Planned Parenthood gives us this breakdown of services provided by their affiliated health care centers (download pdf):

As pie charts go, it’s not terrible. It demonstrates MoveOn’s point, that abortion accounts for only 3% of Planned Parenthood’s treatments. The remainder goes for contraception and health services, mostly for poor women.

The data is also well-presented using a bar chart. This chart uses larger type and yet requires less space than the pie. The data is sorted in an easier-to-read layout, and it’s easier to highlight one value (abortion) using a darker color, because the chart uses shades of only one color.

The bars show all of the components of the whole block of services provided, as indicated by the horizontal axis label. To show that the bars add to 100%, another data series can be added and connected with a line, to make a Pareto chart. The top four items, all unrelated to abortion, account for 96% of Planned Parenthood services.

Peltier Technical Services, Inc., Copyright © 2011.
 
Licensed under a Creative Commons Attribution-Noncommercial-Share Alike 3.0 Unported License.
 

And the Grinch got a wonderful, awful idea.

The chart appeared in 2011: The Year in Review on the Kickstarter web site. Kickstarter is a unique program for funding projects, which lines up proposals with backers who pledge support for them. It actually seems like a great program, but they need to work on their graphs.

The cited article describes their funding activities for last year. The pie chart contains much of this data, but not in a form that is remotely useful.

The “Chart”

This is the first view I had of the pie chart. All of the slices are the same size, which is strange, and no information is visible except for the category names, which is awkward.

More ominous is the label above the chart that reads “Mouse over a color to see that category’s stats.” So you have to mouse over the chart to show any of the data.

I first saw this on my mobile, and, well, I couldn’t mouse over anything. Later I loaded the page on my laptop so I could examine it in greater detail. I moused over the pie, expecting to see a popup with some data. Nothing. I spent more than a minute mousing over the picture before I realized the data was displayed in the same rectangle that first directed me to mouse over the chart.

When you mouse over the wedge, you need to note the category name, and try to memorize three numbers, before you move to the next wedge and its three numbers. And the next, and, uh. . .

The Problem

I’ve been calling this illustration a “chart”, but it’s not really a chart. It’s not really a graph. The best I can call it is a picture, or an illustration.

It’s not a graph, because not a single byte of data is displayed graphically in the illustration. You see the data when you mouse over a segment of the pie, then it disappears when you mouse over another segment. If you can’t remember the text, you cannot compare the values from segment to segment. Maybe I was stupid not to notice the data the first time I moused over a slice of pie, but nobody is smart enough to keep all of this text in their memory.

It’s so bad, that I moused over the picture for several minutes, and had no idea which of the categories had more or less support, and whether support measured by amount pledged tracked with number of backers or number of funded projects.

Interactivity is okay when it’s done effectively. The user should not need the mouse to see the overview of the data. The interactivity should be used to allow deeper investigation into subsets of the data and into alternative views.

A Better Chart

This is a rough draft of a different view of the data. I hardly spent any time on the labels, just enough to give viewers a taste.

It’s a simple panel chart, with three panels of bars, one each for amount pledged, number of projects, and number of backers. The categories are sorted by decreasing order of amount pledged, not alphabetically. The sorting is almost preserved in the other panels, which makes sense: more projects and more backers would gather more pledges.

Every bit of data that is hidden within the original pie is clearly displayed in the panel chart. No need to play cat and mouse to see information, no need to rely on a weak short term memory to allow comparisons and to see trends.

Peltier Technical Services, Inc., Copyright © 2011.
 
Licensed under a Creative Commons Attribution-Noncommercial-Share Alike 3.0 Unported License.
 

The other day I encountered a tweet with a common theme: “Worst Pie Chart Ever.”

So I had to follow the link. It was an article in Atlantic Cities, 2011 Major League Baseball Regular Season: Attendance by Team, which stated that combined attendance at all Major League Baseball (MBL) games was fifth highest in history. The article was accompanied by a chart, not showing a time series of attendance by year, which you might expect after seeing the “fifth highest in history” statement. Instead, the chart compared the attendance of the MLB teams. And it compared the teams using an unexpected approach.

Pie Charts

Here’s the MLB attendance chart. It’s not the worst pie chart ever, but it has substantial deficiencies.

The choice of chart type was strange. When you think about a team’s attendance, you think of total numbers. Showing the teams together as a portion of the league total is unusual, and the actual numbers are obscured.

Comparison of the sizes of the wedges is not easy, especially with so many data points, but at least the data is sorted.

A big problem with this chart is identification of the data points. As the tweet indicated, the randomly assigned colors did not make use of team colors to aid in identification of the slices. There is no legend, but a legend is a poor way to label a chart, because you have to drag your eyes back and forth between chart and legend. In any case there are repeated colors, which would reduce the legend’s effectiveness.

