Carl Westin

7-3 Research challenges and limitations 163 Another challenge was how to display consistency and differences in consis- tency between individuals. Traditional presentations of data, such as boxplots and bar charts, do not satisfactory capture the individual and variations between individ- uals. Previous research has suggested that visual illustrations, such as pictures and films, may be the best way to present individual’s data. 237 In the end, a sunburst chart presentation was used because it provided 1) an overview of the consistent patterns found for a specific conflict, 2) the proportion of inconsistent controllers, and 3) how frequent a certain problem-solving style (see Chapter 6). There may, however, be other more suitable ways to present individuals data. 7-3-3 The effect of advisory timing Consistency was not defined in terms of when in time controllers solved the con- flict across repetitions. However, solutions were investigated to determine if they varied over time. Particularly for a crossing conflict, the heading deviation required to solve the conflict increases over time as the two aircraft approach. Although re- sults revealed considerable differences in when conflicts were solved, results did not suggest that solutions varied over time. Determining when to provide decision support may be one of the most challeng- ing design decisions. In ATC, the inappropriate timing of conflict alerts has been identified as a driver for resistance. 189 It was reasoned that a resolution advisory should be provided before the controller has determined a solution. Analogously, the greatest benefits of safety-net alarms and conflict warnings in autonomous cars have been achieved when made before the driver takes action (e.g., in terms of trust in system and driver breaking performance). 270–273 However, some researchers have argued that advice provided before a decision has been made increases the risk for automation bias and complacency. 48–50 The same researchers note that by measuring peoples’ reaction to an advisory after they have made a decision, the effect of the advisory on changing their solution can be determined. Indeed, there is a risk that controllers accepted advisories because it was convenient and effort saving. On the other hand, an advisory provided after the controller has decided, may confound their decision to either accept or reject. With this in mind, it would be interesting to investigate the effect of conformance on acceptance, when advisories are provided after controllers have decided on a solution. This can serve to validate a person’s conformal advisory. 7-3-4 Dependent measures of trust and safety Although the acceptance of an advisory is an indisputable measure of automation use, its dichotomous measure is a limitation. Therefore, an agreement rating was