A saccade comes across as if the eye is searching for something. It isn't. By the time the eye starts moving, the decision about where it's going has already been made.
The Decision Is Made Before the Eye Moves
Saccadic motor planning is underway well before the eye leaves its starting position. In the frontal eye field, target-specific activity builds toward a movement-triggering threshold. In the superior colliculus, that activity converges with sensory, cortical, and basal-ganglia input into a spatially organized gaze command. Trial-to-trial variation in how fast a saccade gets launched is driven mostly by the state of this pre-movement activity — how high it started, how early it began accumulating, how steeply it rose — not by a late "go" signal at the moment the eye moves. In gap-saccade experiments, removing the fixation target before a peripheral target appears measurably increases this preparatory buildup and shortens latency by roughly 20 milliseconds. Reversible inactivation of the frontal eye field increases saccadic reaction time and reduces that same preparatory buildup downstream in the colliculus, which is about as direct as causal evidence gets in this literature.
The practical implication: saccadic latency is not a measure of how fast an eye muscle responds. It's a measure of how ready the system already was.
Where the Saccade Goes Is Where Attention Already Went
Rizzolatti's premotor theory of attention made a strong claim in 1987: covert attention to a location is the subthreshold preparation of a saccade to that location. The theory has been narrowed since — attention and saccade programming are not perfectly identical, and covert attention can occur without a saccade pending — but the core finding has held up under decades of scrutiny. Preparing a saccade improves perceptual discrimination at the intended landing point before the eye moves. Covert attention to a location, even with fixation held steady, measurably deflects the trajectory of a later saccade elsewhere. Weak electrical stimulation of a frontal eye field site — too weak to trigger a saccade — still improves detection at that site's corresponding location in space, and alters visual processing in area V4 in a way that looks like ordinary spatial attention.
The two systems draw on overlapping spatial priority maps in frontal eye field, superior colliculus, and posterior parietal cortex. A location becomes behaviorally important when the neural population encoding its potential gaze vector is elevated relative to competing locations. If the movement is withheld, that elevated representation still biases perception. If threshold is reached, the same competition produces the saccade. Attention and saccade target selection are two expressions of the same underlying computation, not two separate systems that happen to cooperate.
The Saccade Requires an Accurate, Continuously Updated Spatial Map
Every saccade displaces the retinal image. If the visual system only worked off retinal input, the world would appear to jump with every eye movement. It doesn't, because the brain updates its spatial representation before the eye moves, using a copy of the planned movement rather than waiting for the visual consequences of that movement to arrive.
This was demonstrated directly in parietal cortex: neurons in area LIP begin responding to a stimulus at its future retinal location — the location it will occupy once the impending saccade lands — before the saccade happens. The response can't be visual feedback, because the visual consequence hasn't occurred yet. It has to be driven by an internal copy of the motor command. That corollary-discharge signal has since been traced along a specific pathway from superior colliculus through the mediodorsal thalamus to frontal eye field, and interrupting that pathway degrades the predictive shift in frontal visual processing. The same predictive updating shows up in frontal eye field and superior colliculus as well, not just parietal cortex.
What this means functionally: planning where to look next requires the brain to already know, in advance, what the visual world will look like after the eye lands. A child whose spatial map isn't reliably updated isn't going to have a clean, easily localized "tracking" problem. They're going to have a harder time on anything that requires sequencing more than one fixation — returning to an interrupted target, planning a second saccade before the first has landed, keeping the location of something just seen while looking elsewhere.
Attention and Saccadic Control Run Through the Same Circuitry
Because saccadic planning and spatial attention share substrates, problems with sustained attention usually show up in saccadic behavior — not as sloppier eye movements, but as something more specific: instability in the moment-to-moment engagement of the system that plans and monitors them.
The antisaccade task is the clearest window into this. The participant has to suppress a reflexive saccade toward a cue and instead move to the mirror-opposite location — which requires maintaining a task rule, suppressing a prepotent visual response, transforming the spatial coordinate, and monitoring the outcome. A 2010 study comparing children and adolescents with ADHD, youth-onset psychosis, and controls found that raw antisaccade error rates were not significantly elevated in the ADHD group. What was elevated: how often an initial error went uncorrected, and how variable antisaccade reaction time was from trial to trial. The authors' interpretation was specific — this pattern points to impaired trial-by-trial sustained attention and self-monitoring, not a global inability to inhibit a response or hold an instruction in mind. A 2021 meta-analysis pooling 26 ADHD oculomotor studies found the same general shape: real deficits concentrated in saccade inhibition and memory-guided saccades, and no evidence of a basic pursuit or saccade-metrics problem.
This is a materially different picture than "kids with attention problems have worse eye movements." The oculomotor system isn't degraded in ADHD. The control and monitoring layer sitting on top of it is unstable. That distinction matters clinically, because it tells you that you shouldn't actually be looking for accuracy or speed, as much as you should be looking for consistency and self-correction across trials.
What a "Poor Tracking" Finding Is Actually Telling You
A NSUCO oculomotor exam or a DEM score that comes back low is not just a motor finding. Both are influenced by fixation stability, naming speed, and task engagement as much as by the eye movement itself — a 2021 pilot study comparing NSUCO scores against objective eye-tracking in children with neurodevelopmental disorders found the subjective score correlated well with hypometric saccades and regression rate, which supports NSUCO as a legitimate clinical screen precisely because it's picking up more than isolated motor output.
This is also why "training the eye movement" is a misleading way to describe what a structured saccadic task is actually doing. The eye movement was never the broken part. What a repeated saccadic task demands — hold the rule, suppress the reflexive response, sequence the plan, catch and correct the error, do it again on the next trial without losing the thread — is the exact same trial-by-trial control and self-monitoring that showed up as unstable in the antisaccade data above. You're not drilling a motor skill until it's smooth. You're putting repeated, structured load on the part of the circuit that has to stay engaged from one trial to the next, because that's the part with the actual deficit.
The saccade is just the observable trace of whether that circuit held together on a given trial. Improvement on a saccadic task isn't "better eye movements" as an end in itself — it's evidence that the underlying control and monitoring system is getting more reliable, on the one output where you can actually watch it happen in real time.