DSA Insights

In a recent issue of BusinessWeek , I found the picture shown below. A simple pie chart would show this data, but this adds a certain flare that caught my eye. Maybe it only caught my eye because I live and breathe data analysis and visualization… Regardless, it’s similar to a horizontal stacked bar chart, but with more pizzazz.

I have stated before that I don’t like stacked bar charts, and I don’t. Whenever you compare more than one component over time, it becomes very ineffective. Once you get beyond the first series of data, the baseline is not the same, making a comparison difficult. However, when you only compare a few pieces of data without time on an axis, a horizontal stacked bar chart can be effective. You can visit an earlier post to get more information on stacked bar charts.

In the graphic below, there are some things I might have done differently. For example, the last 3% is a little hard to distinguish because of how small it is compared to the rest. The first and last colors are very hard to differentiate in the print version. In this picture, the colors are more defined, making the comparison easier.

I share this with you to promote more abstract thinking when it comes to presenting data without losing effectiveness. I can tell you this: if I was an Executive and someone brought me this visualization instead of a pie chart, I would be impressed!

Visualize this: you’re watching a presentation and the slides are gliding by with every imaginable abuse of PowerPoint: fifteen bullets per slide, clip-art images, data-packed charts that aren’t even visible, goofy transitions, then along comes this slide. The only thing on the slide is this picture below. All of a sudden ears perk up and slouching turns to posture only a second grade teacher could be proud of. Dare to be different, yet effective!

Without regard to any minute details or scale, I replicated this visualization using Excel, which is shown below. I literally spent about seven minutes creating this in Excel. Granted it looks a little better when the image isn’t modified to fit this blog, but you get the idea. In a future post, I may show how this is done in a screencast, which I can guarantee won’t take longer than a minute.

Stay tuned!

In the December 10, 2007 copy of Businessweek, I came across the horizontal bar chart below. Fundamentally, it depicts employee versus employer annual percent change in health insurance costs from 2004 to 2007. As you can quickly see, the horizontal bar chart is lacking in effectiveness unless you tilt your head ninety degrees to your right. So, I added a few of my own that really only took a few minutes to create.

Businessweek version:

Here is my Excel version using approximate values.

Here is my Xcelsius version with approximate values.

What I found interesting is that when you adjust the size of the graph, you get a more dramatic slope in both lines between 2006 and 2007. I wouldn’t recommend changing the size to maximize your theory or objective. Also, I do not like that the y-axis starts at 2% instead of 0% in the Xcelsius version. (Recommended reading: How to Lie With Statistics)

Below is my version using Open Office , which is an open-source (free) project that contains most of the products commonly found in the Microsoft Office Suite.

Finally, below is my version using Google Docs, which had minimal formatting options that I could find. I’m not even going to post the Many-Eyes version due to its lack of formatting. If you want to see it click here .

My preference is either the Excel or Xcelsius version, which illustrate the 4-year trend much better than the horizontal bar chart. Both took about the same amount of time to fine-tune. If I was limited on budget and didn’t want to spend any money, I would go with Open Office over Google Docs. There are many more formatting options in Open Office that help to create an effective data visualization. The only drawback is the amount of manipulation it takes to get from the default graph to the ones shown above. Both Excel and Xcelsius default to a horrid looking graph that I wouldn’t recommend using (both shown below).

Default Excel Chart:

Default Xcelsius Chart:

Which one(s) do you like the best? Would you suggest another option not found here?

Just the other day, I received the latest copy of the Fidelity Investor’s Quarterly magazine in the mail.  You can see the same chart below if you click on the link and scroll to page - P2.  After dinner, I was reading through the articles when I happened upon this data visualization (shown below).  At first, I was drawn to the amount of pie charts that fit inside one line graph.  It’s no secret that I avoid pie charts like the plague.  However, I found myself thinking that this was a pretty crafty use of pies to show the parts of the whole that made up each asset allocation mix. 

In retrospect, I’m almost ashamed that my first thought was that this may be a good use for pie charts.  That is until I started trying to figure out which shade of green matched to bonds and which to cash.  I found that as my eyes moved across the line graph towards aggressive, I quickly forgot the color for bonds and domestic stocks.  The cross reference between the legend and pie charts while trying to follow the line graph was not working too well.  Also, don’t forget that the more you move away from zero on the x-axis and y-axis, the more risk and return are expected.

Feeling frustrated, yet challenged, my brain started to churn through designs that would work better for this given data set.  My first thought was, keep the line graph and add a bar chart for the mix of investments making up the allocation matrix.  A stacked bar would work just fine.  Then I recalled a recent post I wrote detailing how ineffective stacked bar charts are for making a comparison. 

My next option was to just add a simple 4-row table below the line graph that plainly listed the percentages for the four different investment options.  That would have worked, but I knew there must be something more effective.  Then the light bulb went off.  I remember seeing a similar example on Stephen Few’s website under the examples section found here.

