There is a floor on reaction time; anything faster is anticipation, not a genuine reaction

What it is

Genuine reaction time has a physiological floor: a "response" faster than that floor cannot be a real reply to the stimulus — it must be an anticipatory action the participant had already launched before the stimulus appeared, which merely happened to land close in time. For a simple visual stimulus, this floor sits around 100ms (varying with sensory modality and response method); any keypress noticeably faster than that should be treated as anticipation rather than a genuine reaction.

Why it happens

The floor is set jointly by the physiological speed limits of sensory conduction, central detection, and motor conduction — nerve conduction velocity, synaptic delay, and muscle activation time all have physical lower bounds, and there is no "faster genuine reaction" that can beat them. When a response comes in faster than that, the only sensible explanation is that the participant did not actually wait for and process the stimulus, but instead guessed and launched the action ahead of time based on the rhythm or probability of when stimuli tend to appear.

Studying it

A common way to detect anticipation is the catch trial: on a random subset of trials, no stimulus is presented at all, and if the participant still "responds," that shows the response pattern is rhythm-driven rather than stimulus-driven. Another approach is to set an outright floor threshold on the RT distribution (e.g., discard or flag any trial under 100ms as anticipation) before running statistical analysis on reaction time, so these outliers don't drag down the average and manufacture a spurious "got faster" conclusion.

Where it stops holding

The floor value itself depends on the specific sensory modality, stimulus intensity, and response method — there is no single number that generalizes across tasks, so applying a floor measured in one experiment to a different modality or response device does not hold. In addition, participants who are highly trained or highly familiar with the task's rhythm may genuinely have faster sensorimotor processing overall; this needs to be distinguished from pure guessing-based anticipation using catch-trial data, and cannot be settled just by noting that a response was fast.

Applying it

When evaluating user-testing data on interface response speed, if the statistics turn up an extremely short "reaction time" (far below 100ms), the first hypothesis should be that the participant jumped the gun by anticipating the rhythm, not that this is evidence of a genuinely fast system response — and it should not be kept in the dataset as such. Such abnormally fast samples should be excluded or flagged separately before analysis. How to check: build a number of blank trials into the test design where the expected prompt never appears, and observe whether the participant still "responds" — this tells you whether they are genuinely reacting to the prompt or following a rhythm they anticipated on their own.

Related

• Same group: A8.01.1 Reaction time breaks into sensory conduction, central processing, motor conduction, and muscle activation · A8.01.2 Reaction time and movement time are two independent stages that must be measured separately · A8.01.3 Simple reaction time: one stimulus, one response · A8.01.4 Choice reaction time: RT grows with the number of alternatives · A8.01.5 Discrimination reaction time: the stimulus category must be judged before deciding whether to respond · A8.01.7 Individual differences in reaction time exceed what most interface optimizations can gain
• Nearby: A8.03 Expectancy and preparatory signals
• Search terms: anticipation error · catch trial · reaction time floor