Hick's law

The law rests on two small experiments. Hick (1952) ran the first on a single subject — the experimenter himself, who notes at least 8,000 practice reactions before the ones he reported. His mean reaction times fitted a logarithm of the number of alternatives, about 156 milliseconds per bit of choice, and the same function fitted Merkel's 1885 data at about 188 ms per bit; Hick summarised the result as a rate of gain of information of the order of five bits per second. His second experiment, in which subjects deliberately traded accuracy for speed, had two subjects.

Hyman (1953) is the replication that put the second name on the law, and it is barely larger: four male undergraduates responding by voice to lights given nonsense names. He varied the information per stimulus three ways — the number of equally likely alternatives, the relative frequencies of particular alternatives, and sequential dependencies between successive stimuli — and reaction time came out linear in bits for every subject. The slope, the one number a designer could use, ranged from 127 to 215 ms per bit across those four people. That is the provenance of the law: practised keypressing, not a time budget you can put on a screen.

The shape is the part that gets dropped on the way into design advice. Because the cost is logarithmic, doubling the number of options adds one bit and therefore adds a fixed increment — on Hyman's slopes, roughly a seventh to a fifth of a second — whether you go from two options to four or from sixteen to thirty-two. Liu, Gori, Rioul, Beaudouin-Lafon and Guiard reviewed the choice-reaction literature for CHI 2020 and concluded that Hick's law “speaks against, not for, the popular principle that ‘less is better’”, that logarithmic growth in an observed timing curve is not by itself evidence that the law is operating, and that in HCI tasks choice-reaction time can often be assumed constant. The extra option is not what is making your screen slow.

A first encounter with a menu is a search task, and search is linear. Bailly, Oulasvirta, Brumby and Howes (CHI 2014) had 21 participants select labelled items from menus of 8, 12 and 16 items under eye tracking, in unordered, alphabetical and semantic layouts, with items repeated so practice could build. Selection time rose significantly with menu length, and that rise shrank from the first block of trials to the last. Unordered menus were consistently slower than alphabetical or semantic ones, targets further down the list took longer until the layout was learned, and trials where the target was absent took longest, which is what exhaustive scanning predicts. Their model puts the conclusion plainly: with practice, performance shifts from search, which is linear in the number of items, to decision, which is logarithmic and is where the Hick–Hyman law lives. Ordering and labelling a list buys more than shortening it.

Holding the number of options fixed makes the other lesson visible. Fitts and Seeger (1953) tested 72 airmen, crossing three eight-light stimulus panels with three eight-position response panels. Every combination carried the same three bits of information, so Hick's law predicts the same time for all nine. Mean reaction time ranged from 0.39 seconds for the best-matched pair to 0.77 seconds for the worst, and errors from 3.4% to 18.8%. How the options map onto the controls moved performance by nearly a factor of two while the count never changed.

So: well replicated inside its own paradigm, badly over-applied outside it. Proctor and Schneider's 2018 review lists the standing challenges to the law — stimulus–response compatibility, practice, very large set sizes and sequential dependencies — and notes, with some irony, that it is now regarded as fundamental to the design of interfaces that did not exist when Hick ran it. None of this work measures what design advice claims for it. These are millisecond-scale results about one practised keypress; nobody here tested whether cutting options from a pricing page or a navigation bar raises conversion, and none of it is about whether people choose at all rather than how fast — that is choice overload, a separate and considerably weaker finding. Treat Hick's law as a reason to order, group and label a list well, not as evidence for deleting things from it.