A Rant About The Ridiculous HBAR3D, VBAR3D and PIE3D Statements in PROC GCHART
To the developers of SAS PROC GCHART: Have you lost your minds?
I can think of two possible reasons to include such ridiculous statements:
poor abysmal statistical practice. In fact, I (and many others) have never seen a three-dimensional bar chart or pie chart that was more effective than its two-dimensional counterpart. Thus, we are left with reason number 2. Hardly a compelling reason — no one considers Excel to be a highly regarded way to analyze data — let’s emulate that other package that statisticians deplore; although Excel is a very popular way to analyze data.
What got me started on this particular rant is that I saw the latest VIEWS Newsletter (.pdf), in which Leroy Bessler pointed out how misleading a three-dimensional pie chart (Bessler’s Figure 1) was, and how a two-dimensional pie (Figure 2) chart fixed the problem. I agree. But then Bessler goes on to critize a three-dimensional bar chart (Figure 3), recommending another three-dimensional bar chart (Figure 4 shown below) in its place.
Sadly, I have to disagree with Bessler. His Figure 4 is not a good chart either. It contains useless information and adds structure to the plot, but does not provide any more insight into the data compared to the two dimensional version of the same plot. Tufte speaks of the “data-ink ratio”, that you want it as high as possible, but the additional ink and structure to make it three dimensional adds no more information to the plot. The third dimension interferes with the transmission of information — your eye sees how the colors change in the bar light to dark from top to bottom; it notices the ovals at the right edge of the bar that give the plot its three-dimensional appearance. The ovals at the right of each bar would also prevent you from accurately reading the height on the bar on an axis (although there is no axis in this plot). Now I admit that for this simple bar chart, making it three-dimensional certainly is not a fatal flaw, but it still violates the principles of a well-designed chart. The third dimension is unnecessary and in other plots might be potentially confusing (see below). It is like trying to explain what happened at your child’s soccer game while running around the room playing air guitar like a rock star. It interferes with the transmission of the information. It is like putting a Bart Simpson background on a plot. It interferes with the transmission of information.
Consider this ridiculous and atrocious three-dimensional bar chart from the PROC GCHART documentation (click on the plot to see it full size):
Wainer, in his classic article “How To Display Data Badly”, points out that the human eye registers the area of a colored region, and the three-dimensional nature of the plot adds to the area, in defiance of the area needed to be portrayed by the data. The tops of the bars, visible on only the topmost of the groups, also adds area to this one group, and the fact that the bars overlap left-to-right subtracts some areas. All of this extra area deceives the eye into a visual assessment that is at odds with the data. All of these things, plus the three-dimensional background, interfere with the proper understanding of the meaning of the chart. Wainer, Tufte and many others implore you to create plots where the meaning is obvious; instead of creating plots that use “glitzy” features of your software.
I have a plea — no, a command — to all GVSUG members. Do not use VBAR3D, HBAR3D or PIE3D, or any other type of three-dimensional pie or bar chart. It simply is unacceptable. Using the powers vested in me by GVSUG, if I catch you using one of these plots, we will have you condemned to eternal damnation, you will be hung, drawn-and-quartered, burned at the stake, and then I will subject you to public humiliation. You have been warned.
I can think of two possible reasons to include such ridiculous statements:
- You want to encourage SAS users to employ the most highly regarded statistical and plotting methods.
- You want to compete with that other well-known and “highly regarded” data analysis tool known as Microsoft Excel.
What got me started on this particular rant is that I saw the latest VIEWS Newsletter (.pdf), in which Leroy Bessler pointed out how misleading a three-dimensional pie chart (Bessler’s Figure 1) was, and how a two-dimensional pie (Figure 2) chart fixed the problem. I agree. But then Bessler goes on to critize a three-dimensional bar chart (Figure 3), recommending another three-dimensional bar chart (Figure 4 shown below) in its place.
Sadly, I have to disagree with Bessler. His Figure 4 is not a good chart either. It contains useless information and adds structure to the plot, but does not provide any more insight into the data compared to the two dimensional version of the same plot. Tufte speaks of the “data-ink ratio”, that you want it as high as possible, but the additional ink and structure to make it three dimensional adds no more information to the plot. The third dimension interferes with the transmission of information — your eye sees how the colors change in the bar light to dark from top to bottom; it notices the ovals at the right edge of the bar that give the plot its three-dimensional appearance. The ovals at the right of each bar would also prevent you from accurately reading the height on the bar on an axis (although there is no axis in this plot). Now I admit that for this simple bar chart, making it three-dimensional certainly is not a fatal flaw, but it still violates the principles of a well-designed chart. The third dimension is unnecessary and in other plots might be potentially confusing (see below). It is like trying to explain what happened at your child’s soccer game while running around the room playing air guitar like a rock star. It interferes with the transmission of the information. It is like putting a Bart Simpson background on a plot. It interferes with the transmission of information.
Consider this ridiculous and atrocious three-dimensional bar chart from the PROC GCHART documentation (click on the plot to see it full size):
Wainer, in his classic article “How To Display Data Badly”, points out that the human eye registers the area of a colored region, and the three-dimensional nature of the plot adds to the area, in defiance of the area needed to be portrayed by the data. The tops of the bars, visible on only the topmost of the groups, also adds area to this one group, and the fact that the bars overlap left-to-right subtracts some areas. All of this extra area deceives the eye into a visual assessment that is at odds with the data. All of these things, plus the three-dimensional background, interfere with the proper understanding of the meaning of the chart. Wainer, Tufte and many others implore you to create plots where the meaning is obvious; instead of creating plots that use “glitzy” features of your software.
I have a plea — no, a command — to all GVSUG members. Do not use VBAR3D, HBAR3D or PIE3D, or any other type of three-dimensional pie or bar chart. It simply is unacceptable. Using the powers vested in me by GVSUG, if I catch you using one of these plots, we will have you condemned to eternal damnation, you will be hung, drawn-and-quartered, burned at the stake, and then I will subject you to public humiliation. You have been warned.
posted by Paige @ 5:33 PM
<< Home