Familiar. Collaborative. Managed.
Author: Jenifer Underwood, Sr. Technical Product Manager, Microsoft SQL Server Business Intelligence
In this new world of exploding data volumes, the ability to make sense of all this data and effectively communicate insights from it is a highly valued skillset. Communicating trustworthy insights includes choosing the appropriate data visualizations to tell a story or make a key point. That may seem trivial at first, but in fact, it is quite powerful. In some fields such as research, healthcare or military, the use of data and visualizations has specific guidelines since misinterpretations could impact human lives. Most of the time getting data visualizations right is not a life and death matter, but it is important. There are several highly-regarded thought leaders with excellent reading material on this topic, including Stephen Few and Edward Tufte. If you have not read any of their books and you are in an analytics/business intelligence profession, consider this a “must do” before you build another report or dashboard. In the meantime, here are a few of the most common mistakes and some best practices to keep in mind.
I confess that I love pie charts…but they should very rarely be used. I do violate this rule to showcase Excel Power View pie charts and the new cross-filtered pie charts. These visualizations are stunning and can be useful in the right scenario. The product teams at Microsoft tried several times historically to avoid pie charts to help ensure best practices were used with their products. Alas, they caved in each time to the general public pressure to add pie charts and rely on your knowledge to make the right choice of when to use them. Most often, pie charts are misused to communicate part-to-whole scenarios where line or bar charts would be much more effective.
3D charts are best used in the scientific community for seeing patterns in n-dimensional data. Do not use 3D bar charts just for the sake of doing so because you think they look cool. 3D charts can easily be misinterpreted and often distort 2D data. The best practice is to keep your charts simple.
I also admit that I adore bright colors. My clothing and accessories typically are bright and cheery! Bright colors are fun. However, when it comes to reporting, it is a best practice not to use bright or highly saturated colors, but instead to tone it down a bit. There are a few reasons why, but top of mind, lighter colors are much easier on the eyes, and they also showcase better when presentations are displayed on projectors. You can certainly use a bright color to highlight a specific condition but don’t overdo it.
More about Colors
Another point to keep in mind with regards to color is that more people than you realize are color blind. I am lucky not to be colorblind, but many of my peers are. The most common color blindness is an inability to distinguish reds and greens. That is a pretty important point for reporting performance or conditions – your reader might not be able to detect a difference or interpret the color! Color Lab is an online resource that you can refer to for color palettes that are friendly for color blindness.
Another tip I learned from the healthcare industry is to also use different shape icons along with colors to display status visually. A color blind person can be trained on the meaning of a circle or triangle for deciphering status easily. With KPIs, there are usually options to choose from various indicators. Keep these points in mind when choosing how to effectively communicate KPI performance.
Sparklines, Data Bars and Indicators
While on the topic of KPIs, let me quickly note that a KPI typically communicates the here and now, but it does not effectively showcase historical performance or trends. To add context to your KPI, it is a best practice to supplement it with sparklines. Sparklines are data-intense, design-simple, word-size graphics that provide a quick sense of historical context. When designing sparklines in reports, it is helpful to also highlight the minimum point and the maximum point.
Trellis charts are a small series of charts that much like sparklines also provide a very fast visual comparison of trends over time periods. It is best to keep the time axis on the bottom of trellis charts to not confuse your audience.
Scatter plots are great options for displaying relationships between two quantitative variables, even with exceptionally large sets of data. A trellis chart of scatter charts can further reveal hidden patterns across many different attributes at the same time. Taking analysis one step further, Power View has the fantastic capability to play scatter chart visualizations over time periods. This enables you to see variable interactions change over time.
Best practices around scatter plots include removing fill color where possible, visually identifying groups when multiple groups are plotted together (shapes, images, shades of color), displaying trend lines and using trellis charts to reduce complexity.
Maps are fantastic for understanding geospatial context in reporting. In the real world, I often see thematic maps used with red, yellow and green colors. My color blind peers can’t decipher those reports. You can use thematic maps with colors if you appropriately use light and dark variations of a color, along with a color scale or consider doing an overlap with a bar chart on a map. If you do choose map overlays, be sure not to clutter the map. Simple maps with light colors are better than the dark color or detailed road maps for reporting, unless you need to see roads to understand context of the data. Excel Power View mapping does allow you to toggle between various map views; some are optimized for reporting, and also to drill down to zoom in and narrow the map report scope to a specific location of interest. In the map below images below, it is much easier to visually detect the value in Arizona on the lighter greyscale map since the orange color blends right into the map tan colors.
There are many other chart types, network diagrams, treemaps and infographics. Best practices typically tell you to avoid the other visualization choices. However, in the big data world I am seeing that some creative, alternative data visualizations that break the traditional rules can be quite effective. If you do need a visualization that is not “in the box”, don’t forget about Visio. Visio and Visio Services in SharePoint can be used with data sources for unlimited visualization possibilities. Another great new option with Office 2013 is Office Apps. With Office Apps you can now add your own data visualizations into Excel or Office. I am genuinely excited about the potential of Office Apps in the new data analytics world. I will be building some myself soon to see how these can be used in the new wave of SharePoint 2013, Office 365/Office 2013 and SQL Server BI to really take reporting to the next level.
I know that I barely scratched upon the surface here with regards to data visualization choices and best practices. I hope this was enough to inspire you to take another look at your data visualization choices and the endless possibilities if you need to break outside the box. To review more information on this hot topic, feel free to download the detailed presentation here.
Jen, loved the post. Your a kindred spirit and this is a very timely topic with so much happening in the BI space, our users need guidance. The more BI and visualization work I do, the more I believe that "more is not better". Reduce the noise and use visuals that effectively convey the important messages. Thanks
This post goes into the right direction after years and years of bad charts. However, some of the examples you used are not appropriate from my point of view. The Trellis charts on the left as well as some of your scatter plot examples lack of simplicity. I mean what's the purpose of visualizing data if you cannot grasp the meaning / message immediately? Still a good read, Jenifer.
Jen...nice introduction to charting and visualizations. I tend to follow the motto of "just because you can doesn't necessarily mean you should" which falls in line with what Paul said below of "more is not better". I would say that pie charts aren't all that bad...it's just how you use them... :)
Read up Cleveland and use R with ggplot2. Also, you might get the wonks, wherever they are, to integrate with R a la PL/R (that's with Postgres).
This article should be required reading for every data analyst and presenter!
Like the blog. Have you ever heard of the SUCCESS-Rule? www.hichert.com/.../success
Jen, I miss reading your insightful contributions to the MSFT internal aliases. You always had something great to offer!
And this blog post is just as awesome.