(by Kaiser Fung)
A reading of the arguments by Kosara and Gelman/Urwin (PDF link) reveals one essential dividing line between statistical graphics and infoviz: the division of labor between the graphics creator and the reader. The infoviz designer expects readers to work the data while the creator of statistical graphics works on behalf of the reader.
To capture this difference, I would like to introduce a new metric for evaluating graphics: the return on effort. Effort, conceptually, is the amount of work a reader needs to do in order to understand the graph. Such work includes reading the axes, the scales, the legends, the text, and sizing up the columns, the bubbles or other forms, and understanding the context of the analysis, such as the population and the metrics, and grasping the general question being addressed, and so on. The return on effort is the amount of pleasure gained from the time spent on learning how to read the chart.
Even acknowledged “perfect” statistical graphics can require a high degree of effort. For example, a newcomer to the famous Napoleon’s march through Russia chart (Tufte’s version here) must take time to absorb the multiple dimensions and admire the efficiency of presentation. In this case, the reward justifies the level of effort, and the return on effort is high.
In the matrix shown below, the famous chart occupies the top right quadrant indicating high effort, high reward graphics.
Many statistical charts demand little of readers; the advice is to avoid low-reward charts like simple pie charts. The justly-celebrated Gapminder is a set of statistical charts with rich insights needing little effort. (or see my review here.) Focusing on Tufte’s data-ink ratio will generally push charts to the high reward half of the matrix.
By contrast, it appears that most infoviz charts fall into the high effort, low reward region. Kosara’s two charts (PDF link) require readers to interact with the data, and discover the hidden insights. The much circulated “themeriver” graphic of box office sales over time also places stringent demands on the reader who is intent on learning
something from the data.
I hope that as the field of infoviz develops, the graphics would move into the high reward region. The opportunity is there, as the availability of more data allows us to learn complex patterns, which in turn would require more sophisticated graphics. But readers will not be willing to expend effort if the reward is not sufficient.