The ‘Before’ Chart
The chart above was shown to me by an ex-colleague of mine. Details were modified to protect the confidentiality of the original company. It displays six pie charts representing the KPI (key performance indicator) performance of departments of a division in August 2006. The chart shows the percentages of each department’s KPI performance in four categories represented by colour codes.
In KPI reporting, it is common to represent performance measures by three colour codes, red, amber, and green. Each of this colour code represents poor, average, and good performance, respectively. Poor, average, and good, are of course just general meanings for the colour codes. Organizations may use different terminologies at their discretion, such as under target, within tolerance, and on or over target for the colour codes, or any other terminologies. In the chart above, a fourth colour, grey, was added to represent KPIs that exist but are not reported for the particular month, as stated in the “note” at the end of the chart.
Identifying Design Choices
Let us try to identify the message behind this chart, the original intention of the chart maker. If we look at the chart from a holistic point of view, we would realize that the ultimate intention is to compare the six departments’ performance for a particular month. That is why six equal-looking 3D pie charts are displayed in one view.
There is no time-based information, such as months prior to August, as well as future predictions. We can infer that the chart maker only intends to focus on one particular month, in this case, August 2006.
Every department is measured by a different number of KPIs. This number is represented in two ways. One, as a total, situated in the top left of every pie chart. Two, as data label on the slices of each pie chart. A pie chart would display its data to show part-to-whole relationships, no matter if the labels are based on volume (number of KPI) or percentage (percentage of KPI). In this instance, since the medium of display is pie chart, we can infer that the chart maker intends to focus on the part-to-whole relationship, rather than the volume. In this case, the part-to-whole is represented by the percentages of the KPIs. We may assume that the number of KPI information is included for reference purposes.
A pie chart typically orders its dimension from the top going clockwise. Normally, a chart maker would list the most important dimension from the top. In this chart, we can observe that the dimensions are ordered from red, and then followed respectively by amber, green, and grey. We can infer that the poor performing KPIs (red) is the most important dimension that readers want to see, followed by the rest of the colour codes.
There are no dimension labels that readers can use to identify what the colour codes refer to. We can assume that the readers for whom this chart is presented for would be familiar with the KPI colour codes.
In summary, the chart maker’s intention for this chart is to compare the six departments’ KPI performance by displaying their parts to whole relationship – the percentages of KPI performance in each colour code dimensions.
Evaluating Design Choices
Let us now examine the design choices identified above and evaluate them with respect to the chart maker’s intention.
The first of this is the choice of 3D pie chart as the medium of display to compare the six departments’ performance. Consider the aspect of comparing KPI performance between different departments – what sort of things come to our mind? We need to be able to rank the performance of the departments in all KPI performance categories (the colour codes). A way to sort this ranking by category would be an additional bonus.
The pie charts provide an excellent view of identifying the size of the part-to-whole in each slice. For example, we can easily observe that in Department 6‘s chart, red, amber, and grey together make up 50% of the performance, without the need to combine the percentage values. This ability to immediately identify parts-to-whole data is the core strength of pie charts.
However, if we try to compare between two departments, this is where we may run into problems. For example, let us try to compare the red KPIs between Department 4 (33%) and Department 5 (29%). Visually, the two slices look identical in size. This is because the human brain is not so adept at comparing two-dimensional data, in this case, the areas represented by the slices. This is even further exacerbated by the use of 3D pie charts which distorts certain slices to appear to contain more volume than they actually are. Consider Department 1‘s green and grey slices. They visually look similar in size but of course each slice contains different values, 29% and 43%, respectively. The 3D angle has distorted our ability to comprehend their true sizes.
The data labels in each slices of the pie charts can lead to confusion. It is only after we study the chart for a bit more time that we realize that the data label is in the format “number of KPI, percentage of KPI“. There is no indicator on how to read the labels anywhere on the chart. This may seem like a trivial matter, however, when we present charts to our audience, it is imperative that they would be able to focus on the message of the chart as immediately possible. We do not want them to ponder on how to effectively read the chart.
I mentioned in the previous section that there are no dimension labels to identify the KPI colour code categories. I made the assumption that this is because the audience may be familiar with it. However, I feel that it may be wise to include dimension labels so that even layman readers would be able to understand the chart.
In summary, we need a chart that can display, rank, and order the KPI performance of the six departments in the four categories. We also need a better data label that clearly shows number and percentage of KPIs. Also, it may be wise to include dimension labels to the chart.
A Better Alternative
I have explained above that pie charts provide a poor way of comparing individual pie charts. Although pie charts are good for identification of parts-to-whole data, in this case, it would not be appropriate. This is because we also need to allow the data to be ranked and ordered for proper comparison to be made. A bar chart would be a more proper choice for our purpose.
Specifically, we need four bar charts for the four KPI categories. This would allow the performance in each category to be ordered and ranked exclusively while still retaining each category’s relationship with each other. Furthermore, comparison within each KPI category seems to be more important than comparison between KPI categories.
There are two flavours of bar charts, horizontal and vertical, as shown below.
We can infer from the original chart that the four KPI categories are relatively fixed. The departments, on the other hand, may be more flexible, since this chart may need to be generated for other divisions that may contain any number of departments.
If the data is to be displayed in a horizontal bar chart, the display may be too stretched for data with large size of departments, since the departments would be displayed in the x-axis of the horizontal bar chart. This in turn would make visual comparison more difficult. A vertical bar chart would provide the flexibility of adding more departments without losing the visual cohesiveness of the chart.
The chart would also need to have clear demarcation of the data labels, uniquely displaying percentage of KPI performance and number of KPIs. Since we can assume that the number of KPIs are there only for reference purposes, their label would not be as prominent on the chart. The visual measure (the bars of the bar chart) would measure the percentages of the KPIs.
We also need the dimension labels for layman readers. In my ‘after’ chart, I would label these labels as poor, average, good, and not reported.
The ‘After’ Chart
The chart above is my proposed improvement to the original chart. Readers can clearly see the separation of each KPI performance categories. More importantly, readers may be able to manually rank and order the data with greater ease compared to the pie chart.
The data labels are more clearly demarcated through the use of different vertical lines. The legend in the lower left provides an indication as to what the data labels mean.
There are also dimension labels that identify the performance categories of each of the four vertical bar charts.
The chart can be further improved through the use of sorting option, making the chart even more useful for the purpose of comparison. The animation below provides an example of how this would look.
In summary, as I have noted many times, always use the appropriate chart for the message you intend to deliver. In this example, since the message is to compare the departments’ performance, we need a chart that is suited for comparison, in this case, vertical bar charts.