A Design Problem

The following chart was found on www.sharebuilder.com. It shows a breakdown of asset holdings by sector for a popular ETF (exchange traded fund).

My Analysis

This display is bit of a hybrid. The left side serves as both a legend for the pie chart and as a table that can be used to look up the actual percentages of the sector allocations. The primary problem with this presentation is the pie chart.

To be useful, the pie chart should enable us to quickly see the basic allocation breakdown of the ETF. However, in order to understand which slices represent which sectors, we must look to the table. There are two problems with this design: (1) it requires us to do unnecessary work, looking back and forth, matching the pie slice colors to the small colored squares in the table to understand which slices belong to which sectors, and (2) because we must look up each slice on the table anyway, the pie chart is useless. Once you’ve read the sector allocation in the table, why bother to look at the corresponding slice in the pie?

In order to accurately decode the chart, the 2-D areas of the slices must correspond accurately to the allocation percentages. Below, I have labeled the slices in order from largest to smallest, based on the values in the table.

Take a close look. Do you see anything wrong with this picture? Slice 3 appears to be almost twice the size of slice 2. However, the values in the table indicate that slice 2 is 18.09% while slice 3 is just 14.17%. Similarly, although slice 7 appears smaller than slices 8 and 9 the values in the table reveals that it is actually slightly bigger. Also, slice 5 appears slightly smaller than slice 6, even though it’s supposed to be bigger. These misrepresentations could cause people to make misinformed decisions. These particular discrepancies are likely caused by a problem in Sharebuilder’s charting engine. However, even if the chart were rendered accurately and the distracting 3-D effect were removed, it would still be unnecessarily difficult to assess the sizes the pieces correctly. This is because humans have difficulty accurately comparing 2-D areas, such as the slices of the pie.

Another problem with this graph is less obvious.  Pie charts are used to show parts of a whole. The sum of the parts must add up to 100%. When I add all of the table’s allocation percentages together, I get 98.99%. I was ready to write this off as a rounding error, until I found the following graph for another ETF:

When the slices of a pie chart don’t add up to 100%, it’s misleading. The pie chart above suggests that this particular fund is allocated entirely to one sector. Yet when we review the table, we see that only 45.76% of the allocations are in this group. Where’s the rest of the data?

A Solution

Here is my redesign of the graph:

This simple design shows all the sectors and their corresponding allocations in a sorted table. Although the table data can be easily associated with the corresponding bars to determine each sector’s percentage, discerning individual values is not the purpose of the bars. Rather, they provide a means to rapidly compare the relative magnitudes of the values. People can quickly look at the overall shape of the bar chart and see if it’s relatively steep (as in this case), relatively even, or if it has one or two bars that tower over or are dwarfed by the rest, which provides a useful overview of the fund’s allocation percentages. If you were looking for a fund with a balanced distribution across many sectors, a quick glance at this graph would tell you keep looking. You could even scan an entire page or screen of these graphs to quickly identify funds that were allocated in a particular way. Notice finally that I added an “Other” category to show the remaining 1.01% needed to bring the total to 100%, assuming that the original graph excluded a set of sectors with very small percentages.

Where the original display’s two-column table design and misused graph made users work to find their desired information, the new design simplifies the data, allowing viewers to swiftly ascertain basic information as well as easily access the precise values, if desired. And I didn’t even charge Sharebuilder anything for my services!