This chart addresses the labeling problem by popping up a single label as the mouse passes over each pie slice. Unfortunately you have to wander around the pie to find your favorite team, then wander around again to see how other teams compared, then try to remember what you found before. This cognitive load overpowers the ability to interpret the data.

Since the screen capture didn’t capture the cursor, you don’t even know from this image which slice corresponds to the label. It’s the key lime pie slice in the lower right of the pie (it was black in the first view above).

Interactivity is a nice way to let a user find additional information that would otherwise clutter a graphic. However, forcing a user to interact with a chart simply to extract necessary information is a waste of the user’s energy and a waste of interactive effects.

Granted, showing each category label in a 30-point pie chart takes up lots of room, as you can see in my recreation of this chart below.

If you want to show the actual attendance values, well, the labels take over the chart, almost becoming chart junk themselves (see below). The chart has essentially become tabulated data, but not in a nice tabular grid.

Without showing the attendance values, you have no way to estimate attendance, since the wedges range from 2% to 5% of the total pie circle. Do we expect the reader to multiply an estimated percentage by the posted league total? I just don’t think that a percentage of league total is the best way to display attendance values.

How do we label the categories (the teams), and also provide a numerical measure of attendance?

Bar Charts

A bar chart lines up all the names in a neat list, and encodes attendance by the length of bars along a horizontal scale. Now without mousing around the pie, I can easily see how the teams rank. I can see that Philly had around 3.7 million attendees and Boston had just over 3 million.

The data was sorted, but by average home game attendance, not by total attendance. Because of rained out games which were not rescheduled, the relative rankings by the two measures are slightly different. In the total attendance chart above, Washington is listed higher than Chicago but has a shorter length bar, while in the average attendance chart below, the bars are ranked according to length with no exceptions.

We can also look at average road attendance. This chart is also sorted by average home attendance. Notice that there is much less variability in the data. Each team plays in each other team’s parks, and it’s the home team’s demographics (population and fan rabidness) that drive attendance.

We can plot both series together to compare average home and road per-game attendance by team. I don’t find bar charts very well suited for a dense chart like this: the shorter bars are obscured by the longer bars of the other color.

Of course, a pie chart is much less effective at this kind of comparison. One pie chart wouldn’t be enough, you’d need two side by side. This requires a great deal of eye movement and short-term memory to recall the individual values while switching from one pie to the other. When I see two pies side by side, it reminds me of bicycle tires and spokes, and I want to draw in the bike frame.

Is there a better way to compare data like this?

Dot Plots

We can show this data effectively using a dot plot. The values are encoded by horizontal position of a marker rather than by length of a bar. Longer bars do not obscure shorter bars, and even overlapping points can be resolved.

Dot plots can be used for any of this data, not just for comparisons. Below is a dot plot showing total home attendance by team.  It’s as effective a graph as the bar chart in the previous section.

Worst Pie Chart of the Week?

I stated that the MLB attendance pie chart was not the worst pie chart ever. In fact, it wasn’t even the worst of the week. That honor is bestowed on an organization that has the ironic initials “BI”. In CHART OF THE DAY: Goldman Reveals What You’re Getting For Christmas, Business Insider reported results of a Goldman survey about planned Christmas giving. The pie chart shows which percentage of respondents planned on giving which category of gift for Christmas. What’s wrong with this chart?

Somebody forgot that pie charts are supposedly good for showing portions of a whole, that is, they show values that add to 100%. Since each respondent could provide multiple answers, not a single answer, the sum of the percentages in the chart above was 434%. Oops.

This data should have been presented in a bar chart or dot plot.

Well, we might have lowered our expectations, had we noticed the Engage-O-Meter gauge in the sidebar of the Business Insider site. Such a gauge is totally useless, since we don’t know what the minimum and maximum values are around the circumference of the gauge.

Without the large number above the gauge, we would know nothing. I suspect the needle never moves, and only the number updates. Even if the needle moves, it’s only gratuitous animation, because only the number has meaning.

Peltier Technical Services, Inc., Copyright © 2011.
 
Licensed under a Creative Commons Attribution-Noncommercial-Share Alike 3.0 Unported License.
 

What’s wrong with this chart? Well, it’s two pies used to try to show different values over time. Why is this bad? Despite their proximity, each pie is a separate entity, and we have to look at one of them, load that into short term memory, then look at the other, then look back. In order to make the comparisons, we need to look back and forth many times, as if watching a table tennis tournament. It’s hard to make a lot of comparisons, because within a given pie, and from pie to pie, the different slices are misoriented. We could count minutes around the circumference, but this confuses the false measurement of minutes with the angular measurement that represents percentages.

This graphic is also very cluttered. One pie has icons and labels, while the other only has icons, so we must use the first pie as a legend. Since the topic is time, the pies are decorated like clockfaces, complete with tick marks for hours and minutes to confuse us, and hour, minute, and second hands to disrupt out view of each chart. In addition, the two pies are shown in different sizes, so we wonder whether the overall pie diameters are significant. And one pie overlaps the other, but the damage was already done.