Finally, I found a better alternative and off to Excel I went.  Below, you will see my version of Stephen’s design with a few minor changes using this data.  This design, using bar charts instead of pie charts, makes it much easier and faster to compare the six different asset allocation strategies.  Some of the formatting is a little bit off due to converting the Excel workbook to a .jpeg file.  Since some of the numbers are a little hard to read, I have included a link here to the Excel version.  Just when I think there may be room for pie, I quickly dismiss the notion.

Here is a good example of someone trying to get creative with displaying the amount of caffeine in different types of drinks.  I applaud their effort because it’s different and visually ties the amount of caffeine in each drink to a brand. 

Below is the original data visualization:

One issue I see is in looking at Pepsi Max and Sobé, which have 46 and 48 mg of caffeine per serving respectively.  The problem lies in the bar displayed in Pepsi Max that shows the 46 mg being higher than the 48 mg bar.  I can only guess how that may be possible, but think it may be an error or oversight.

Below is my rendition of a better design:

I have no way of getting rid of the bars they used in the drinks.  If possible, I would have removed the bars and added them to below the picture as shown in my version above.  Also, I removed the names of the drinks because they are self-evident.  Granted, this data could be shown in a simple bar chart, but adding the pictures makes it a little more eclectic.

Here is a recap of the posts for this series:

1. Bar vs. Line Chart Part 4 – My Preference
2. Bar vs. Line Chart Part 3 – Stacked Bar Example 2
3. Bar vs. Line Chart Part 2 – Column Panel, Sparklines and Waterfall Chart
4. Bar vs. Line Chart Part 1 – Small Multiples Line Graph

Here is the original stacked bar chart that GE used in its 2006 annual report. 

This is my enhanced version of the data using a line graph.  The only part missing, which could be added easily is the total for all business segments. 

The main reason I prefer this graph over the stacked bar chart is because I can quickly see the change for each business segment over the five year period.  There is no referencing letters to a legend or illusion from stacking the segments.   If you’re color blind it wouldn’t matter with this graph.  Another plus is the minimal amount of real estate that this one takes up.  Best of all, this is very close to a default Excel line graph with minimal formatting changes.

Here is a recap of the posts for this series:

3) Bar vs. Line Chart Part 3 – Stacked Bar Example 2

2) Bar vs. Line Chart Part 2 – Column Panel, Sparklines and Waterfall Chart

1) Bar vs. Line Chart Part 1 – Small multiples Line graph

The stacked bar chart below came from a recent issue of Businessweek.  There are three things that concern me about this chart.

1) The use of a stacked bar chart makes any analysis of the individual segments extremely difficult.  As I mentioned in part 1, if the lowest level in the bar increases, it can give the illusion that others have gone up.  At least there isn’t a lot of chartjunk…

This question should help prove my point.  Can you tell what the change is from 2010 to 2020 for Ultra-Heavy Crudes (hint, it’s in grey)?  I can’t without getting a ruler or micrometer.  For that matter, what is the change in Biofuels from 2010 to 2030 (hint, it’s in bright red)?  It’s almost impossible to identify these changes with the use of a stacked bar chart.  What may have helped, although I wouldn’t advise it, would have been to add values to each stack in the bar.  This way I could tell if the segment increased, decreased or didn’t change.  Adding values would be better, but not optimal.

2) I don’t think there is any intent to deceive the observer.  However, if you look closely, you will notice that under the history section, the span starts at 10 years (1980 to 1990, 1990 to 2000).  Yet the span to the fourth bar is only four years.  Deception?  You be the judge.  It was probably just the last year that they had data, or that’s what I tell myself.

3) What if you are color blind like an estimated 7-10 percent of Americans?

On a positive note, at least they notated that 2010 through 2030 are projections [hint of sarcasm].  We wouldn’t want anyone getting the wrong idea or forgetting what year it is…

Finally, upcoming in part 4 of this series, I will unveil my preferred graph to visualize the data shown in GE’s stacked bar chart from part 1.  I may be biased, but it’s off the charts (pun intended)!

In the first part to this series, I revealed an alternative to the stacked bar chart that GE used in their 2006 annual report.  This post adds a few additional options to consider instead of the original stacked bar chart I found to be ineffective.

The first illustration (below) is another attempt at a vertical panel chart instead of the original stacked bar chart.  The advantage to using this style is how clean it looks and how effectively it shows the trend of each business segment over the five years.  The downside is the amount of manipulation it takes to format the charts to be effective.  I would almost always prefer a line graph when time is present along the x-axis.  A bar chart uses too much ink, which takes longer for your eye to see the trend. 

The next chart (below) is a simple attempt at showing the business segment trend using sparklines.  I am not a big fan of this chart for a few reasons.  First, it uses bars, which I said above is not my first choice.  Secondly, there are only five years of data, which makes for a small sample to depict in a mini-graph.  Sparklines are designed to show as much data as possible using minimal real estate.  You will typically see sparklines used to show a time series of stock prices or indices.

Finally, I have added a waterfall chart to show the breakdown of revenue between GE’s business segments for 2006.  This chart is great when you don’t need time across the x-axis.  The primary purpose of waterfall charts is to show how an initial value is increased or decreased by a series of values.  For this illustration, I modified the chart so it starts at zero.  It can also replace the dreaded pie chart or just serve as something different.

