Pie charts are one of the most popular and commonly used charts not just in BI dashboards but in a variety of other areas as well. As school children, we have all learnt fractions by looking at pies sliced in various ways and decoding the ratio (quarters, half, three-quarters, etc) of each slice. In today’s blog we will look at why pie charts are not suitable for dashboards for a variety of reasons. However, before we delve further into this aspect it is important to go back into the history of pie charts and understand how they came to be used.
History of pie charts
The earliest use of a pie chart can be traced back to a publication in 1801 by William Playfair titled The Statistical Breviary. In this publication, this chart was used to depict the proportion of the Turkish Empire located in Asia (See figure below).
In 1858, the French engineer Charles Joseph Minard was the first to use pie charts in maps. The pie charts in the map below are used to represent the cattle sent from all around France for consumption in Paris.
Although William understood the intuitive ability for a sub-divided circle to represent part-to-whole relationships, he was not aware of the problems in visual perception that arise when we try and compare the parts of a pie chart.
Disadvantages of pie charts
Pie charts (a.k.a circle charts) decode quantitative values using two visual attributes:
- The area of the slice
- The angle formed by each slice at the center of the pie
However, both these visual attributes are not easy to compare. Our built-in powers of visual perception are great at identifying differences in two-dimensional locations and differences in line length. However, when it comes to identifying differences in 2-D areas and angles we fare quite poorly.
Here is a simple test to illustrate this. The figure below has two circles with different area. If the area of circle A is 1, what is the area of circle B?
It’s not easy to even come up with a reasonable estimate. In sample surveys, we get answers ranging from 6 to 50 whereas in reality the area of circle B is 16 times larger than circle A. Stephen Kosslyn in his book titled Graph Design for the Eye and Mind has explained how the perceived area is usually equal to the actual area raised to an exponent that is approximately 0.8 times a scaling constant. He goes on to state that in comparison to areas, differences between relative line lengths are perceived almost with 100% accuracy provided the lines are oriented in the same way.
A similar problem occurs even while dealing with relative angles made by the pie slices at the center of the chart. In the pie chart below try to order the slices in decreasing order of magnitude.
I am sure you will run into problems since comparing the angles of the slices in the above case is quite difficult. In her book Creating More Effective Graphs, Naomi Robbins has explained how our judgment of angles tends to get biased. In general, we tend to underestimate acute angles and overestimate obtuse angles. Also angles with horizontal bisectors appear larger than angles with vertical bisectors.
Here is another example that illustrates how the judgement of the slice magnitudes is impacted by the starting position of the slice angles. In the figure below it is quite easy to determine that the value of Grapefruit (the grey slice) is 25%.
However, in the figure below it is no longer easy to recognize that the grey slice contributes to 25% of the pie chart. Nothing has really changed in the figure below except that the slices have been sorted in decreasing order of size.
In general, it is easier to judge the magnitude of a slice when they are closer to 0%, 25%, 50%, 75% or 100%. Also it is easier to discern the magnitude of these slices when they start at the 3 o’clock, 6 o’clock, 9 o’clock or 12 o’clock positions as these 4 positions are quite familiar to our eyes and the mind. This explains why the size of the grey slice (value of Grapefruit) that was easy to judge in the first figure was more difficult to judge in the second figure.
One could of course argue that this problem can be easily solved by labeling the values of each of the slices as shown in the figure below.
Now in this pie chart notice the extra visual effort required to continuously shift the focus between the legend values on the right and the pie slices on the left to figure out which slice represents which fruit. Also what we have really done is added more labels to a pie chart to overcome the difficulty in reading the chart.
Alternatives to pie charts
Studies have shown that our judgement of relative line lengths such as the length of bars is very accurate. That’s why in the following figure we find it easier to compare the fruit values in the bar graph which displays the same data that was used by the pie chart.
Notice that the frequent context switching required between the legend and the slice values is absent while reading a bar chart. Also the need for assigning distinct colors to the category values does not exist in a bar chart since the data labels are placed adjacent to the bars.
In summary, pie charts are not appropriate for dashboards for the following reasons:
- There exist alternatives like bar charts that communicate insights about the data more effectively than pie charts.
- It is important for dashboards to be able to present information at a glance in a single screen without compromising clarity and efficiency. This requires the use of display media that exploits the principles of visual perception and guides us to make correct judgments about the data. Pie charts fail on many of these aspects.
Before my readers conclude that I am totally biased against pie charts, I would like to add that pie charts are not without their strengths. In a subsequent blog, we will see different use cases where pie charts can be used quite effectively. We will also look at some useful defaults that pie charts can use to improve their readability.
The faulty variance chart
In my earlier blog, I had posed a question about an Actual-to-Budget variance chart.
Although the purpose of this chart was to highlight the variance of actual revenues from budget, the chart displays them as separate lines thus making it difficult to detect the variance.
Here is the improved version of the variance chart. Here the budget values have been plotted as a 0% reference line while the actual values have been converted to the percentage difference between the budgeted and actual revenues. As the variance line (yellow line) moves above and below the budget line (blue line), this depicts the variance on a week-on-week basis much more clearly than the previous chart.
A Question for readers
The following grouped bar chart displays the month-wise revenue and cost for the 12 months of the year. But there seems to be an obvious mistake here. Can you spot it?