Data visualisation: alive, visual, uncertain words

Last week, I very happily took on the challenge of being a discussant at the LSE Public Lecture titled “Data Visualisation: alive visual words”. The speaker was Federica Fragapane, an award-winning independent information designer 📊🎨

I got to know Federica in 2021, and since then I never stopped looking with awe at her visualisations. I felt she was using different shapes from what I got used to see, and with those she had the capacity of making me stick a bit longer to a caption, an article, a topic. We all know that concentrating is getting increasingly difficult in the era of social media, and I know that for me - spending so much time on my laptop - distractions are always a click away. Anyway, Federica’s visualisations hooked me, and with time I got to know her philosophy and ideas. So, when the Stats department at LSE was looking for possible Public Lecture proposals, I immediately thought about her.

Fast forward of one year, and it became reality. Federica went through many of her visualisations, talking about the construction of an alphabet with graphics (visual words) and the difficulty of people to read graphs (not a skill we are taught at school!). She explained how one of her clients defined her visualisations as “fragile”, because of the shapes she was using, and she made this term her own and kept on with the idea of using, when useful, organic shapes rather than purely geometric ones.

It was this idea of a fragile alphabet and fragile data that grew on me. I knew that I wanted to talk about uncertainty in data, cause I think it’s an important topic which is often overlooked. I knew that I had the technical alphabet to talk about it (after all, us Bayesian are always talking about uncertainty quantification!), but at the same time I knew I had to do a step further to talk about it to a non technical audience and to link it to Federica’s work, who did not use the shapes we are used to show, as statisticians, to communicate uncertainty. My discussion revolved around how to represent uncertainty in a quantitative way to explain how we should expect election polls to always have a % of error, and how we could leverage this error to represent more faithfully the various outcomes that we could expect from a complicated experiment such as elections (I fished one of 538 great representations from this article, the one with the possible simulated election outcomes accounting for polling errros). Sometimes, though, we cannot reliably quantify this uncertainty. For example, for fragile data - in this case, data that talk about delicate topics and/or whose collection is extremely hard to accomplish - we might want to give up on this quantification, but instead acknowledge their uncertainty by using qualitative representations. I gave an example using Federica’s work on Gaza, where she represents the change in population density in Rafah from October 2023 to February 2024. She uses soft, organic shapes that are inherently more difficult to compare (think how this would have looked as a bar plot!) but nevertheless pass a strong message. Also, she did not put any numbers in those visualisations! I believe this is also a respectful way to communicate that these data are extremely uncertain (the numbers were in the caption, of course).

Well, I guess I’ve have said enough! If you want to catch up on the full lecture you can do it on YouTube.

Thanks again to LSE and the Department of Statistics for this incredible opportunity, to Federica for the inspiring talk, to Prof. Irini Moustaki for chairing, and to Marta Foresti and her great LAGO Collective for the policy impact prospect. It was the first time I was challenging myself with something like this. It was a challenge, yet extremely fun and enriching (I got to think hard about a topic that is a bit out of my confort zone), and I am really looking forward to doing this again in the future.