The study, led by Canling An of the Doctoral Program in Psychology at the University of Tsukuba together with Soichiro Matsuda of the university’s Institute of Human Sciences, tackled a question that sits at the intersection of two established research traditions. The first concerns value-driven attention: a large body of work has shown that stimuli previously associated with reward continue to capture attention even when they are no longer relevant to the task at hand. Faces associated with social reward, monetary outcomes, and emotional expressions can all bias visual selection in lasting ways. The second tradition concerns interpersonal distance, a fundamental dimension of social perception. Approaching stimuli are oriented to faster, and stimuli in near, peripersonal space measurably alter neural markers of attentional engagement. What had remained unclear was whether repeated exposure to distance-related social outcomes—someone seeming to come closer or move away in response to being looked at—would leave a measurable trace in later patterns of gaze allocation.

To answer that question, the researchers designed an experiment that carefully disentangles the initiating role of gaze from the ongoing experience of watching a stimulus change. Fifty adults, all students at the University of Tsukuba with normal or corrected-to-normal vision, completed two experimental blocks in a within-subjects design: one Approach block and one Recede block. Each block contained a Baseline phase, a Training phase, and a Test phase. In the Baseline and Test phases, participants simply viewed two static neutral female faces drawn from the AIST Facial Expression Database, each measuring roughly 9.75 degrees of visual angle and displayed on either side of, or above and below, a central fixation point. The dependent variable was the proportion of total dwell time directed at whichever face had been designated the gaze-triggered face in that block.

The Training phase was where the manipulation lived. Participants viewed a single face on each trial, and if they fixated on the gaze-triggered face for 300 milliseconds, a two-second animation was launched. Crucially, once triggered, the animation proceeded independently of any further gaze behavior. In the Approach condition, the face grew to approximately 635 by 635 pixels—a simulated advance toward the observer, from an estimated apparent distance of roughly 110–130 centimeters down to about 70–85 centimeters. In the Recede condition, the face shrank to roughly 152 by 152 pixels, corresponding to an apparent withdrawal to around 300–360 centimeters. The static comparison face remained unchanged for the same duration. Because the animation ran on its own after the 300-millisecond trigger threshold was met, the design avoided the continuous gaze–stimulus coupling of classic gaze-contingent paradigms and instead captured a discrete, everyday feature of social interaction: sustained gaze can mark a transition point after which interpersonal events unfold without depending moment to moment on where the eyes go next.

The technical execution reflected considerable methodological care. Eye movements were recorded with a Tobii X3-120 screen-based tracker sampling at 120 Hz, with calibration accuracy required to be below one degree of visual angle before testing. Each block consisted of four baseline trials, forty-eight training trials—twenty-four with the gaze-triggered face and twenty-four with the static face in randomized order—and four test trials. Baseline and test trial counts were deliberately kept low to limit fatigue and minimize additional exposure that might itself alter gaze patterns. Participants received no verbal instructions and were never told about the contingencies linking their gaze to the animations. Condition order and the assignment of face identities to the gaze-triggered role were counterbalanced across participants. After the experiment, participants completed the Japanese version of the Autism-Spectrum Quotient, whose scores served as an individual-difference measure given prior evidence linking autistic traits to variation in gaze behavior.

The analysis centered on dwell-time proportion—the fraction of dwell time on the gaze-triggered face relative to the total dwell time on both faces—averaged at the participant level for each Condition-by-Phase combination and entered into a linear mixed-effects model fitted with the lme4 package in R. Fixed effects included condition, phase, and their interaction, along with centered AQ scores and their squared terms, with gender and condition order as exploratory covariates. Participant identity was modeled as a random effect with random intercepts and by-condition slopes, and the model was fitted by restricted maximum likelihood. Fifty participants remained in the final sample after two were excluded for insufficient valid trials, and the overall valid trial rate was 99.4 percent. At baseline, dwell-time proportion sat close to chance in both conditions—0.52 in the Approach condition and 0.51 in the Recede condition—confirming that participants began with no inherent preference for either face.

