DSA Insights

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.

In the May 21, 2007 issue of BusinessWeek, you can see an excellent example of how to improve presenting data.  Below is a copy of the pie chart, which is titled, “The Other Banking System.  A growing number of Americans seek financial services from alternative institutions”.  Looking at this chart, it is difficult to quickly compare the seven different slices.  Here are a few points why I think pie charts, in general, fall short:

1. Shades of blue are hard to distinguish

2. Handouts will be hard to read without a high-quality color printer

3. Scale of slices are difficult to interpret and compare

4. Takes much longer to create with all of the customization needed to make meaningful

5.  Can be too cluttered when you have many slices or use different colors for each slice

Check out the bar chart below that has the same data and only took me a few minutes to create using the Excel chart wizard.  Formatting the pie chart takes a lot longer and is much more difficult.  I find the bar chart is straightforward to read and is simple to create in MS Excel.  For some reason people just gravitate to pie charts; probably because they are prettier.  I would presume that the pie chart is just what everyone expects when trying to show data similar to the one above.  By using the bar chart below, you can easily see the scale, highest and lowest percent, and labels.

Here is yet another option to use the in-cell graphics functionality that is very easy to create in MS Excel.  This is my personal favorite and is simple to format to your liking.  The charts in Figure 2 and 3 reduce “chartjunk”, a popular term coined by Edward Tufte that simply means “unnecessary or confusing visual elements in charts and graphs.” [Wikipedia] 

After looking at these three options for presenting the same data, which one do you think is the most effective? 

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

In the March 12, 2007 issue of BusinessWeek, there is a great example of taking charts and analytics to the next level.  If you are looking for that wow factor, then start creating charts like the one shown here.  Impress your audience by going beyond the basic chart wizard.  The best part is you don’t need anything more than Microsoft Excel.  If you are interested in creating charts like this, drop me an email.