2.3 Data Visualization Theory

First, "minimize the ink." In other words, reduce unnecessary lines, color, graphics, or anything else that doesn't directly help the ability internalize and interpret the data. This makes for very "clean" visualizations. Examine the dashboard below for search engine optimization efforts. There are no graph lines or unnecessary axes. Just the KPIs and very simple text. Remember, if it's unneccesary, remove it.

Along these same lines, do not use 3D effects or fancy patterns unless they directly improve the ability to internalize the data. Just because Excel or Tableau can do it, doesn't mean you should. See below. This chart was created in Excel.

Next, what type of chart should you use for each metric? Well, it's often self-explanatory, but there's some decent advice here if you want it; line charts for data over time, bar charts for categories, etc. However, what we will share with you (based on Tufte's research) is that you should avoid pie charts. Research has shown that pie charts are notorious for skewing our perception of data. However, donut charts are even worse. Donut charts are the devil. If you should avoid pie charts as much as possible, you should run away at the sight of donut charts. They are even worse when it comes to promoting misinterpretations of data. So why do people use them? Because they look cool. Never trade off the ability accurately and efficiently process data visualizations for "cool" when it comes to organizational dashboards.

This leads to an interesting topic/question: How do people misinterpret data? And how can you "lie" with visualizations? Well, other than the sub-conscious biases our brain suffers from when examining certian types of charts, chart-makers can also intentionally (and unintentionally) leave out important contextual information that significantly skews, or flat-out lies about, the data interepretation. Tufte gives the following good examples (Tufte, Edward, and P. Graves-Morris. "The visual display of quantitative information.; 1983." (2014).):

How are these charts above deceptive to the consumer? Well, the first chart indicates that the percent of all family doctors in California in 1964 was 27 percent. In 1975, it was 16.8%. If you do the math, then that middle doctor image should be 62.2 percent as large as the 1964 doctor. However, if you "chop up" that 1975 doctor into pieces, how many of those 1975 doctors would it take to fill the space occupied by the 1964 doctor? About 4; which should indicate that the percent of family doctors shrunk to 25 percent of what it was in 1964 by 1975. That is false. These images give the impression that the percent of family doctors is shrinking at a much more alarming rate then they truly are. This is because the graphic image used to represent the precent of family doctors is two-dimensional. If As a result, if you want to maintain proportions, then if you shrink the height to 16.8 percent in 1975, then the width is automatically adjusted as well--giving it the appearance of shrinking much faster than it truly is. The same effect is happening in the second image representing the purchasing power of the dollar.

How else can you lie with a chart? Well, take a look at the two bar charts below. They have followed the "minimize the ink" guidline and they are very simple. But the first chart makes profit margin appear to be increasing more rapidly than it truly is. Why? Because the X-axis starts at 5% rather than 0. But you don't know that because the axis has been removed. To be clear, sometimes it's a very good idea to adjust the X-axis to highlight differences. But make sure you clearly identify where the axis begins if you do that.

Next, take a look at the stock prices in the chart below for ABCorp and XYZCom. If you could go back in time and invest $1000, which company would you invest in?

The answer is that it doesn't matter which you invest in; both return the same rate becasue both are increasing at the same rate. Whether you buy 10 shares of ABCorp at $10 per share or 100 shares of XYZCom at $1 per share, you end up with $1000 either way. So how should this chart be modified? The Y-axis should be percent change from year over year rather than actual stock price.

So now you know how to make a decent chart. Next, how should charts be arranged on a dashboard. This is fairly simple and it has to do with how our brains are programmed to process information. Depending on the culture, we read from left to right and top to bottom. Therefore, the most important information (tpyically 1-2 most important KPIs) should go at the top-left. Why waste that space with a brand logo? You already know who you work for so use the space for something useful. Next, arrange the information in the order that they should be interpreted in; tell a story. For example, arrange measures and charts in this order: 1) What is total revenue?, 2) How much of that is net the cost of goods sold?, 3) How much is overhead? 4) Operating expenses? 5) What are the predicted returns for the same period?

Next, you know that you need to keep dashboards simple. But you now also know that they can be "too" simple (e.g. not reporting the axis start value). So how much information should you provide about a measure? In other words, how much context should you add? It needs to be enough to properly evaluate a measure. For example, let's say the primary KPI of a dashboard is total revenue and it's currently reported at $1,231,990. Obviously, there is some context you MUST have. Is this today's revenue? This week's revenue? But you can also provide other contextual information to compare this to. For example, how does this compare to yesterday's revenue? What about this day last year? What about the last time we used this same promotion that is going on today? What about the last time we issued a new product to market like we did today? Find the right amount of context that doesn't clutter the dashboard or begin to inhibit the ability to focus on the KPIs themeselves.

Lastly, dashboards need to change over time. If you're not constantly trying out new metrics and charts, then you're going to miss something; because your competitors likely are. So how do you introduce new metrics and charts on a dashboard? Generally speaking, we introduce new charts in the lower right portion of the dashboard. We offer help in understanding them (e.g. clickable tool tips) . And, we track their usage (e.g. record clicks). If it's a new metric that supports an existing KPI, then use a smaller font and place it in the lower-right portion of the space occupied by the primary KPI. Notice in the image below how the number 1980--the target goal for new customers introduced in the period--appears in a smaller font and different color in the lower-right portion of the space occupied by the large KPI in the top-left of the dashboard indicating the actual number of new customers earned.