When Probability Meets Depth: Statistical Learning Reweights 3D Attention

Does selection history reshape attentional priority along the depth axis as it does in the image plane? Across three immersive virtual-reality visual-search experiments (total N=65), we tested whether observers implicitly learn depth-based regularities and exploit them to guide attention. Participants searched for an oriented target among distractors distributed across stereoscopically defined planes while target probability was manipulated between the depth planes. In Experiment 1 (N=37), with two eligible frequent target planes (near vs. far; 80/20 between subjects), responses were reliably faster when the target appeared on the frequent plane, regardless of whether it was near or far. In Experiment 2 (N=15), when all three planes were eligible and the target probability was concentrated at the fixation-aligned middle plane (10–80–10), no benefit emerged for the frequent plane. Experiment 3 (N=13) reinstated a between-subjects 80–10–10 probability distribution across all three planes: robust prioritization reappeared when the frequent plane was near or far, but again failed when it coincided with the middle plane. Together, the findings show that statistical learning can reweight plane-level priority in three dimensions, yet fixation depth constitutes a boundary condition where probabilistic biases do not accrue. A two-stage account is proposed consisting of plane-level gating followed by within-plane selection. This account reconciles the results with planar contextual-cueing findings and informs VR interface design.