Lee Wilkinson is most famous for his book, The Grammar of Graphics, whose ideas were implemented in the widely used R package ggplot2 as well as Tableau, the popular commercial graphics program, and all sorts of other places. Arguably as important as the book itself is its title. The idea that graphics has a “grammar”—that’s a real breakthrough. It’s related to ideas in statistics (graphs as comparisons, graphs as model checks, graphs as exploration and being surprised, with surprise being implicitly defined relative to expectations and modeling) and to ideas in computer science (here I’m not so familiar with the literature, but I’m thinking of the idea of a graph being defined not by what it looks like but rather by the steps used to create it). The modern era of statistical graphics was begun by Tukey in the 1960s and 1970s and continued by Cleveland, Tufte, and others in the 1980s. I think of Wilkinson as the key figure in the next wave of work in this area: once the general messages (exploratory data analysis is important; graphs can be clear and beautiful) had been absorbed, there was space for new ways of thinking about the process of creating statistical graphics. Along these lines, I think it was valuable that Lee straddled the worlds of academia and commerce.
I did not know Lee well—I’m not sure exactly how many times we actually spoke, but it was less than ten times—but I considered him a friend. We corresponded by email from time to time. He was a sincere and thoughtful person, both soft-spoken and lively, if that is possible. He could be critical (see for example here) but only because he cared so much about statistics and its applications that it bothered him when people were being annoyingly stupid. I think it’s safe to say that Lee’s ideas will continue to influence data analysis for many decades to come.
Good post as always, Andrew.
One package I quite enjoy that’s been inspired by the Grammar of Graphics is AlgebraOfGraphics.jl. If you take the formalist position in the philosophy of mathematics, that mathematics is just a special kind of grammar, AoG takes the logical next step of seeing what concepts we can borrow from mathematics to build better graphs. Not as big an advance as coming up with the grammar of graphics itself, but definitely the most innovative implementation I’ve seen.
In the package theme, I really like using plotnine: https://plotnine.readthedocs.io/en/stable/. Very good ggplot implementation for Python.
What a loss! His 1999 APA task force report on statistics, which strongly emphasizes the importance of depicting data in graphs — and depicting them properly –, is on the reading list in my lab. It’s a shame that APA never really came around to making sure that the recommendations are followed in their journals. And after I discovered SYSTAT during my time as a grad student in the 1990s (ancient times!), I never went back to SPSS. It was so enormoulsy helpful to be able to visualize data with just a few clicks or a line of syntax that was easy to memorize. It changed my whole approach to data analysis.
Kudos to Lee Wilkinson!
Glad to see you write something about Lee on his passing. I knew Lee while in graduate school. Interesting fellow. He had a divinity degree (or some such) before he started his PhD work, and while his PhD I believe is in Psychology, where I knew him from was work being done with a Professor in the School of Public Health. My first course with him was actually on Stochastic Processes, based on Chiang’s book. But where I really got to know him was he helped teach a course on multivariate statistics. While the course was based on Morrison’s text, the meat of he course was based on using a statistical system that he was developing that went on to become SYSTAT. I had taken a multivariate course in the stat department, and while I learned a lot of fancy math, I didn’t learn much on how to analyze data. SYSTAT was a revelation, particular if you either lived back then, or know about the 1970’s, and how limited computational resources were at the time. (The only other course that I took that came close was a clustering course that John Hartigan taught, because we had code that implemented the algorithms in his book).
Anyway, saddened to hear of his death.