“Improving” the Pie Charts

Paresh reworked the chart and came up with the following pair of pies. He criticized the original pies for having no baselines, but then neither have his. He removed the hour and minute tickmarks, but added his own markers at 45° intervals. At least there’s no clutter on the face of each pie, so you can see that the dark blue wedge decreased a lot in size, and the, uh, other dark blue wedge increased by almost as much. The rest of the wedges look about the same. I discovered later that there were three fill colors used, for large, small, and no changes.

I  put on my Chart Busters cap, and made my own pies, partly to correct the ills of the two previous sets (I removed the markers around the circumference,  and sorted the wedges), and partly to demonstrate once again that pies aren’t so effective at showing changes. Sure, we can see the large changes I mentioned before, and the two smallest slices seem to have grown. But all the other slices have changed positions around the pies, so as stated, there’s no baseline.

Hint: “Watching Films” (red) increased its value by 10%, but you wouldn’t know it from this graphic.

I don’t know about anyone else, but when I see two pie charts, I want to draw on a nose and mouth.

Unsuccessful Line Charts

The general consensus about plotting trends over time tells us to use line charts. I put this data into a line chart, and got a result which was not very good. There are obviously some points that are hidden behind the ones we can see, and it’s hard to identify the series.

To identify chart series more clearly, I like to remove the legend, and place data labels by each series in the chart. That didn’t help in this case. Out of nine labels, three pairs are completely overlapped, and another pair is touching. Conventional wisdom has not helped much, though at least we can see that most series saw changes from start to finish.

Semi-Satisfactory Bar Charts

Okay, we know bar charts can be effective when comparing values. In this bar chart, clustered by year, it is easy to compare percentages within each year, but not between years.

Transpose the data and group by category, and it’s easy to compare by year, and we haven’t lost much within each year. Time on the vertical axis is a bit irregular, though, and rotating the chart will result in hard to read category labels.

Stacked Bar and Area Charts

My comment to Pradesh was that his use of markers around the circumference of a pie chart was like adding a numerical scale. Might as well straighten out the scale and make stacked bar or stacked area charts. These aren’t too bad, but it’s hard to tell whether “Watching Films”, “Reading”, “Shopping”, and a few others changed during the five years.

Panel Chart – Hey, This Works!

The line chart is still best at showing the trends, but what if we didn’t have to overlap all the series? Let’s put each one into its own panel of a panel chart.

This graph clearly shows even the smallest change, and preserves the percentages as well. Apparently everyone has become bored with their friends and family, and is spending more time on other activities.

Peltier Technical Services, Inc., Copyright © 2011.
 
Licensed under a Creative Commons Attribution-Noncommercial-Share Alike 3.0 Unported License.
 

New York Times Donut Charts

It’s a series of donut charts showing the market share of Biotronik over the period during which the physicians were hired as consultants. From a market share of zero in 2007, Biotronik zoomed up to 95% in two years, forcing out their main rival Boston Scientific.

I’m not going to analyze the scandal, but I do want to examine the New York Times’ choice of graphics.

We all know the rationale for donut and pie charts: they show proportion of a whole. And as far as they go, each of these charts is reasonable for its purpose, to show the market shares of two or three vendors (treating “other” as a single entity). In all cases one segment of the chart is much larger than the others, and we get a good qualitative sense for the market at each point in time.

If I had seen any of these charts in isolation, I would not have though much of it. However, this visual uses several standalone snapshots to show how the market evolved. We can get a rough qualitative sense of the change from chart to chart, but to show trends it makes more sense to show data in a single chart than in a series of charts.

Line Charts

My first cut at the data was with a line chart, market share against year for the three entities.

The lines in these charts draw the eye to the trends from year to year. To further clarify the trends, I decided to replace the blanks in the line chart with zero values, connected by dotted lines to the original values.

This chart clearly shows how rapidly Biotronik grew its market share and how precipitously Boston Scientific’s share fell.

Area Charts

Suppose we still wanted to show a percent of the whole kind of chart, while maintaining the clear trends. I don’t use stacked charts often, but with only a few areas to consider, I thought I’d try a stacked area chart.

That shows the changing market shares pretty well, but an artifact of the stacking order (a danger of using such charts) makes it seem almost as though Boston Scientific is increasing. So I changed the order of the series.

This chart gives a clear sense that Biotronic started from nothing, and expanded rapidly, pushing Boston Scientific off the bottom of the chart.

