Chart Deception: Part 1
Slide 1
I believe that graphics are an import way to get results to marketing managers. Now that I’ve provided the basics of chart design, I’lldiscuss how researchers can design charts that can deceive marketing managers about underlying processes. Often these deceptions are unintentional, so alerting you to these practices should help you to avoid them.
Slide 2
I’ll begin by defining two basic terms, which are conflated in common English usage: information and data. Information equals data plus labels. In the example, there’s a 10 digit number, which seems meaningless until I provide a couple parentheses and a dash; suddenly, it’s a phone number. With the parentheses and the dash it becomes information. A string of ten numbers is merely data. To a large extent, that’s what graphs do: basic data and added labelsconvert the data into information that managers can use to make better decisions.
Slide 3
Graphics are wonderful for summarizing data. You must be carefulwhen summarizing data that you haven’t hidden so much detail and that the summary is deceptive of the underlying process.
Slide 4
Here’s an example of hiding detail. The bar chart at the top of the slide shows last month’s versus this month’s sales. Seemingly the trend is upward, which if true would make the retailerhappy. Unfortunately, monthly sales by salesperson suggest a different story, as this month’s sales are almost entirely generated by one salesperson. That should cause the retailer much concern because next month it’s unlikely that the one salesperson would be responsible for such a large amount of sales. As a result, next month is likely to be a real problem. Although the first slide suggests things are going well, the second aggregated slide suggests there may be a future problem.
Slide 5
The value of a pie chart is to denoted percentages. Therefore, you should use percentages not raw values when labeling pie chart slices.
Slide 6
The next two slides show you pie charts in which raw values, rather than percentages, are used.This example shows why this is a bad idea. If you add all the numbers for those slices together, you’ll discover they indicate a 7 ½ hour day; thus, something has been omitted. It’s difficult to tell, if you’re looking at raw figures, if something has been omitted or doesn’t sum. A pie must sum to 100%.
Slide 7 (No Audio)
Slide 8
In this case, the failure to use percentages is not about omitting a slice, as each year is represented. The problem is what percentage is that? The viewer must guesstimate from the area of the slice the percentage that corresponds with that slice. That’s asking the viewer for too much. The 1990 slice seems to comprise more than 25% of the pie, but there’s no way to know how much more than 25%. To clarify and to make maximal use of a pie chart, you should include percentages, which indicatethe percentage of the whole comprised by each slice.
Slide 9 (No Audio)
Slide 10
Using multiple pies is a bad idea, but trying to fix that problem by including raw figures, rather than percentages, just compounds the problem.
Slide 11
As most viewers of your pie charts will be familiar with languages that read from left to right, people looking at your pie charts will start at the left and move their glance to the right, where they’ll finish viewing your chart. Because that’s where they finish, that’s where you should place the most important slice. If you use an exploding slice, one to further parse what’s contained in that right-most slice, then youalso should place it on the right-hand side.
Slide 12
This slide shows an example of what I mean by an exploded slice. Notice that the right-most slice is the most important slice and the exploded slice is then depicted to its right.
Slide 13 (No Audio)
Slide 14
This black and white scan of the original pie chart stresses why you shouldn’t use 3- D effects in pie charts. As I’ll advocate throughout this lecture, 3-D effects to improve aestheticscan distort perceptions about a graph. In this case, the 3-D effect causes the bottom slice to appear bigger than it is and the top slice—due to foreshortening—seems smaller than it is. Thus, all you’ll accomplish with 3-D effects isto mislead viewers.
Slide 15
The problem with multiple pies, even if you attempt to make the areas of the pies proportionate to their true size, is deceptive because your eye cannot discern changes in areas well. When people design multi-sized pies so the first pie is supposed to be half the size of the second pie, they make the diameter of the first pie one half the diameter of the second pie. As you may recall from high school geometry, the area of a pie is not twice but more than three times its diameter. Doubling the diameter creates a pie more than three times the size of the original pie. This deceives your eye as you try to compare slices between different sized pies. Rather than constructing two pies that are appropriately proportioned to one another, just avoid using multiple pies on the same graph. In addition, becauseyour pie slices should sum to 100%, the miscellaneous slice may become too large and obscure too much detail. If you resort to a miscellaneous slice, ensure that slice doesn’t hide important detail.
Slide 16 (No Audio)
Slide 17
It’s acceptable to combine pie and column charts if your goal is to provide more detail about the most important slice, which should be located on the right side of the pie chart.
Slide 18 (No Audio)
Slide 19
A donut chart is a series of stacked pie charts. You may remember this toy from your childhood: a series of progressively larger colored rings that you stacked by size on a post. If you starred down that post, you’d see a sliceless donut chart.
The issue with such charts is the increasing area associated with each subsequently larger donut. Refrain from using donut charts for the same reason you should refrain from using multiple pie charts: it’s almost impossible to create multiple pies proportioned appropriately for cross-pie comparisons.
Slide 20 (No Audio)
Slide 21
When designing pie charts, be careful how you treat missing data if missing data comprises a meaningful fraction of the data you attempted to collect. By deleting that slice, you’ll increase the percentages associated with all remaining slices. If the missing data slice is small, then it’s not a problem. However, if the missing data slice is large, then you’ll need to decide whether or not to exclude it from your pie chart.
