avoid confounding factors introduced by uncontrollably influencing the degree of pedigree or expertise in each source. However, because of the limited information participants may not have contemplated much about the credibility or performance of each source. More background information may have made the presence of different sources more salient and realistic. Although information about the functionality of the heading band SSD was provided, which was necessary in order to use it, the SSD was used as the mediating interface for both the automated and human source.