The Box Plot Lesson
Early in my career, I put together an analysis for a client. Box plots. Violin plots. Not, by any reasonable standard, the most complicated visualizations in the world. I was confident the analysis was solid.
I showed up. I shared the slides. I got blank stares.
The room had no idea what to do with the box plots. The feedback was direct: "Too complicated. We just need histograms or bar charts." We scrambled, rebuilt the analysis with the simplest possible visualizations, presented again — and then the insight landed and the recommendation was accepted.
The Lesson
The lesson is not that box plots are bad. Box plots are perfectly good visualizations. The lesson is that the right visualization depends on your audience, not just on your data. I had picked a chart for myself, not for the room — and in doing so, I had buried the insight in a format the decision-makers couldn't read.
Almost every data scientist I know has at least one version of this story. The mark of experience is not that you stop having these moments — it's that you start anticipating them.
Before you pick a visualization, ask three questions. The answers determine the chart, not your personal preference or the sophistication of the analysis.
- Who is looking at this? A data scientist? An executive? A regulator? A clinician? A patient? Your default visualization should be calibrated to their visual literacy.
- What decision are they trying to make? A board approving a major investment needs the headline number front-and-center. A scientist reviewing a study needs the full distribution. A clinician making a treatment decision needs the patient's specific data, not aggregate trends.
- What do they already understand about plots? If the audience has never seen a box plot in their life, throwing one at them is malpractice. If the audience reads scientific papers all day, a bar chart of means with no error bars will annoy them.
Audience Defaults
- Technical and statistical — full toolkit: violin plots, pair plots, faceted plots, log scales, complex multi-panel figures.
- Executive and product — default to bar charts, line plots, histograms. Simplify aggressively. One insight per chart.
- Regulatory and compliance — ask first. There are often field-specific conventions you're expected to follow.
- Clinical — ask first. Medicine has its own conventions (e.g., Kaplan–Meier curves for survival analysis), and using the right one matters for credibility.
The same support-team response-time data shown as a bar chart (executive view) and a box plot (technical view). Toggle between views to see how the packaging changes what's communicated.
You've built a careful analysis using violin plots and pair plots. Your audience is the company's executive team, none of whom have a statistics background. What should you do?