Info & Interaction Design:
Vis Theory

Mick McQuaid

2023-11-27

Week FOURTEEN

grammar of graphics

Wilkinson (2005)

chapter 2, how to make a pie

The chapter is titled How to make a pie and it illustrates two different ways to make a pie chart. Wilkinson begins by noting that pie charts are ubiquitous and loved by managers while hated by statisticians, both without justification. They are just a way to visualize data.

One approach to making a pie is simple. Start with the information, do the process of making the pie chart, and render the result. If we follow this process, the emphasis is on the result and this is common but, according to Wilkinson, inadequate graphical practice.

Many programs have ways to select pie chart, then fill in the information, then see the pie. Wilkinson suggests that this is wrong. We should choose the pie much later in the process.

Instead, Wilkinson suggests this approach, taking quite a few steps before choosing the particular pie shape. Let’s review those steps.

(Review steps)

This picture only shows the ingredients and their dependencies. It does not tell how to assemble them.

Here is the data for our pie. The main conclusion we can draw from this data is that more females than males believe there is bias favoring males in peer review. It takes a while to arrive at this conclusion because there is a lot to read. It might be nice to have a quick way to see this.

The first process is to extract the data we need from the data as a whole. This is the process labeled “variables” in the data flow diagram and it takes us from data to a variable set.

The next process is called “algebra” and consists of crosstabulating, or joining the two subtables together, giving a more manageable variable set.

The next process is called “scales” and converts the variable set into integers. In practice, software usually does this for you but it is still an important step and you must at least know how software does it. For example, I have often seen people inadvertently come away from a software package with a counterintuitive scaling where 5 means none and 1 means frequently.

The next step is statistics and here we see the computation of the summary proportions or percentages.