The Best Tip for Effective Visualization Creation (MVP)

As data scientists we act as translators — turning large amounts of unstructured data into qualitative actionable items. Most people, however, don’t want to read your Python code or can easily gain insight from glancing at columns and rows. A good way to share your results is by creating an easy to follow and clearly labeled visualization.

Incremental Development

A practical tip I learned for creating effective visualizations, especially in a time crunch, is dream big, but start small and cover all the bases. Then continually improve your visual based on order of importance. That way you always have a ‘working’ finished product. This is commonly referred to as a minimum viable product.

A few years ago, Henrik Kniberg, drew this picture to describe misconceptions in agile development.

Notice under the “Like this!” heading, during the first iteration, a working product is presented that will get the customer from point A to B. They are not 100% satisfied, but what the developer gains from each iteration is usable feedback to make improvements. Under the “Not like this…” heading, the customer is first delivered a wheel, which they can do absolutely nothing with by itself. Once the car is finally produced, you can be sure that it's possibly full of flaws or bugs because the car has not been adequately tested yet.

Example

After collecting and analyzing data on student loan debt in the U.S., I wanted to create a graph that showed the relationship between the number of total borrowers based on their age as it corresponded to the amount of debt they owed. Take a look at some off my iterations before I got to my car.

My skateboard here has all the information I want to convey in a color separated, stacked bar chart. My title and axes are labeled and I have a legend. Is it easy to follow what’s going on?

For my intermediate iterations, I played around with different chart styling elements like size and color.

Finally, my car! You see I’m still conveying all the details I wanted my readers to understand. However, my graph is much more visually appealing without the generic colors or monotone color scale; it’s easier to see where one age range ends. I changed the axis labels and title so the reader better understands what they are looking at. I increased the size of the legend, added edge lines to the bars, labeled the subsections of each bar with the corresponding percent, increased the figure size to stretch the y-axis, and set the visibility of the top and right spines to false. I was satisfied with my skateboard, but I’m in love with my car.