Randomised versus fixed trial order in repeated exposure VR studies of pedestrian interaction with automated vehicles

Alam, M. S., Bazilinskyy, P.

Working project. (2026)

ABSTRACT Repeated exposure experiments often use the trial position to study learning, fatigue, habituation, or carryover. Such analyses are difficult to interpret if the order of conditions is fixed, because later trials may contain a different mixture of conditions from earlier trials. We examine this issue in a virtual reality study of pedestrian interaction with automated vehicles. The same experiment was available in two runs: 50 participants completed the 40 experimental trials in participant specific randomised orders, and 50 participants completed the same trials in one shared fixed sequence. We asked where this ordering difference changed the conclusions. The average trigger based responses were highly similar between the groups. Mean continuous unsafety was 0.34 in the randomised order group and 0.34 in the fixed sequence group, and the fraction of time marked unsafe was 0.34 and 0.34, respectively. No behavioural or head yaw outcome differed after correction for multiple comparisons. The ordering difference mattered more for trial position based conclusions. In the fixed sequence group, the position of the test was related to yielding, eHMI status, visibility, and distance to the pedestrian. The mixed effects models showed a corrected ordering scheme by trial position interaction for Q3, the rating of understanding of vehicle intention ($eta = 0.33$, $q = 0.01$), while the corresponding interaction for fraction of time marked unsafe was weaker after correction ($eta = 0.04$, $q = 5.30 imes 10^{-2}$). These effects weakened when trial position was replaced by prior exposure to the relevant condition. The results show that a reused fixed sequence may preserve broad average responses while making apparent learning effects ambiguous. For repeated exposure VR studies, participant specific randomisation or stronger counterbalancing is especially important when the research question concerns learning, expectation, carryover, or other trial by trial changes.