What are the limits of small multiples? At what point do the charts and graphs lose their individual meaning and blend back into a single, collective image/pattern? This is a quick experiment using bar graphs with random values to test the visual limits of small multiples.
A couple important caveats. I am not looking at various forms of colour blindness in this assessment, and, I am viewing these graphs on a laptop computer as opposed to a mobile/tablet or other size screen.
Lets take a look at the graphs in the simplest form and work up to multidimensional representations.
Single (1×1) charts with increasing data density
Starting the test by looking at single (1×1) charts with varying ranges of individual data points. The data in all cases are generated from Processing’s random() function.
Multiple (N x N) charts of varying data density.
3 x 3
This matrix of charts is very easy to visually assess. Each chart is The changing density spectrum for each chart is intentional to provide us another way of comparing charts in a small multiples test.
4 x 4
Again, at this level of density, each chart is visually accessible and relatable to its context.
7 x 7
This is starting to feel like the sweet spot. At this level, each chart is still distinct but our visual perception is beginning to cluster the charts by nearest neighbours and by rows and columns. It is not as easy to compare charts that are in different rows and columns.
11 x 11
From here on down the graphs are becoming more of a collective chart than individual graphs. Similarly to the 7 x 7 example, the individual data points in each graph are mostly distinguishable, they are not easy to read, interpret or compare. There is definitely a time and place for this level of density but only in cases where each graph contributes trend information, not unique data points.
18 x 18
This view is useful for seeing outliers or intra-graph patterning. Definitely a gestalt view here.
29 x 29
Aesthetically I like this quite a bit but the density has moved beyond even seeing trend data. This type of view is useful for conveying a sense of the magnitude of data being grappled with.
Observations and Ideas on the Limits of Small Multiples
It became clear to me at the ‘7 x 7′ level that we were finding a transition point between viewing individual graphs and views the graphs as a continuum or collection of graphs. And certainly by the point of looking at 18×18, each graph can only be seen for general qualitative features since the bars’ widths and heights are far too small to see. At this point, the colours and respective areas are more telling than specific features.
This exercise is a useful starting point for discussions with clients and stakeholders when talking about how to represent a large amount of data.
A few ideas for taking this further
Using randomly generated (and non structured randomness, not perlin) is useful here in order to focus on the chart representations and not the data but the next step is clearly to re-test this with real data. I’d like to see something like every NHL player’s stats represented in someway.
Test with greyscale and with various colour spectrums.
What impact does colour have, especially at higher dimensions of small multiples, eg, north of 7 x 7.
Animation, always animation
Does animating the charts make a difference? I’m guessing most definitely yes, and my assumption is that it will drive down dimensionality but that’s just a guess at this point.
Thanks for reading.
A few extras, having fun with background colours and styling
At this point it also becomes an aesthetic exercise. I really like the retro colour / feel of these versions.