Carl Westin

4-4 Results 75 32 and 71 seconds). This was thirteen seconds after the conflict warning provided by the safety net function. Based on this data, all participants received the reso- lution advisory 20 seconds after the scenario started in the experiment simulation. Advisories were intended to appear before participants made their own decision. This deliberate design choice was driven by two assumptions. First, it was essential that participants did not solve the conflict prior to the advisory, as it would have invalidated the resolution. Second, providing advice after participants have made a decision forces them to question their own decision. This study sought to shift the focus, and have participants questioning the advice received from the system before they had made their own decision. 4-4 Results Because of the small sample size it was decided to not use any inferential statistics. The analysis is therefore limited to descriptive statistics. 4-4-1 Simulation data Figure 4-4 provides descriptive data for the conformance and advisory source condi- tions on acceptance, agreement rating, response time, and difficulty rating. Overall, simulation data was inconclusive. Figure 4-4(a) shows that the acceptance rate was very high with only one out of twenty advisories rejected (95%, a conformal au- tomation advisory). This observation shows that neither conformance nor advisory source affected the acceptance of resolution advisories. Figure 4-4(b) shows that agreement varied little between source conditions. Participants agreed slightly less with conformal advisories than nonconformal advisories. Figure 4-4(c) shows that response time varied little across conditions. Across both advisory source, response time was slightly lower (i.e., faster) for the confor- mal advisories. The variability in response time was larger with the human source than the automated source. Figure 4-4(d) shows that difficulty variability was large across all conditions. The perceived scenario difficulty slightly decreased with the human nonconformal scenario as compared to the human conformal scenario. For the automated source, the data was reversed, with automation nonconformal sce- nario being perceived more difficult than the automation conformal scenario. Fi- nally, no separation losses were recorded which indicates that safety performance was unaffected by source and conformance.