The fixation point is not the locus of attention
Aliases: covert attention · looking without attending · gaze versus attention
What it is
An eye tracker reports where the fovea is aimed, not which object attention is serving. Covert attention can leave the fixation point: someone can hold a central cross and prepare a target in the periphery, or leave the eyes on a sentence already read while attention moves to the next decision. Treating gaze coordinates as a sufficient statistic for “what they are thinking or about to do” collapses seeing, attending, and intending into one thing.
Why it happens
Attention can shift in space without an eye movement. Posner cueing showed this repeatedly: a valid peripheral cue speeds a later target even when the eyes stay on center. Before the next saccade, attention typically moves first to the intended landing site, so the impending eye jump itself creates a brief gaze–attention split. Antisaccade tasks require looking opposite the stimulus, deliberately tearing the two apart. Everyday interfaces produce messier splits: during proofreading the eyes sit on a typo while attention retrieves the intended spelling; during a captioned figure, attention is on the narration. Pupil, EEG, or verbal report can sometimes catch the mismatch; gaze coordinates alone cannot.
Gaze remains a strong prior—people usually send the fovea to what they are processing—but the relation is correlation, not identity. The correlation is stronger in idle browsing and weaker while waiting, listening, overloaded, or deliberately monitoring the periphery.
Studying it
Posner cueing, antisaccades, and “fixate center, report periphery” probes are the laboratory tools that split gaze from attention. When eye position is a dependent measure, the question is not the landing point itself but whether that point agrees with report, with action, and with secondary-task accuracy. Interface studies look more like dual tasks: the primary task drives gaze (reading, search) while a probe asks “which item was just highlighted,” or a change that only peripheral attention can catch is inserted while gaze stays on a control. The disagreement rate is the measure that belongs to this claim, not mean fixation duration. A methodological risk is that instructions turn participants into attention detectives: once they know they will be asked what they saw, covert splits are replaced by overt staring. Mind-wandering, waiting, and listening as they occur in the wild are hard to recreate intact in the lab.
Where it stops holding
When high spatial frequencies must be resolved (small type, adjacent icons), attention almost has to take the fovea with it; the room for dissociation shrinks. Driving, monitoring a large display, or glancing at a notification while listening is the opposite: dissociation is the default. Strabismus, amblyopia, or failed monocular tracking can make the reported point not even the fovea of the seeing eye. The product error of equating gaze with attention is especially common in advertising heatmaps and “looked equals interested” conversion stories—that is a commercial measurement habit, not a perceptual fact. Some brain injuries allow the eyes to move while attention transfers poorly, or the reverse.
Applying it
- Do not treat “gaze is on the control” as sole evidence of an imminent action or of comprehension; when comprehension matters, cross-check with paraphrase, choice, or a later action.
- Adaptive interfaces that push content or steal focus from the current gaze point will misread people who are monitoring the periphery or listening.
- Verify by inserting probes that are offset from gaze (peripheral changes, auditory instructions) in real tasks, and counting how often people can report the probe while fixation stays elsewhere.