Week 27 · Human Interaction Psychology
End of experience
In 1947, competent pilots kept retracting the landing gear when they meant to lower the flaps — on clear days, in serviceable aircraft. The accident report always said the same two words. Two psychologists proved the report was wrong.
Part I
In June 1954, Paul Fitts had subjects tap a stylus between two metal plates as fast and as accurately as they could, and varied only two things: how far apart the plates stood, and how wide they were.
Below is the same test. Twelve targets, sized between 28 and 96 pixels, placed at distances your last tap sets up. Tap the true center of each one. Miss, and it simply asks again — a miss does not count against you, it retargets. Do it on a phone if you can; that is where this law now lives.
Tapping board
12 targets, 28–96px wide
Twelve targets. Tap the true center, as fast and as accurately as you can.
Run the board above at least eight times through and this paragraph will fill in with your own bits-per-second, set beside Fitts’s 1954 average of 10.1.
Part II
Fitts needed one scale that could compare a short reach to a wide plate against a long reach to a narrow one. He borrowed it from Claude Shannon: treat distance as a signal and target width as noise, and difficulty falls out as a bit count.
His own measure was ID = log2(2D / W), where D is the distance to travel and W the target’s width. The version this piece computes with, live, is I. Scott MacKenzie’s later refinement — ID = log2(D / W + 1) — which never goes negative for a target you are already resting on, and fits Fitts’s own 1954 data slightly better (r = .9937 against .9831). Every bit figure on this page uses MacKenzie’s form. Where the two disagree is only at very close range; for a distant target they converge.
Movement time then follows a straight line in that difficulty: MT = a + b × ID. Across sixteen conditions in 1954, that line held steady at roughly 10 bits of information per second — a bandwidth for the human arm, the same way a modem has a bandwidth for a wire.
Predict
Which reaches the thumb faster: a 96px button 480px away, or a 30px button 150px away?
That trade has an exact shape under Fitts’s original formula. Double the distance, or halve the width — either move adds exactly one bit: log2(2·2D / W) = log2(2D / W) + 1, and log2(2D / (W/2)) = log2(2D / W) + 1. The same one bit, either way. “Make it bigger” and “move it closer” are not two separate design tickets — they are one lever, marked off in bits.
Which is also why “Fitts’s Law means make everything huge” misreads the law. The cost is logarithmic: each doubling buys one bit, and past a comfortable size the gains all but vanish while the layout keeps paying rent. Size is a budget to spend where taps are frequent or a miss is expensive — not a virtue to maximize everywhere at once.
Part III
The logarithm in Fitts’s equation is not decoration — it is the fingerprint of a two-part movement that R. S. Woodworth described back in 1899, decades before Fitts gave it a number.
A fast, coarse ballistic phase throws the hand most of the way there on momentum alone. A slower, feedback-guided correction phase then homes in on the target — and the smaller the target, the more of these corrective loops it costs, each one buying a little more precision for a little more time. That correction phase is very likely what you felt as hesitation on the narrowest targets in the board above.
Two movements
same reach, different difficulty
Easy — 1.08 bits
Hard — 4.00 bits
The ballistic/correction split shown above is illustrative, not measured — it dramatizes the mechanism the literature describes (Crossman & Goodeve, 1963; Meyer et al., 1988), which this dossier itself flags as standard-but-secondhand: the primary papers were not independently checked. Graded accordingly below.
Part IV
On a phone, the pointer is not a cursor you steer — it is a thumb, anchored at a corner of the device you are also holding.
Steven Hoober watched 1,333 people use their phones in public and found three grips — one-handed (49%), cradled (36%), two-handed (15%) — switching constantly as tasks change, with roughly a third of one-handed use on the left thumb. Tap a zone below for what that means for placement.
Thumb reach
right-hand grip, one-handed
The thumb's natural resting sweep. Thumbs drive roughly 75% of all phone interactions (Hoober, 2013) — your highest-frequency, highest-consequence action belongs here, full width, nothing else crowding it.
Cursors get a trick thumbs cannot use: the screen edge is a wall. A target flush against it has effectively infinite width along the axis of approach — the pointer cannot overshoot, so there is no correction phase to pay for. Apple pinned its 1984 menu bar to the absolute top edge; contemporary Windows attached menus to each window, leaving dead pixels above them.
Bruce Tognazzini claimed edge-pinned menu acquisition runs roughly five times faster on the Mac. The directional claim — edges win for a cursor — is uncontroversial Fitts arithmetic; the specific “5×” is his own figure, widely repeated but not, in this dossier’s sources, a peer-reviewed result. And the edge trick is a cursor privilege only: on touch, the same edge is bezel, palm-rejection territory, and frequently slower to hit, not faster.
Part V
Same one-bit budget as the tapping board, now yours to allocate directly.
Slide either control and watch the difficulty and the predicted time move together — and watch the movement below visibly grow its correction phase as the ratio gets harder.
One budget
Move either slider
Your movement
This lab uses MacKenzie’s Shannon form (as does the board above), and the “+1 bit” identity from Part II is exact under Fitts’s original formula, only approximate here. It will not visibly matter for any distance many widths away from its target.
Challenge · Checkout
OpenPlace order and Cancel both sit in the top row right now — hostile territory, and adjacent to each other. Move Place order into the comfortable bottom row, and put real distance between it and Cancel.
Placing
Insight
The miss was never yours. It was always the panel’s.
In 1947, Fitts and Jones read through 460 reports of pilots operating the wrong control, and found that the common factor was never the pilot — it was cockpits with identical, adjacent levers for gear and flaps. The fix was not retraining. It was shape-coded knobs. Every rage-tap, every mis-click cluster in your own analytics is that same report, filed eighty years later: a pattern of competent people making the same mistake is a property of the interface they were handed, measurable in the same D and W you have been reading all along.
Mental model
Whenever a user misses, hesitates, or says they ‘can’t find’ a control,
look for the ratio of distance to width behind it, not their attention span,
because you control both sides of that ratio, and every halving of one or doubling of the other costs exactly one measurable bit.
Tomorrowmeasure your app’s primary call-to-action against the 44pt / 48dp floor. If it is smaller, you already know the fix, and now you know why it works.