I’ve come to get the sense in the data visualisation world, and in fact in almost all aspects of life, that we are too used to accepting what people tell us as gospel.
When I talk about this in relation to data visualisation, I mean in scenarios where best practices are discussed, and we fail to challenge the underlying theory as to why.
We’ve all been told ‘never use pie charts of more than 3 segments’ or ‘never use circle size for comparisons’, the reason we often receive following these comments is extremely high level and along the lines of ‘well its because humans are poor at interpreting x’ or ‘well its because humans find it difficult to perceive x’.
Well last Thursday, during a London Tableau User Group and a critique of a visualisation (shown in the image below) I got told one too many times that ‘we shouldn’t use circles for comparisons’, without an in depth discussion as to why (I just want to clarify at this point I am not criticising the individual that was presenting on this topic, it would have been a bit much if they had gone on a 10 minute explanation as to why in front of a group of 100 people).
I then spent the rest of the evening googling, looking for a proper theoretical reason as to why. Some actual research evidence if you will.
And this is where I stumbled on the notion of ‘Visual Crowding’.
Whilst visual crowding was not necessarily the immediate answer I wanted in regards to ‘circle comparison theory’, it was still a very interesting read and something that we should be more aware of in the data visualisation space.
The term ‘Visual Crowding’ refers to our inability to recognise objects in clutter, which sets a fundamental limit on conscious visual perception and object recognition throughout most of the visual field.
Well that’s a proper definition, in ‘normal speak’, clutter affects how we perceive objects in a space.
At present crowding effects are seen to have a detrimental affect on our ability to determine the orientation (Andriessen & Bouma, 1976; Parkes et al., 2001), size and colour (saturation and hue) of an object (van den Berg, Roerdink, & Cornelissen, 2007).
Of course, this has several practical implications for any visualisation designers reading this post.
I’m not going to give a comprehensive overview of the literature on this subject. My time writing 5,000 word uni essays is over. But I will give a quick summary of one specific paper on the subject area and the conclusion that they came too.
The authors, Ronald van den Berg, Jos Roerdink and Frans Cornelissen used a series of visual tests and assessed user responses. These tests look to assess how by changing the degree (distance between the objects) and eccentricity (location in our visual field) affected recognition.
The results given by the study clearly identify an affect on our ability to distinguish objects with the change to spacing and eccentricity. The above visualisation shows that for all encoding attributes, the ability to correctly determine their comparable size begins to diminish once passed a ‘critical spacing threshold’.
The flat line you see on the above chart shows that when eccentricity is low (i.e. directly in our line of site) there is little significant change in our ability to distinguish differences between objects. The authors hypothesised that a low level of eccentricity led to crowding effects being small because they were obscured as a result of the low ‘signal-to-noise’ ratio.
Further assessment of the results (visualised in the image below) shows that the affect is much smaller (though still important) when reviewing hue and saturation encoding (though they themselves are comparable) verses orientation and size encoding (again, they themselves are comparable).
So what implications does this have to us? Well as ‘data visualisation people’ we need to be aware of how human perception is limited as a result of crowding. How we should think more carefully before using size and colour encoding, especially as the number of marks (and as a result the level of crowding) increases.
Like all other aspects of our visualisation, we need to make decisions on what to add and what to remove, and perhaps we need to implement some sort of plan for making encoding decisions.
Of course, there is still an issue, how do we define whether a visualisation is ‘crowded’, here I have referred to a crowded visualisation as one with many marks, but many marks doesn’t always mean a crowded visualisation, a concentrated area of any number of marks can lead to a crowded visualisation. Is there a way of giving your visualisation a crowding score? Possibly and this is something I plan to look at next.
I will finish with a one line summary.
Think colour shelf first, then size; and only ever encode if appropriate.
Andriessen J. J. Bouma H. (1976). Eccentric vision: Adverse interactions between line segments. Vision Research, 16, 71–78, doi:10.1016/0042-6989(76)90078-X.
Parkes L. Lund J. Angelucci A. Solomon J. A. Morgan M. (2001). Compulsory averaging of crowded orientation signals in human vision. Nature Neuroscience, 4, 739–744, doi:10.1038/89532 .
van den Berg R. Roerdink J. B. T. M. Cornelissen F. W. (2007). On the generality of crowding: Visual crowding in size, saturation, and hue compared to orientation. Journal of Vision, 7(2):14, 1–11,http://www.journalofvision.org/content/7/2/14, doi:10.1167/7.2.14.