What Is Choice Reaction Time?
Reaction-time tasks come in two classic flavors. In simple reaction time, one stimulus maps to one response: wait for the light, tap the moment it appears. In choice reaction time, any of several stimuli can appear and each maps to a different response — find the one that changed and tap it, fast. The rules are easy to state; the gap between the two versions is where the science lives.
That gap was the founding observation of mental chronometry. Donders, in the 1860s, proposed subtracting simple RT from choice RT to isolate the time the decision itself takes. A century later Hick and Hyman showed the choice component follows a clean law: RT grows linearly with log2 of the number of alternatives — two options cost one bit of decision, sixteen cost four. Together these tasks make up the processing-speed paradigm family. An honest caveat, same as everywhere in this literature: RT tasks are excellent measurements, but evidence that practicing them transfers to anything beyond the practiced task is thin.
What makes a real reaction-time implementation
A few details separate an honest RT task from a tapping test:
- A randomized foreperiod. If the wait before the stimulus is fixed, users learn the rhythm and respond on prediction, not reaction. The foreperiod must vary — and premature taps must be caught. Simple RT conventionally discards anticipations and retries; choice RT scores them as failures. Either way, an app that can't tell a false start from a real response isn't measuring RT.
- Millisecond-grade timing. Reaction time lives in the 200–700ms range. Timestamps should come from a monotonic media clock taken at stimulus onset and tap completion — anything coarser swamps the effect.
- Varying set size per trial. The Hick–Hyman result only shows up if the number of choices changes trial to trial. N should be drawn randomly within a range, so each rep pairs an RT with that rep's set size — you measure the slope directly rather than a single point.
- Windows that scale with the task. The response deadline should grow roughly linearly with choice count (flat per-choice time), not collapse as N grows — otherwise you're confounding speed pressure with decision load.
- RT as the score. The headline metric is mean RT of clean trials, with timeouts, wrong choices, and false starts excluded — not an accuracy percentage.
Reaction time in Avoken
Avoken ships a simple-RT and a choice-RT workout built as companion
measures of the Donders contrast. The simple version is a single circle
that turns green after a randomized 0.5–1.5s foreperiod; the response
window starts at 1000ms and adapts per user down toward a ~350ms floor —
near the young-adult simple-RT floor including touch latency — and
anticipations are discarded, not scored. The choice version draws 2–16
circles per rep (each rep's N randomized within the level's range, so
the Hick's Law slope is measured directly), with the response window
scaling ~500ms per choice, harder color discrimination at upper levels
(green-vs-teal, then green-vs-yellow), and decoy flashes that tempt
premature taps. Both timestamp with CACurrentMediaTime()
and score on mean RT of clean trials.
Avoken is a training app, not a treatment — what it trains is the task itself, and we don't claim more than that. If you're fond of the paradigm, it's implemented with care here.