The actor has multiple views of a single model, each showing a different aspect of the same underlying data. Relating these views is a manual task: find the item in one view, re-find it in the other, hold the correspondence in mind. Coordinated views wires them together so that selection, navigation, and highlighting propagate — what is designated here lights up there, and relationships between aspects become something the actor sees rather than remembers and infers.
Solution
Couple the views so that what the actor designates in one — a selection, a focus, a highlight — carries to the others, propagating across chains of views; brushing and linking are the canonical mechanics. Baldonado’s eight guidelines are the trade-off machinery — four for when another view earns its place (different users and tasks, complementary aspects, complex data worth decomposing, all tempered by parsimony — merge views with near-identical semantics, and couple only where coupling demonstrably helps), four for how to run the ensemble (spend space and time deliberately; make the coupling self-evident — propagation under ~100 ms, or visibly decouple, greying out the stale view; keep affordances and state consistent across views; manage attention so priority events pull the actor to the right view).
Resources & references
- Wang Baldonado, Woodruff & Kuchinsky (2000) Guidelines for using multiple views in information visualization — the eight guidelines; “two or more distinct views of a single conceptual entity”
- Roberts (2007) State of the art: coordinated & multiple views in exploratory visualization — the survey of what grew from the guidelines; queued as the deepening read
- Oh, Lee, Kim et al. (2022) An empirical study on the relationship between the number of coordinated views and visual analysis — evidence on the parsimony trade-off as view count grows
Consequences
- designating something in one view marks it in every view where it appears; cross-view relationships — correlations invisible in any single view — arrive perceptually instead of being reconstructed
- every added view charges for what the coupling buys, in learning time, working memory, screen space, and upkeep
- each coupling is now a correctness surface: a false cue implies a relationship that doesn't hold, and propagation slower than ~100 ms reads as no relationship at all — the cues, not just the data, can lie
Related patterns
Related
- Regeneration — parallel variants are judged side by side
Preceded by
- Pan and zoom — everything keeps its place in one coordinate system, so nothing breaks the spatial sense — and orientation becomes the actor's problem: at any zoom the view carries no evidence of where it sits in the whole, which is what a second, wider viewport kept in step is for
Enabled by
- Selection — the coupling payload is a propagated designation — sometimes a staked selection (acted on as a target), sometimes mere focus (navigated to as a position); which one decides what the coupled views may do with it
Instantiated by
- Overview and detail — overview-detail is the two-view coupled system where one view is compressed and the other full; what travels between them is focus — the current item — not a selection