5 Minute EDA: Practical Use Case for 3D Visualizations
They look pretty cool but are they useful?
What is a 3D visualization?
A 3D visualization adds an added layer of dimensionality to a 2D visual. For example, 3D plots can often be used to show a relationship across three variables. While most conventional plots displays the relationship between two (plotted on the x and y axis), 3D plots leverage three axes (x, y, and z). However, there is also a growing trend to displaying 2D data in 3D format for pure aesthetics.
The debate on 3D visualizations— how useful are they?
Many data analysts and data scientists dislike 3D visualizations. Why? In one blog article titled “Should You Use 3D Charts?” the author writes:
“The short answer is NO. With all the magical tools available at our disposal, it is very easy to create awesomely complex looking charts and graphs that add no value. 3D charts are one of them.”
In another blog that explores 3D graphics in more detail, the author walks you through examples of cases where turning a perfectly informative 2D visual into 3D can make things rather complicated, confusing, and unnecessary.
For example, here is an example of a 3D bar graph that is referenced in the blog:
At first glance, it looks kind of pretty. However, it is rather difficult to read the precise fluorescent intensity metric for each bar. Some bars are hidden behind others, and the shadows of a bar appears to have a slightly elevated value compared to the non-shadowed portion. This is a case where a 2D visual would have conveyed the same information — perhaps more clearly — than a 3D graph.
The general sentiment shared amongst the data science community is that in most instances, 3D graphics create confusion rather than to add valuable insight.
When is a 3D visualization useful?
I think there are two instances in which it would make sense to use 3D. They are:
- Multivariate analysis (to see how clusters are formed)
- Surface plotting (to display the coordinates of a 3D object)
**However, both cases should ideally have an interactive feature that allows the user to rotate the visual as an animation.
Multivariate Analysis
When attempting to discover the possible correlation of clusters across three dimensions, 3D plots can be quite useful. Here is a blog that highlights this argument well.
However, even this is not always a good idea, because unless the graph is interactive like the one shown above, it can project misleading information depending on the angle of how the visualization is presented.
I would much rather recommend remaining with a 2D plot, plotting each variable of the third axis on a separate graph. Another alternative would be to play with the color or size of the dots to convey further labels of a third variable.
Surface Plot
Displaying a 3 dimensional object is perhaps the most obvious use case for using 3D visualization. If you wanted to highlight the surface area of a mountain, for example, a 3D plot makes the most sense.
Conclusion
Before incorporating a 3D visualization into your analysis, it may be a good idea to first assess whether the same level of insight can be conveyed just as well (if not better) using 2D. While visualizations can be fun, if the purpose is to convey important information, it’s always best to stick to the simplest and most straight-forward methods.
This is part of my 5-minute EDA series, where I run quick exploratory data analysis on an interesting dataset. Thanks for reading!
