Law of proximity

The perceptual claim is old and unusually well quantified. Wertheimer listed proximity among the factors that organise an array in 1923, and Kubovy, Holcombe and Wagemans turned it into a measurable function using dot lattices, where competing organisations differ only in inter-dot distance. As Wagemans and colleagues summarise, grouping “depends only on the relative distance between dots in competing organizations” and “decays as an exponential function of (relative) distance,” regardless of “the overall configuration in which the competition occurs,” and the slope of that function is “a person-dependent measure of sensitivity to proximity.” Kubovy and van den Berg found proximity and similarity additive. So even in the clean case the law fixes the shape of the effect, not its size for a given user: a spacing scale has to work as a set of ratios, because ratios are all the visual system reads.

Whitespace is the weakest grouping cue available, which is where the usual design advice goes wrong. Villalba-García and colleagues equated the strength of each cue for the visuomotor system first, then pitted cues against each other, and found “a consistent pattern of processing dominance of: (1) luminance similarity over proximity and (2) common region over connectedness.” Quinlan and Wilton found persuasive effects of both proximity and common colour, weak evidence for common shape, and that “both common colour and common shape were shown to override grouping by proximity, under certain conditions.” Palmer's common region — elements inside one bounded area group together — is reported as “capable of overcoming the effects of other powerful grouping factors such as proximity and similarity,” though on demonstration displays rather than a quantitative test. The build reading is the reverse of the folk rule: a card border or a shift in background tone beats a gap, so carry the grouping you can least afford to lose on the border, not the margin.

Connecting elements does not buy the same override. Palmer and Rock argued that uniform connectedness partitions the image before the classical principles act at all; Han, Humphreys and Chen tested that ordering with global-letter discrimination and found it half right. Connecting the elements speeded grouping by shape similarity, but “the discrimination of stimuli grouped by proximity was unaffected by connecting the local elements,” and only against weak proximity did connectedness come out faster. So a bounded region overrides proximity and a connecting line does not add to it.

At the interface level the measured record is thinner than the principle's popularity suggests, and it is almost all visual search. Niemelä and Saarinen found that “both the presence of icons and their grouping had a significant positive effect on the scanning speed.” Hornof compared labelled and unlabelled visual hierarchies and found the labelled ones “searched much more efficiently,” with “a fundamentally different strategy for each of the two structures” — the label, not the gap, is what let people skip whole groups. Spacing is not monotonically good either: Everett and Byrne's CHI paper is titled for its result, that varying icon spacing changes users' visual search strategy, and Reddy and VanRullen found the dependence task-specific, with more spacing significantly improving an upright-among-inverted face search, leaving a bisected-disk search poor, and doing nothing to two classically parallel searches.

The display literature that generalised proximity is the literature that found its limits. Wickens and Carswell's proximity compatibility principle pairs close display proximity with close task proximity, so grouping is only right when the task integrates the grouped items; for a value that must be read in isolation the prescription reverses. In the experiments behind it, Wickens and Andre varied how three aircraft stall indicators were displayed and found “spatial proximity had little effect on either focused attention or integration, whereas a distinct color code improved focused attention and disrupted integration.” Distance was the manipulation that did least.

What is missing is any controlled evidence for the interventions this principle is invoked to justify. No study here shows that regrouping form fields, boxing plan cards or widening a section gap moves completion, comprehension or conversion, and the narrower label question is openly unresolved: Tangmanee and Ruchikachorn eye-tracked 312 participants across left, top and in-field labels and found left-aligned labels noticed fastest but “the top or the infield alignments still outrank the left one” on fixations, while Das, McEwan and Douglas, on “a small sample of UK university-educated users,” recommended right alignment against prevailing guidance. Placement does move behaviour at scale where it has been measured, but as position rather than grouping: Utz and colleagues ran consent notices past more than 80,000 unique users and found “users are more likely to interact with a notice shown in the lower (left) part of the screen.” Treat this as a reliable fact about perception and an untested theory of conversion: use it to decide what reads as one thing, then test the thing you changed. The law runs whether or not the relationship is real, which is why putting a consent control next to an unrelated button is a legal problem rather than a style choice.