It might just be me, but I've always thought that we have a couple of different terms to describe one concept in data visualization: having two or more visualization views and somehow synchronizing them. Here are the terms that I believe are somehow related:
(multiple) linked views
multiple coordinated views or coordinated multiple views
brushing and linking
On top of that, I believe the following can be special cases of the above:
overview and detail
focus + context
This post documents my path of trying to get to the bottom of the terminology. I'm mainly curious whether each of these implies a slightly different way of synchronizing multiple views, or whether the fact that we have multiple terms is purely historical.
tl;dr: (1) There's no single preferred term for linked views. You can use whatever feels appropriate for your case, but 'multiple coordinated views' is probably a safe bet. (2) Brushing and linking has a rather muddy history in data visualization literature.
Where to even start...
To give a bit more context, I'm interested in this topic because of my current work in genomics data visualization. So many genome browsers depend on linking, but the terminology is not really clear. For example, Higlass allows "view synchronization" (sharing pan and zoom) and rectangular brushes that define the extents of another view (this is called "viewport projection" in their docs). Similarly, Gosling implements locking of zoom and pan interactions between views, as well as linking views through a rectangular brush. Although Gosling's documentation calls it brushing and linking, it really only can be used to implement an overview and detail type of interface.
Since I'm interested in the genomics case, I thought I'd look at what I consider a seminal genomics visualization work: the MizBee paper from Miriah Meyer and co-authors.
“MizBee is the first synteny browser to provide linked views across the genome, chromosome, and block levels, allowing the user to maintain context across all of these levels when exploring conserved syntenic data.”
Linked views, got it.
“Their analysis needs motivated our highest level design decision of using multiple linked views [18], a visualization approach that is well suited for exploration of large datasets that have clearly defined levels of structure.”
This references a paper from 2007 by Roberts: 'State of the art: Coordinated & multiple views in exploratory visualization'. I saw this paper referenced all over the place in other works as well. Let's take a look.
First of all, right from the beginning there's a slight but confusing inconsistency:
"Multiple Coordinated Views is a specific exploratory visualization technique that enables users to explore their data."
"These interactive systems have at their heart the Coordinated and Multiple Views technique."
So which one is it? Multiple Coordinated Views or Coordinated and Multiple Views? When I see capitalization, I expect a proper noun with a concrete meaning.
Then we finally get to some definitions:
"The term multiple views is the general name used to describe any instance where data is represented in multiple windows. It usually implies that different representations are placed in subsequent windows and that operations on the views are coordinated, hence the longer term multiple coordinated views may be used."
There's also a mention of overview+detail and focus+context, as two (out of 6) variants of multiple views.
Let's consult the book
Of course, another place to start probably should have been THE book from Tamara Munzner. What are the definitions there?
Chapter 12 looks at faceting data into multiple views; the multiple views that are of interest to me are based on juxtaposition (placing side-by-side).
"Using multiple juxtaposed views involves many choices about how to coordinate between them to create linked views*."
The footnote gives some clarification:
"* Linked views, multiple views, coordinated views, coordinated multiple views, and coupled views are all synonyms for the same fundamental idea."
Got it: as I probably could have guessed, a lot of these terms are overlapping and don't necessarily imply any specific subtype of synchronization.
Next, the book established that there are 4 ways to realize the linking:
sharing visual encoding vs. different encoding
is highlighting linked?
same data vs. subset vs. each different set
is navigation synchronized?
Sharing encoding is pretty straightforward.
The second category mentions linked highlighting (aka brushing or cross-filtering) as a special case of sharing the visual encodings. It is described this way: you interactively select some items in one view and these items are highlighted ("using the same color") in all other views.
More context is given in the notes at the end of this chapter:
Linked highlighting was proposed at Bell Labs, where it was called brushing [Becker and Cleveland 87]; a chapter published 20 years later contains an in-depth discussion following up these ideas [Wills 08].
The third category mentions overview-detail, where one view shows an entire dataset and another view shows only a subset. Often, overview-detail combines shared visual encoding with "navigation support". An example given is of geographic maps, where the overview view is small compared to the detail view. Other terms mentioned in this section are: small multiples, detail-on-demand. This is exactly what Higlass and Gosling (which is based on Higlass) do.
Although this part of the book is pretty general, what could be added is that the selection "rectangle" can in general have any shape, or might even be somewhat procedural. But the rectangular brush is probably just mentioned in connection with the maps example. There's mention of unidirectional or bidirectional navigation linkage.
There's surprisingly little said about synchronizing navigation, with just a reference to an earlier chapter that goes in detail about the different ways of changing visualization by navigating.
Coordinated vs. Linked
So what's the deal with using 'coordinated' or 'linked'? My guess is that 'coordination' is a much more general term, while 'linking' can imply that there is almost a physical aspect to it: perhaps there are actual visible links between the elements.
That's just my speculation though, haven't found anybody discussing that (yet).
