A dwell threshold that is too short causes false selection; too long causes a standing penalty
Aliases: dwell timeout · false dwell selection · dwell wait
What it is
The dwell-selection threshold is a cut: fixations shorter than it count as browsing; longer ones count as commit. Too short, and ordinary reading fixations and the back-and-forth of comparing neighbors cross the line as false selections. Too long, and every intentional choice waits out a clock—a standing penalty. There is no single millisecond that is jointly optimal for all targets and all people; there is a tradeoff curve that slides between false selection and waiting.
Why it happens
The duration distribution of natural fixations overlaps the distribution of intentional dwells; they are not two separated peaks. Reading a word often sits at 200–300 ms; comparing two buttons or checking a price spontaneously exceeds 400 ms. A cut in the overlap region yields both false positives (browse counted as select) and false negatives (select counted as browse); which side dominates moves with the cut. A short cut assigns most of the overlap to “select,” and false selections rise. A long cut assigns it to “browse,” intentional waits rise, and people start fighting the clock—staring, leaving early, abandoning sequences.
Target geometry shifts the curve. Sparse large buttons make it easy to fill the threshold; dense small ones interrupt the integrator, so the same milliseconds feel longer. Calibration error shifts it too: a systematic offset parks the aim point on the target’s edge, the clock breaks, users feel a penalty, and the log shows the threshold was never reached.
Studying it
Sweep a set of thresholds on the same targets (for example 300, 500, 800, 1200 ms) and plot error rate and selection time against threshold. The curves usually cross; the crossing moves with width, spacing, and user group. Split errors into two kinds: selected without intending to (false positive) and intended but not selected (false negative); they move in opposite directions with the threshold. Adaptive thresholds that follow recent errors or individual fixation statistics need a separate stability check: if the cut moves every minute, users cannot form a strategy. Threshold searches in gaze-typing papers are a ready template, but typing keys are equal-width character targets and do not transfer unchanged onto graphical buttons of mixed size.
Where it stops holding
Children, older adults, fatigue, and dry eye change natural fixation durations; crossing points from young-adult labs fail. For people with motor disabilities who communicate only by eye control, acceptable waits far exceed what an office user calls “slow.” In a game or while driving, an 800 ms penalty may be intolerable; for a rare home command such as toggling a lamp it may be fine. Motion in the scene (scrolling, animation) breaks the clock, so nominal and effective thresholds part company. A sudden offset from slipping glasses produces a burst of false selections at a short threshold and “I can never select” at a long one.
Applying it
- Band thresholds by target risk: navigation short, destructive actions long; do not share one “rule of thumb” across the whole UI.
- Offer a user-adjustable range and place the default at an explainable point on the false-selection–wait curve, not as a bare millisecond.
- Verify by sweeping thresholds separately on dense small targets and sparse large ones, counting false positives and false negatives apart, and checking that the crossing still sits near the default.