When several principles act at once, the outcome is set jointly by each cue's relative strength

What it is

Real layouts rarely rely on one grouping cue alone: spacing, color, connecting lines, and containers usually appear together. When they all point to the same grouping, the percept is very stable. When they point to different groupings, the grouping you actually see is not decided by one cue "absolutely beating" another — it is pulled out by how strong each cue happens to be in that specific display. A strong cue pulls harder; a weak one pulls less; whichever side pulls harder wins.

This is different from asking "which cue always ranks first." Take the same pair of cues, change the actual spacing and color values, and the winner can flip, because the outcome is set by each cue's actual strength in that display, not by a fixed rank the cue type carries with it everywhere.

Why it happens

It helps to picture each cue as casting a vote for how to group, with the vote's weight set by how strong that cue is in the current display — a bigger spacing difference gives proximity a heavier vote, a bigger color contrast gives similarity a heavier vote. When the cues agree, their votes add up, producing an especially stable percept that almost nobody misreads. When they conflict, the outcome leans toward whichever cue's vote is heavier; if the two are close, the percept becomes unstable — different people, or the same person at different moments, may report different groupings, similar to how a bistable figure flips back and forth.

Studying it

The standard way to test this "jointly decided by strength" account is the competing cues paradigm: build a grid or lattice where two cues act along two directions — say, spacing decides grouping across rows, color decides grouping down columns — and have participants make a forced choice, reporting whether they see the pattern grouped by row or by column.

The key move is to change only one variable at a time: hold the size of the color difference fixed, and vary only the spacing ratio (the ratio between within-group and between-group spacing), stepping it from nearly 1:1, where grouping is hard to call, up to a ratio large enough to make spacing-based grouping obvious. Everything else — element count, overall size, the color difference — stays fixed, so that whatever change shows up in the results can be attributed to the one variable that moved rather than to some other property of the display changing along with it.

Reading the results: plot the proportion of "grouped by spacing" reports as a function of the spacing ratio, and it traces an S-shaped curve. The point where that curve crosses 50% is where this particular spacing ratio exactly balances the fixed color difference — the two cues' pulls are equal there. Repeat that measurement at several different color-difference levels, and the resulting set of crossover points traces out a trade-off curve: how much spacing ratio it takes to cancel a given amount of color difference. That curve is the real, comparable "exchange rate" between the two cues — far more useful than a bare statement like "proximity beats similarity," because it answers by how much, and over what range of values that holds. Response time is a useful secondary measure too: even when the final report follows one cue, the strength of the "losing" cue often still leaves a trace in slower responses — meaning the losing cue was not ignored outright but kept exerting a pull, which supports weighted combination rather than a winner-take-all account.

Methodological cautions (the part that matters most here): an exchange rate measured with one specific set of spacing and color values only holds for those specific physical values — the absolute spacing scale, the number of elements, viewing distance, and exposure duration can all shift the rate itself, so a ratio measured with one dot-lattice study cannot be dropped unchanged onto an icon grid at a very different scale; it has to be re-measured for the target scenario's actual numbers. The forced binary choice itself has a limitation: allowing only "rows" or "columns" as answers hides genuinely intermediate or mixed percepts, which are common in real interfaces — supplementing with open description or confidence ratings helps recover that. And this exchange rate is a population-level average tendency; individual crossover points can vary considerably, so a claim of "this cue is stronger" based on one or two people's subjective impressions is not reliable evidence.

Where it stops holding

This weighted-voting picture fits best when a cue is continuously adjustable and can be given a graded strength — spacing, color difference. Cues that are closer to an all-or-nothing presence, like a connecting line or a truly closed container, tend to behave more like an outright lock on the outcome once present, rather than dragging the percept smoothly in proportion to their strength — the resulting curve tends to be steeper, closer to a step than a smooth S-shape.

This method typically manipulates only two cues at a time in a clean experimental display, while a real interface usually stacks alignment, whitespace, borders, and icon style on top of each other. The workload of measuring pairwise exchange rates for every pair grows fast, and results from a two-cue experiment should not be assumed to generalize directly to a complex layout carrying five or six cues at once.

Applying it

When a layout has two candidate grouping cues and it is unclear which will win, a full lab study is not required — a scaled-down version of the competing cues check works: fix one cue's strength (say, keep the color scheme unchanged), vary only the other cue's strength (say, the spacing ratio) across two or three variants, and show them to five or six users, asking each to immediately say how they see the elements grouped. That locates the rough crossover point for this specific case, rather than borrowing an abstract priority ranking from some paper.

Related

• Same group: A2.10.2 Grouping intent has to be layered across principles, not carried by a single one · A2.10.4 Continuity and closure together resolve toward whichever reading is simplest overall · A2.10.5 Gestalt principles describe general tendencies, not a fixed lookup-table priority · A2.10.6 Cue-conflict outcomes in a specific layout need user testing, not just reasoning from principles
• Nearby: A2.01.2 Proximity dominance over similarity · A2.07.2 Common-region grouping beats proximity grouping · A2.08.2 Connectedness is the strongest grouping cue
• Search terms: cue combination · competing cues paradigm · point of subjective equality