Up Is Not Down: Head Movements Pull Attention in Opposite Directions Depending on Axis

When someone turns their head to the left while keeping their eyes locked on you, your attention does something strange: it drifts in the opposite direction. This counterintuitive phenomenon, known as the overshoot effect, has fascinated vision scientists for decades. Now a new study suggests that this quirk of social attention is not a universal rule of the human brain but a peculiarity of horizontal space. When heads tilt up or down instead of turning left or right, attention follows the head rather than fleeing from it. The finding, published in the journal Attention, Perception, & Psychophysics, reveals a striking dimensional asymmetry in how the mind reads the attention of others, and it challenges long-standing theories of how gaze and head cues are combined.

The research, conducted by Felix J. Götz, Paul M. Heineck, Anne Böckler, and Andreas B. Eder at the Julius-Maximilians-University of Würzburg, set out to answer a deceptively simple question: does the brain integrate gaze and head information the same way in vertical space as it does in horizontal space? The dominant models of social attention, developed largely to explain horizontal effects, implicitly assume that it does. Hierarchical accounts propose that the eyes are the privileged signal, with head orientation playing a subordinate role. Holistic accounts instead argue that gaze, head, and body cues are merged in parallel into a single estimate of where another person is looking. Both frameworks predict that a misaligned configuration, in which the head points one way and the eyes another, should produce similar attentional consequences regardless of whether the misalignment occurs along the horizontal or the vertical axis.

The Würzburg team suspected otherwise, and their reasoning drew on the biomechanics of real eye-head coordination. In everyday life, people glance left and right with ease while keeping the head nearly still, but looking up or down typically requires a tighter coupling of eyes and head, because vertical eye mobility is comparatively restricted. Misalignments between eye and head direction therefore occur far more often along the horizontal axis than the vertical one. If the visual system has internalized these ecological regularities, it should treat a vertically tilted head as a more reliable predictor of where attention is heading, while in the horizontal plane it may lean more heavily on gaze information and anticipate how the eyes will move next. To test this, the researchers needed stimuli in which the head itself visibly moved, rather than the static photographs used in most classic overshoot experiments.

Across four preregistered experiments, participants viewed centrally presented faces of four models wearing uniform black shirts against a white background. Each trial began with a fixation cross, followed by a neutral head image for one second. The image was then replaced by a shifted version: the head turned left or right, or tilted up or down. After 300 milliseconds, a blue circle or square appeared at one of two locations on the relevant axis, and participants classified its shape as quickly as possible. Crucially, the cues were entirely non-predictive; targets were equally likely to appear at cued and uncued locations, so any response-time differences reflected spontaneous shifts of attention rather than learned expectancies. Gaze visibility was manipulated between experiments, with the models either gazing directly at the observer or keeping their eyes closed.

The first two experiments probed the vertical dimension. In Experiment 1, with direct gaze maintained throughout, forty-three participants identified targets significantly faster when the target location matched the direction of the head tilt, a head-congruency effect of roughly twelve milliseconds, with an effect size of dz = 0.67. Attention, in other words, followed the head upward or downward rather than overshooting away from it. But a rival explanation remained: the vertical tilt also displaced the eye region toward the target, so the effect might have been driven by the position of the eyes rather than the head itself. Experiment 2 eliminated this confound by repeating the procedure with closed eyes. The head-congruency effect not only persisted but grew dramatically stronger, reaching dz = 1.60, with targets at head-congruent locations detected about seventeen milliseconds faster. Vertical cueing, the authors concluded, genuinely reflects head orientation, not displaced eye position.

Experiments 3 and 4 then turned to the horizontal axis, and here the classic pattern re-emerged in unexpected form. With direct gaze and dynamic head turns, Experiment 3 replicated the overshoot effect: participants responded faster to targets appearing opposite to the head direction, with a significant effect of dz = 0.65. This confirmed that the overshoot survives the transition from static images to moving heads. The surprise came in Experiment 4, where the researchers removed all gaze information by closing the models’ eyes. Standard cue-weighting logic predicted that without eyes to compete with, head orientation should dominate and the overshoot should reverse into head-following. Instead, the opposite happened. The head-opposite bias grew even stronger, reaching dz = 0.92, with head-incongruent targets identified about thirteen milliseconds faster. Attention fled from the direction of the head turn even when there were no eyes to read at all.

Exploratory comparisons across the four experiments reinforced the dimensional dissociation. Pooling the vertical experiments, the head-congruency effect was significantly larger with eyes closed than with eyes open, an interaction confirmed by linear mixed models fitted to log-transformed single-trial response times. The same pattern appeared horizontally, where the overshoot-like bias was modestly amplified without visible gaze, though the authors flag this horizontal modulation as tentative given its smaller size and exploratory status. Taken together, the results suggest that direct gaze acts as an additional attentional signal that attenuates head-based cueing in both dimensions, but that it never changes the direction of the effect. Vertical attention follows heads; horizontal attention flees them, with or without eyes.

What could produce such a bizarre horizontal bias in the complete absence of gaze? The authors propose that observers do not merely encode the current geometric relationship between eyes and head but engage in anticipatory inference about how another person’s attention will be realigned. A dynamically turning head implies that attention is in transit. In the vertical dimension, where eye and head movements are tightly coupled, the head is a trustworthy forecast of where the person will look, so observers orient toward it. In the horizontal dimension, where the eyes roam independently, a turning head may signal that the person is disengaging from one location and redirecting attention elsewhere, prompting observers to orient toward the likely source of that shift. Removing gaze information may increase the ambiguity of the observed movement, encouraging richer, intention-based interpretations, such as avoidance or disengagement, rather than simple directional reading. This account aligns with the idea of attention reading, in which observers dynamically integrate head, gaze, and motion cues to predict where someone’s attention will land next.

The findings are difficult to reconcile with strictly hierarchical models, which cannot explain head-following in the vertical dimension, or with holistic models, which predict no dimensional asymmetries at all and certainly cannot accommodate a horizontal head-opposite bias when no gaze is visible. Instead, they favor flexible, predictive frameworks in which cue integration is constrained by the biomechanics and ecology of eye-head coordination. The authors acknowledge limitations: response times offer only an indirect window onto attention compared with eye tracking, the stimulus set comprised just four models, and the horizontal and vertical manipulations were run in separate experiments for methodological reasons. Yet the consistency of the dissociation across four preregistered studies, with all materials and data publicly available on the Open Science Framework, makes a compelling case. The next time a friend tilts their chin toward the ceiling, your attention will dutifully follow. But should they swing their head to the side while their eyes stay on you, your mind, ever the fortune-teller, will already be looking where they are about to look next.

Subject of Research: Dimensional asymmetries in social attention and gaze-head cue integration during dynamic head movements

Article Title: Dimensional asymmetries in social attention: Dynamic head movements elicit opposite attentional biases in vertical versus horizontal dimensions

Article References: Dimensional asymmetries in social attention: Dynamic head movements elicit opposite attentional biases in vertical versus horizontal dimensions. (n.d.). https://doi.org/10.3758/s13414-026-03351-3

Image Credits: AI Generated

DOI: 10.3758/s13414-026-03351-3

Keywords: social attention, gaze-head integration, overshoot effect, spatial cueing, head movements, eye-head coordination, joint attention, anticipatory inference, gaze perception, attentional orienting, dimensional asymmetry, psychophysics