Making Data Visual

Chapter 7. Case Study 1: Visualizing Telemetry to Improve Software

The previous chapters of this book explore the data counseling process: how to move from an ambiguous question to a more precise one, and how to refine a design through iteration into a final visualization. This description, however, has missed some of the twists and turns in the data counseling process. Data counseling goes through a series of iterations, each of which casts new light on the questions, but encounters dead ends. Ideas that seem insightful in a sketch turn out not to scale, or are not interpretable when used with real data. At each of these steps, the goal itself may change as new aspects come to light.

To see how this process can evolve in the real world, this chapter reviews a case study from a team that Danyel worked with at Microsoft. To protect sensitive information, this study obfuscates some images and details slightly. In addition, this telling reduces some of the complexity.

Introduction

One of Danyel’s roles at Microsoft is to consult with teams from the rest of the company about visualization. Jacqueline, who works on a data science team, emailed him a question: “How would we show distributions so that they pop for users?”

It can be hard to answer such a question without more context. What sort of distributions? Are they based on data, or are they abstract functions? What aspects of them should pop out? For example, Jacqueline might want to let her users see whether a given distribution ...