The next post, part 3, in this series will reveal my preference for a chart to replace the GE stacked bar chart.  Stay tuned.

How much data can you cram into one bar chart?  The bar chart below, taken from GE’s 2006 Annual Report, attempts seven; six different business segments and the total for all segments.  The only bar that is relatively effective is segment F, Industrial.  Other than Industrial, I can see that the total revenue rises from $108 Billion in 2002 to $159 Billion in 2006.  In general, I find it much easier to see trending over time by using a line graph versus a bar chart.  As soon as you add more than one object to a bar chart with time across the x-axis, the graph becomes worthless. 

 Here is why I think this bar chart is ineffective:

¨     There are no labels on the y-axis for reference

¨     It takes entirely too long to cross reference a letter to business segment

¨     It is nearly impossible to differentiate scale after the first segment

¨     It’s deceptive because a segment may appear higher in subsequent years, only due to the lower segment increasing.

¨     The chart has too much unutilized (white) space in key areas

I have added two Excel graphs that I believe are much more effective.  The only downside is that they use a bit more real estate. But, I welcome the tradeoff.  The only difference between the two Excel charts is one is more horizontal and the other vertical.  I prefer the one that is horizontal because I think it’s easier to view left to right.

The benefit to using the Excel version is each line graph is highly effective.  Within seconds, I can see the trend of each business segment as well as the total trend for GE over the five-year period.  If I were to use a tool that has a selection option, where the chart changes as a different segment is selected, I would be no better off.  The problem is there can be no comparison when you only see one segment at a time.  Edward Tufte popularized the term, small multiples, to represent a series of small similar pictures, making a point through repetition.  See the examples and make your own conclusions.

Fact: Did you know that pie charts date back over 200 years to 1801?  

I found this example in AIG’s annual report.  My intention isn’t to be critical of AIG, it’s to show a more effective solution.  Recently, I wrote a post that ranked the top annual reports based on a few data visualization basics and AIG fell towards the bottom of that list.  Take a look at the pie chart below.  Let me ask one very simple question.  Why include the pie chart?  The data is already present and organized in descending order in the table, which is really ample. 

I can only presume they did because people have been conditioned to always think, pie chart, when looking at parts of a whole.  When I look at the pie chart below, all I want to do is slap down a $20 bill, pick a color and spin it!  

Statisticians tend to regard pie charts as a poor method of displaying information. While pie charts are common in business and economics, they are uncommon in scientific literature. One reason for this is that it is more difficult for comparisons to be made between the size of items in a chart when area is used instead of length. …This suggests that length is a better scale to use, since perceived differences would be linearly related to actual differences.

      source 

One of the purposes of a chart or graph is to give the person looking at it a quick and effective means of comparing multiple values.  Using the pie chart alone in the example below, it’s nearly impossible to compare the values accurately.  The one thing I can quickly deduce is that it has been a long time since visiting the money wheel at Foxwoods!

Below is the chart that I would prefer to see when looking at parts of a whole.  I like this style because it’s easy to create, very effective and doesn’t use Excel Charts, which can be maddening to get just right.  Albeit not perfect, it’s pretty close and much easier to create.  You really don’t even need the numbers, but for this example, I wanted to show them to attest the scale works. 

In the last Golf Magazine issue, there was an interesting graph that illustrated the results of an online poll.  The title is, “Who Will End Tiger’s Reign as World Number One?” and is posted below as a thumbnail.  In my humble opinion, I think the only one to end Tiger’s number one rank is [Tiger’s] retirement. 

Here are a few comments on this graph:

The Good:

• Allure.  The graph catches your eye and is visually appealing; probably much more than a simple bar chart or (cringe) pie chart.  Kudos to Golf Magazine for not opting for the pie chart!

The Bad:

• Chartjunk (“The decoration of graphics generates a lot of ink that does not tell the viewer anything new,” Tufte) that does not add any true value.  Some specifics are:
  • Lines that are drawn from the name of the person to the slice of the graph.
  • Fill within each of the slices is distracting and adds no value.  Even if you were color blind, all of the fill styles are the same.
  • 3-D is almost always a bad choice in data visualization and usually just skews the depiction and makes the data difficult to understand.
• Layout - I don’t understand why Phil Mickelson’s slice is slightly behind the Grim Reaper’s slice.  It seems more logical to have it separated like the rest, so the spacing is even. 
• Scale - It is nearly impossible to quantify the difference between Adam Scott and Jim Furyk without having the percentage labels.  So, if you need the labels, then why not just display the data in a table (see below)?  They probably didn’t because the table is much less attractive.

In this instance, I think most readers can get the same value, but it just may take a little longer.  I have posted two different Excel versions of the same data that took no time to create.  This is an interesting approach to displaying data and much better than a pie chart.  I believe laying the data out horizontally is more effective because you can see the scale more efficiently.  Click on the thumbnails to see a larger view.

Click here to send us an e-mail to subscribe to DSA Insights and get updates by e-mail.