The results hinged on a significant Condition-by-Phase interaction, β = 0.07, with a 95 percent confidence interval from 0.03 to 0.12 and p = .002. A main effect of Phase was also present, indicating overall lower dwell-time proportions at test. The pairwise comparisons told the more interesting story: in the Approach condition, dwell-time proportion significantly decreased from baseline to test, falling from a mean of 0.52 to 0.48, a change of roughly five percentage points that reached significance with a Bonferroni-corrected p value of .013. In the Recede condition, the numerical shift went the other way, rising from 0.51 to 0.54, but this increase did not reach statistical significance. Robustness checks using a logit-transformed outcome reproduced the same pattern, and model diagnostics, including DHARMa residual tests, showed no substantial deviations. Notably, none of the individual-difference or demographic covariates—AQ scores, gender, or block order—produced significant effects.

The direction of the approach effect deserves particular attention. A reward-learning framework might lead one to expect that a face that reliably responds to gaze with social engagement—an approach, an apparent closing of distance—would come to attract more looking, much as stimuli paired with reward continue to capture attention after the reward is gone. Instead, the participants looked less at the approaching face at test. The authors are careful about interpretation. Reduced gaze allocation, they argue, may reflect a redistribution of attention rather than subjective dislike or a stable negative evaluation of the face. One plausible account involves monitoring: during training, participants may have actively monitored the face to verify that its approach had completed, and when both faces were later presented as static, that monitoring demand was gone and attention was reallocated elsewhere. On this reading, gaze-triggered distance changes redistribute later gaze rather than simply increasing looking toward the approach-associated stimulus.

The authors are equally explicit about what the data cannot establish. Because the same face identities were consistently paired with specific outcomes during training and then shown as static at test, the design cannot definitively separate effects of face identity from effects of outcome history. Nor can it determine whether any association was attributed to the face itself or to the experienced gaze-triggered event. A further constraint is that approach and recede were implemented through changes in face size, so distance changes were necessarily accompanied by visual size changes; the study cannot determine whether the observed shifts reflect processing of interpersonal distance, of size, or of both. The use of neutral female faces in a screen-based task may also have limited the social salience of the signals and constrains generalizability to naturalistic interactions. The authors note that the observed effects do not imply subjective evaluation, value acquisition, or a specific learning mechanism, and that the null effect in the Recede condition could mask individual differences in the direction of gaze shifts that canceled out at the group level.

Even with these caveats, the paradigm opens a controlled window onto a phenomenon that most theories of social attention have treated only indirectly: the aftereffects of gaze-triggered social outcomes. In everyday life, looking at someone frequently precedes consequences—a step closer, a turn away, an approach or withdrawal—that then proceed without depending on continued gaze. The Tsukuba design isolates exactly that structure, separating the discrete triggering role of sustained gaze from the unfolding of the outcome and from any concurrent motion during assessment. The authors suggest that future work incorporating non-gaze-triggered control conditions, such as randomized onset of size changes, could establish whether the effects depend on the gaze-triggered structure itself rather than on motion per se, while varying facial expressions could test whether the modulation of gaze allocation generalizes across social and affective contexts. For now, the study delivers a provocative core finding: the eyes do not merely select the social world—they help set in motion the very outcomes that later select where the eyes will go.

Subject of Research: How gaze-triggered changes in apparent interpersonal distance affect subsequent gaze allocation to face stimuli

Article Title: Condition-specific changes in gaze allocation following gaze-triggered interpersonal-distance outcomes

Article References: An, C., & Matsuda, S. (2026). Condition-specific changes in gaze allocation following gaze-triggered interpersonal-distance outcomes. Attention, Perception, & Psychophysics, 88(7), Article 185. https://doi.org/10.3758/s13414-026-03322-8

Image Credits: AI Generated

DOI: 10.3758/s13414-026-03322-8

Keywords: gaze allocation, interpersonal distance, eye tracking, social attention, gaze-triggered feedback, value-driven attention, approach and avoidance, dwell time, face perception, attentional bias, linear mixed-effects model, social cognition