Brushing and linking
Although Tamara Munzner's textbook answers a lot of my questions, what about brushing and linking? Where did this term come from, specifically the "and linking" part? Is that a strict subcategory of Multiple Coordinated Views? Given the significance of Muzner's book, why isn't the term 'linked highlighting' used more instead?
As the textbook says, Becker and Cleveland 1987 indeed proposed the brushing interaction in scatterplots, but the specific term "brushing and linking" isn't featured in this paper. Similarly, also Wills' chapter on linked data views does not explicitly mention brushing and linking. The closest I can find are "brushing with linked highlighting" and "linked brushing" in other chapters of the same book.
At this point, I almost thought that I myself hallucinated that this exists as a significant term. However, the paper by Koytek et al. from 2018 actually uses it! They take a definition from another publication:
Hearst defined brushing and linking as “the connecting of two or more views of the same data, such that a change to the representation in one view affects the representation in the other views as well” [37].
The Hearst publication itself references two other works (Eick and Wills 1995 and Tweedie 1994) for brushing and linking as a "main information visualization technique", before giving the definition above. Although these two works describe brushing, neither of them establishes---or even uses---the full term 'brushing and linking'.
But I know that there are other papers that specifically use 'brushing and linking', just from a quick keyword search in titles and abstracts. Let's see which papers they cite for this technique:
Hajizadeh et al. (TVCG 2013):
'brushing and linking' cited with [Li and North 2003] and [North and Shneiderman 2000]
Chandrasegaran et al. (Int. J. Hum.-Comput. Stud 2017):
'brushing and linking' cited with [Li and North 2003]
Zhou et al. (TVCG 2021)
'brushing-and-linking' is used, but no citation is given
Doerr et al. (CGF 2024):
cites 'brushing and linking' with [Roberts 2007]
Quijano-Chavez et al. (VRW 2024):
cites 'brushing and linking' with [Buja 1991]
That's interesting! We know the Roberts 2007 paper: it discusses multiple coordinated views and brushing, but not 'brushing and linking' specifically.
Buja et al. 1991 describes the linked highlighting (aka brushing) technique. However, they again did not call it 'brushing and linking'! Instead, they use 'focusing and linking', and these are considered two separate techniques rather than a single method. Interestingly, the authors also use the term 'painting' for the action of selecting the items of interest.
That leaves us with the two papers with Chris North as co-author. First, Li and North 2003 compare two methods for filtering a visualization using another data view (sliders vs. brushing histograms). While interesting, the paper isn't giving any definitions useful for our investigation.
Much more interesting and relevant is the earlier paper: North and Shneiderman 2000. Here the authors actually establish what 'brushing and linking' means!
Brushing and linking: an exploratory data analysis (EDA) technique used when displaying a set of data items in multiple visualizations. When users select items in one visualization, those items are automatically highlighted in all the visualizations. A common example is brushing scatterplots (Becker & Cleveland, 1987).
So there we go, a definition of what brushing and linking means from within a data visualization community. There's a reference to the Brushing Scatterplots paper that we already know, but since we already discussed that that's not where the term came from, we can say that the North and Shneiderman 2000 paper is another origin of that naming.
Conclusion
The conclusion I take from this is that two papers sort of independently (at least without acknowledging each other) define brushing and linking as a term:
- 1.
Hearst's 1991 chapter from the Modern Information Retrieval is the earliest origin of this definition, based on two earlier examples that however didn't use that specific wording.
- 2.
North and Shneiderman 2000 formulated the term for data visualization audience.
I'm aware that this is a really surface-level literature survey and I might have missed some important paper or connections that will clear the confusion up. I might update or extend the post as I discover new things.
Overall, here's what I take from this whole exploration:
Multiple views, linked views, coordinate views, juxtaposed views etc. all have pretty much the same exact meaning. It's not that linked views means one thing and coordinated views something different.
I think the most general term should be multiple coordinated views, as a catch-all phrase that allows ways of synchronizing not only based on physical linking.
The origin of 'brushing and linking' is less clear than I expected. I take it that it has come up rather organically and unlike other concepts there's no single paper to cite.
It's still a little unclear what do the two parts of 'brushing and linking' refer to. Like my labmate noted, brushing can for some people be associated with just the act of selection (for example, based on https://d3js.org/d3-brush), requiring the 'and linking' part, or even better 'brushing with linked highlighting'. However, the papers that I discussed here sometimes seem to bundle both actions under 'brushing'. Gotta read up more to have a conclusive answer!
I do hope that somebody finds this useful. I remember when I was new to the vis community and people were throwing various terms and mantras around as if it's common knowledge. Well, turns out that sometimes the meaning can be pretty fuzzy. I found it fun to dig into past works and be pedantic a little.
If you read up to this point and have thoughts, feel free to reach out and correct me or point me in the right direction.
Disclaimer: This text was completely rawdogged, without the use of LLMs or any other tools beyond a system/browser-level spellchecker. Any grammatical or stylistic issues are left on purpose as a proof of this.