Slide 22 (No Audio)
Slide 23
Graphics can communicate much information in a short time. There’s a good reason for that. Our culture isso accustomed to certain graphing conventions that we don’t need to figure outgraph conventions. We understand X and Y axes, plotting points over time, and the like. There’s a lot assumed in our ability to read graphs. If you think about a standardized exams like the SAT or ACT, the authors assume you can read graphs. Most such exams include several questions about interpreting charts.Thus, you shouldconsider existing people’s predispositions toward chart reading when designing chartsand be careful not to violate those predispositions. The next two slide show what happens when you violate those predispositions.
Slide 24
Charts typically flow from left to right, so peopleassume progress and time advance in that direction. If time is on the X axis, then the left-most point is the first year and the right-most point is the last year. For the Y axis, people assume that up is more and down is less. To design a graph otherwise will just confuse viewers who glance at your graph quickly and assume that it’s consistent with these graphing norms.
Slide 25
This graph is terrible for several reasons. First, it violates basic assumptions about how we create graphs in western society. The years are on the Y axis and they go from most recent year to the furthest year. Years don’t go on the Y axis and we don’t sequence them in this fashion. In addition, sales are increasing towards the bottom, which makes no sense. The other reason it’s terrible is all the chart junk. The motorcycle at the bottom is traveling backwards and the point of this graph is to stress that sales are increasing; the chart junk distracts from this message because the motorcycle is moving from right to left, which is in the direction of less rather than more.
Slide 26
Here’s a quick attempt, with just a bit of cropping in Adobe Acrobat, to fix the previous slide. I’verotated the graph 90 degrees. I’d need to change two additional things to really fix this graph. First, I’d need to place the sales figures in units or dollars along the Y axis. Second, I’d need to kill the 3-D effects, which can only distort reading this graph. At least this is a good first attempt to fix the graph. I also eliminated the distracting chart junk. You may think it adds aesthetically, but if it distracts from the message, it’s harmful.
Slide 27
Horizontally oriented bar charts also violate western norms for reading graphs. That’s one reason why I suggest you try to avoid horizontal bar charts.
Slide 28 to Slide 29 (No Audio)
Slide 30
Here’s another violation of the implicit understanding that people bring to interpreting graphs. Close your eyes and think about a map of the U.S. As you think about that map, you can think about the eastern seaboard, the mid-west, and the far west. In your mind’s eye, you see a map where the eastern part is on the right and the western part is on the left.This graph violates that image; the left bar is east, the center bar is west, and the right bar is central. Over time, people will forget the labels on the X axis and just remember the relative size of the bars. Ultimately, they’ll remember sales were highest in the east, when in fact they were highest in the central. You want to ensure, when designing graphs, not to violate people’s expectations. The expectation here is that east is on the right, west is on the left, and central is in the middle.
Slide 31
Notice how much more intuitively appealing the top bar chart is relative to the bottom one. That’s because it’s vertically oriented rather than horizontally oriented.
Slide 32
The bar graph depicted on this slide is lousy for two reasons. First, its’orientation is wrong; the bars are oriented horizontally rather than vertically. Second, the X axis is discontinuous; there’s a gap between 0 and 40%. Although this gap is denoted, the differences among the various bars appear larger than they truly arewhen you break an axis this way. The unintended message here is that there’s a huge difference between the first and all subsequent bars. In fact, the difference isn’t 50%, it’s only 1/3rd because we’re going from 95% to 65% or less. Looking quickly at these two bars, it seems the top bar is double all subsequent bars, which is incorrect.
Slide 33
In designing bar and area charts, I recommend that you refrain from stacked bar charts, as cluster bar charts depict that same information in an easier-to-understand form.
Slide 34
Here’s an example of a stacked bar chart. The relative area in each of the six stacks corresponds to the score for each of those three brands. It’s reasonable to assume that each stack sums to 100% and that the relative magnitudes of the components of each stack are readily recognizable.
Slide 35
Unfortunately, those stacks are confusing, and this alternative graph gives a much better sense of the original values for each of those means. In this case, you can compare all the white bars to one another on the relative scores for each of the six dimensions for a given product. Within each characteristic is the scores of the three products relative to one another. Yourease in makingsuch comparisons illustrates why I recommend clustered bar charts over stacked bar charts.
Slide 36
If it’s important to communicate the relative percentages of different components in a bar, thenyou might try a 100% bar chart rather than a stacked bar chart. Think of a 100% bar chart as a solution to the multiple pie problems. Pies should indicate percentages summing to 100% for all slices; these bars accomplish same thing. Each bar is identically sized; thus, there’s no problem with distortions or trying to compare slices across multiple pies. If indicating percentages of components is important, try this 100% bar chart style.
Slide 37
If you insist on stacking graphs, be certain to place the least variable component on the bottom. If you place the most variable component on the bottom, it will make all the levels above it seem more variable than they truly are because they are riding on top of a component that’s highly variable. If you place the least variable component on the bottom, then whatever variability the viewer senses from looking at your graph will come from the top-most rather than the bottom-most components. In other words, you distort the gestalt provided by the graph by placing the most variable component at the bottom.
Slide 38 to Slide 39 (No Audio)
Slide 40
As the note associated with this graph indicates, this is a poor graph for two reasons. First,horizontal rather than vertical bars are used. Second, the bars pointing to the left indicate greater magnitudes associated with a more leftward position, which is the opposite of conventional graph norms. Generally, pairing charts in this way is confusing and deceptive because it violates basic norms for viewing bar charts.
Slide 41
The next six slides show excellent reasons for avoiding 3-D effects in graphs. The final two slides are not strictly bar charts, but the message is to avoid 3-D effects in software packages like Excel, Quattro, or SPSS. All such effect can only confuse viewers.
Slide 42 to Slide 47 (No Audio)
Page | 1