Boschi, Tobia Ravà.

Why VR is, and isn’t, the future of Data Viz

As non-gaming applications of VR* continue to gain attention, data visualization has become a topic of some debate. Advocates claim that VR is going to revolutionize how we work with data, while critics maintain that nobody will ever strap on a headset to do something they can already accomplish in a spreadsheet.

At Kineviz, we agree with both camps. We believe VR’s true potential lies not in replacing traditional tools, but in creating new ones. We’ve reached this conclusion by focusing on use cases.

Broadly speaking, we see data visualization falling into two categories — asking questions and telling stories. Either you’re using graphs as a tool to understand something better, or you’re using them to communicate what you’ve learned.

Self Description, XKCD

People have been successfully asking and telling without VR for a very long time. The data-driven world we live in today is a testament to the effectiveness of existing methods. Spreadsheets, dashboards, and infographics (not to mention paper and whiteboards) allow us to derive insight from many data sets and communicate that insight quickly and clearly. If you’re trying to figure out which states voted which way or trace the fluctuation of a stock’s price, VR is probably going to make things harder to see.

But what about things that are already hard to see? High dimensional data sets, relationship (not to be confused with relational) data, and heterogeneous data bump up against the limits of 2D visualization. Think about how you’d represent change over time across 100k entities with 50 dimensions. A single 2D graph, or even a single dashboard, just won’t cut it. Even 3Ds graph rendered on 2D screens are of limited value; they’re visually too dense to parse.

Mouse Bone Marrow data set courtesy Nikolay Samusik

This is the kind of problem where VR becomes a force multiplier. Kineviz was recently charged with streamlining a technique where our client had to plot dozens of 2d graphs to analyze a single sample; doing so took upwards of 2 hours. Part of the difficulty lay in simply keeping track of how the plots relate to one another. Bringing the data into VR shifted the challenge of pattern recognition from a cognitive one, requiring an analyst to mentally model the relationship between 2d plots, to a visual one, where relationships become easy to spot as volumes.

The result: in VR, the time needed to analyze a sample dropped from over 2 hours to under 10 minutes. The number of plots per sample was reduced 8-fold.

Efficiency gains are not the only case for bringing data into VR. Straddling the line between simulation and abstraction holds promise in physical domains like medicine and construction. VR’s power to immerse and engage users offers accessibility to audiences without a data science background. As multi-user experiences in VR improve, collaborating in a shared data environment is another compelling scenario.

At least, these are some of the hunches we’re following at Kineviz. What’s needed now is more experimentation. We know for sure that VR won’t replace bar charts in the data scientist’s tool kit. But we have reason to believe it will become just as integral.

*While I refer exclusively to VR throughout this post, much of what I’ve written is equally applicable to AR. In fact, Kineviz is developing hybrid VR/AR visualization technology. I’m an advocate of XR, extended reality, as an umbrella term for both.