Carl Westin

102 Automation transparency effects Aircraft A Aircraft B Preferred solution ( Participant 1 ) Conformal solution ( Participant 1 ) Nonconformal solution ( from Participant 2 ) F IGURE 5-4: Creation of conformal and nonconformal resolution advisory. The experiment simulation took place in week three. It took roughly 90 minutes and consisted of two sessions, one for each transparency condition, and a question- naire part. In this phase participants solved conflicts using both the HB and TRI SSD. Participants were divided into separate groups encountering the transparency conditions in different order, either HB-TRI or TRI-HB. Advisory conformance was, unknowingly to participants, varied within each session according to a Latin square design. Prior to each session, participants received instructions about the interface representation they were about to interact with. Each session comprised twenty minutes of training followed by five two-minute long scenarios. The five sce- narios consisted of one measurement scenario repeated twice, intertwined by three dummy scenarios. Unknowingly to participants, the scenarios and conflicts encountered in the ex- periment simulation were identical to those used in the prequel simulation. This time, however, participants were assisted by a decision aid that would suggest ad- visories by plotting them in the SSD. Advisories were accompanied by a beeping sound and a dialog window with an accept and reject button, an agreement scale, and a countdown timer starting at 30 seconds. Participants could either accept or reject the advisory, but first after indicating their agreement with the advisory. Participants were instructed that the decision aid would detect most conflicts and that resolution advisories suggested were generated by the decision aid and always would be safe although not necessarily efficient. In reality, resolution advisories were either con- formal or nonconformal and only provided for the scripted designed conflict. 5-6 Results Non-parametric tests were used because of the small sample size and ordinal vari- ables. All tests were conducted with a significance level of α = 0.05.