In 2019, I gave five talks about data visualization. This was great fun, but as I looked back, I felt compelled to confront a difficult question: what are we as actuaries still not getting about this? Haven’t actuaries embraced data visualization? We understand that it’s important. We understand that it’s an essential element of communication. We’ve learned how to avoid some common data visualization pitfalls (I don’t see nearly as many 3-D bar plots as I used to). What’s left for us to change?
Earlier this year, I got an answer. At the 2020 CAS Spring Meeting I experienced the CAS virtual meeting platform. One of the great features is the ability of attendees to make comments in a chat window as the presentation is taking place. One of the attendees said that they had provided data visualization for their boss who replied with, “Next time just give me the numbers.”
I don’t need to remind anyone that this is a silly sentiment. Still, everyone repeat after me: “Numbers may be expressed visually.” So, what was the disconnect? There may have been a lack of comprehension towards what was on the page, or maybe difficulty approaching numeric values that were not expressed in Arabic base 10. Data visualization is often a more interactive and colorful approach; maybe the nontraditional presentation threw this person off. Whatever the reason, they are hardly alone. I’ve heard many similar stories.
So what can we do? We can continue the process of discussion with our clients and colleagues to help them understand the value of data visualization. But first, let’s assess our own relationship with the tool. Do we visualize at all stages of our work, or just near the end when we’re preparing a report? Do we create plots and figures because they are an integral part of our understanding of data, or do we do it because we think someone else is expecting it? Do we spend about as much time considering the type of plot as we do the type of model to use? Do we adjust the scale of our axes with as much attention as we think about model parameters? Does a visual merely represent a model, or does it inform a model?
What I’m trying to impart is that it’s time to take visualization seriously. If it ever was, data visualization is no longer a “nice to have.” It is as much a part of the process as data wrangling, or building and evaluating a model. It’s not a pretty picture to help explain the analysis. Very often it is the analysis. Visualization is a tool to help non-actuaries (and actuaries alike) understand a model.
And let’s retire some of the ideas about how data visualization is subjective. I mean, it is. We can alter scales, pixels, colors, symbols and layouts to mangle a message. But tools like predictive analysis have similar vulnerabilities. Changing the link function of a GLM, handling outliers in sample data, selection of a weighting function for variable process: these are all elements which figure significantly in forward-looking estimates. Actuaries have developed and adopted a rich set of tools to mitigate these risks. The syllabus provides practical and detailed guidance about when modeling decisions can go wrong. The data visualization literature is no different in this regard.
So let’s get serious. The best bit? Fun needn’t be the opposite of serious. Data visualization is fun, but so is building a model. Be sure to follow along as CAS explores the many facets of data visualization this month with the hashtag #DYKdataviz.
Good thinking Brian. I like your argument that visualizations and models both contain subjective elements. Rather than use “Its too subjective” to write off the medium, we can treat it as a skill that can be managed and improved. If we can spend time getting better at modelling, we can also spend time getting better at data visualization.
“Just give me the numbers” illustrates how some consumers of numbers want it both (or perhaps, more accurately, neither) ways. That is, they may lament what they perceive as scientists’ lack of business acumen and social graces, but are equally devastated when confronted with sleeker and more powerful methods of communicating results. The latter may derive in part from a mistrust of scientific views, where the visualization becomes a perceived wall between the number consumer and the perceived truth. Visualizations delivered as interfaces may ease some of the friction around more artful delivery by allowing the consumer to reshape the visual according to varied assumptions.