Fig. 1: Task structure and avoidance rates.

A In the Elevator Task, a trial began with a screen that required fixation on the central arrows (2.5 s). After a jittered interval, the two doors opened and simultaneously one of the two arrows lit, indicating whether the elevators were going up or down (anticipating avatars’ movements after the choice). Participants had 1.5 s to freely choose which elevator to enter (by pressing the left or the right button). After the choice, in the Predictable condition (up arrow lit), participants were led next to the avatar in the chosen elevator during a period lasting 8.5 s (4 s moving inside the elevator, 4.5 s standing next to the avatar). In the Unpredictable condition (down arrow lit), after participants made their choice, the avatars in the elevators walked outside the environment, while two new different avatars (not previously visible) walked inside the elevators. Thus, participants were led next to a new avatar (50% probability of being angry or neutral). At the end of the trial, participants were “teleported” back to the starting position. The scene reset between two trials within 0.5 s. Note that the “Angry” and “Neutral” texts are for illustrative purposes only and were not present in the virtual environment. The picture was created as an illustrative example of the task: avatars’ faces were modeled from the Radboud face database⁴³, and the 3D body models were purchased from the RenderPeople website (https://renderpeople.com). Both sources permit the use of stimuli as examples. B Proportion of avoidance choices, P(Avoid Angry), by Condition (gray: Unpredictable; colored: Predictable). Box-plots represent first, second (median) and third quartiles. Whiskers are drawn within the 1.5 interquartile range. Points represent individual participants. C Proportion of avoidance responses by Condition (Unpredictable vs. Predictable) and experimental block. Vertical lines: within subjects 95% confidence interval. D Predicted probability of avoidance choices as a function of RTs (solid lines). Shaded areas: 95% credible intervals. Points: observed avoidance rates over median RTs for each participant. E Observed proportion of avoidance choices as a function of RTs within a 100 ms sliding window. Shaded areas: 95% bootstrapped Confidence Intervals. On the bottom: defective density distribution (i.e. weighted by the relative frequency of responses) of RTs for avoidance (solid) and approach (dotted) responses. In all plots, the black dashed line represents the chance level (50%). B–E - 1: Experiment 1 (predictable condition in blue, n = 60), 2: Experiment 2 (predictable in orange, n = 30), 3: Experiment 3 (predictable in red, n = 60). Unpred Unpredictable condition, Pred Predictable condition, RT Response Time.