A Proponent of Donut Charts Chimes In

I first saw the New York Times donuts in Pie Charts and Doughnut Charts – Tension Releived! [sic] on Paresh Shah‘s Visual Quest blog, where he made much of the inherent Gestalt principles of stability and equilibrium in pie and donut charts. Paresh followed up in Gestalt Principles. Doughnut charts.Pies, in which he states that pies and donuts “satisfy the human craving for stability and equilibrium”. Well, I wanted to learn more about the relationship between pie charts and various Gestalt principles, so I Googled pie chart stability equilibrium:

The first two links are to the first article by Paresh cited above, the third is to the web site of a woman whose Balance of Self Theory uses a “Balance Pie Chart” to help maintain one’s stability in life. The fourth links to a comment by Paresh in a recent post of mine, Why Do We Love Pie Charts?. Further down the list is Excel chart gallery: a difficult equilibrium by Jorge Camoes on his Excel Charts blog, where he discusses not Gestalt principles but instead the preponderance of “junk” charts in Excel’s chart types dialogs. In the first five search results pages, I found no mention of the stability and equilibrium of a pie chart.

A follow-up search of pie chart gestalt turned up Paresh’s article cited above, his comment on my post also cited above, and his comment on In Defense of Pie Charts, by Robert Kosara on the Eager Eyes blog. But there is nothing else about stability and equilibrium of a pie chart in the first five search results pages.

So I guess there isn’t a lot of discussion abut pie charts satisfying the human striving for stability and equilibrium. For me, the obsession with clearly presented data completely overwhelms any desire I have to seek stability within circles. A poorly conceived and implemented graphic actually upsets my equilibrium. SteveT commented on my Why Do We Love Pie Charts? post that pies may satisfy not Gestalt, but something Freudian.

Peltier Technical Services, Inc., Copyright © 2011.
 
Licensed under a Creative Commons Attribution-Noncommercial-Share Alike 3.0 Unported License.
 

The value axis of a bar chart must include the value zero.

We saw how Microsoft, at least the PR department responsible for the online tutorials, didn’t understand this rule in Misguided “Professional” Charting Tutorial. In addition to encouraging us to use excessive gradients in the fills of a bar chart (and to use 3D bar charts), the tutorial instructed us to start the vertical axis above zero. The reasoning was to illustrate the differences between the values, but the axis minimum was only changed from 0 to 50, while the maximum remained at 400. So we gained hardly any improvement of resolution, while still getting substantial distortion of the smaller numbers.

Another department at Microsoft has done it again. This time it’s the Internet Explorer team, blogging about the improved performance of IE9, in Browser Power Consumption—Leading the Industry with Internet Explorer 9. The blog post includes several bar charts, and all suffer from the don’t-start-at-zero problem.

In the first analysis, this chart was used to compare browser power consumption when viewing a single blank page. This is actually the total power consumption of the system when the only thing running is the operating system and a browser navigated to about:blank. Here is my recreation of the original chart; the original chart did not include System Idle (with no browsers running).

Your first impression, a false one, is that Opera 11 is quite a power pig, using twice as much power as IE 9, Chrome 10, and Firefox 4, and that Safari 5 is nearly as bad. But when you notice the scale starts at 10 and reaches to 12, you might just realize you’ve been deceived.

The vertical axis here has unreadable minor tick spacing, because I based it on the original chart.

When the chart is redrawn and the axis now includes zero, we see that there is not much difference between browser power consumption. Sure, Opera and Safari are higher, but not by too much. The difference between Opera and IE is less than 6%, not around 100%.

What if you do want to focus on the relative values? Since the markers in a line chart encode values based on their position along a scale, we could use a line chart.

The problem with a line chart in this case is that the lines imply a trend between the categories, and there really isn’t a meaningful order of categories over which a trend would be meaningful. I guess we could sort by software version number (Firefox 4 through Opera 11), but this isn’t very meaningful either.

Perhaps a better value to track is the increase of power consumption when a browser is being used. In this chart I’ve replaced the discrete System Idle bar by a background that’s shaded up to the System Idle value.

That doesn’t give us the resolution we need. So I changed my X axis minimum to 10.529, the power consumption at System Idle with no browser running. Our axis doesn’t start at zero, but it does start at a reasonable baseline.

Those axis numbers are not easy to make sense of, so why not just subtract the System Idle value from all of the data, and plot the excess power consumption.

This gives us the resolution we want, while plotting a reasonable value that is zeroed at an appropriate baseline value.

The valid conclusions we reach from this chart are that, under the conditions of these tests, the latest versions of IE, Chrome, and Firefox all use very little power, while Safari and especially Opera use substantially more power. Less power usage means longer laptop battery life. You can follow the Microsoft article and repeat this analysis for all conditions they evaluated.

Thanks to Robert Kosara, aka @eagareyes, whose tweet this morning brought these charts to my attention.

Peltier Technical Services, Inc., Copyright © 2011.
 
Licensed under a Creative Commons Attribution-Noncommercial-Share Alike 3.0 Unported License.