Friday, November 2, 2012

to stack or not to stack

My husband came across the graphical focus of this post in his Google+ stream last week. The original source is a Wall Street Journal blog post summarizing a recent Forrester report, where the main story can be summed up by one of the Forrester quotes within the post, "The future is one where no single OS or vendor is dominant - Microsoft is extremely late to the market expansion into mobile and has lost its dominant position."

Here is the graph included to illustrate this point:

It's true that you can get the evidence to support the claim made from this graph: once you identify the light blue portion of the bar as Microsoft, we see clearly that it decreases over time as the orange portion (Google) and the yellow portion (Apple) become increasingly prominent. But I'm of course not satisfied with it. The color palette is strange. Color in general could be used more strategically here. We can eliminate the work of going back and forth between the legend and the data it describes. I'm also not sure how I feel about the stacked bar chart.

Let's look at a couple variations on this data viz. First, here's what it could look like if we preserve the stacked bar and use color a little differently (note: I didn't have the raw data, so the remakes below are based on me eyeballing the figures and likely aren't entirely accurate):

Other minor changes I made above:
  • Added an action title so it's clear what to look for in the graph (this was included as tiny text below the graph in the original post).
  • Oriented the graph title and legend text at upper left - so reader encounters how to read the data before they get to it.
  • Added a title to the y-axis. Always include this!
  • Added data labels to the Microsoft series. This acts to both draw more attention to Microsoft, as well as to give a quick numerical view of the decrease over time.
  • Narrowed the bars. In the original, they are bordering on too thick so that our eye starts to try to compare the area rather than the height.
When it came to color, I took a look at Microsoft's logo. I nearly always use blue to highlight the areas to which I want to draw attention. In this case, I actually tried venturing out using the red color from the logo and then the green color from the logo. But both just looked a little off (the burnt red looked overly negative to me, the green a little pukey). So I went with the blue from the logo (matching by eye - it's not perfect but close). I chose a grey palette for the remaining series.

I still don't love this, for a couple of reasons.

First, I think I just don't like stacked bar charts. This is actually probably a good use case for this graph type, since this lets us emphasize the percent of total and how that's changing over time. But I still don't love it. Because the bars aren't oriented on a consistent baseline, our eyes are forced to compare differing heights starting from different points. That's fine to get a general view, which is probably all we need here. Perhaps I'm just being overly finicky.

Second, it's still a bit of work to go back and forth between the legend and the data. If you don't recognize that the legend (left to right) is in the same order as the right-hand bar (top to bottom), it could prove difficult to see quickly which series is which.

I thought I would like a line graph of this data better - it would allow me to organize the series on a consistent baseline as well as label each directly. But then I graphed it and reconsidered:

Personally, my issues with this line graph version are greater than my issues with the stacked bar. We lose clear visibility/confidence that the lines sum to the total market, 100%. I also worry that in this case, the lines make it at first glance appear that we have more data than we do. Perhaps three points per series is too few for a line graph? If not in general, than I think that is at least true in this case. The overlapping nature of the lines creates a sort of spaghetti graph (as if I had a handful of uncooked spaghetti noodles and threw them on the ground). I tried to make this better by emphasizing the main three series (Google, Microsoft, and Apple) and de-emphasizing the others. But it still isn't great.

Given this, I'm back to the stacked bar chart. I think the information comes across most quickly with that structure.

What do you think - is a stacked bar the best choice here? Or are there additional options worth considering?

In case you're interested, here is the Excel file with the makeovers.


  1. I like your stacked bar the best. That said, I would probably put Microsoft on the bottom of the stack. Also, it's hard for my eye to match the greys in the bars with the greys in the legend. I'm not sure if the solution is different greys or maybe other colors but dim?

  2. I definitely like the line graph best. The story is about Microsoft's downturn, so you don't really need to explicitly see if it all adds up to 100%.

    Actually, I'd get rid of the "Other" and "Proprietary Android Brands" categories and let it be the typical fight of giants overview. I believe that's what most readers are looking for, and as long as the story isn't about any changes in the "other" categories, that should suffice.

    The great fall of "Other" could easily have been picked up though. That really could be a great story. The fall of Linux? Symbian?

  3. I might like the line graph better if the other lines were not gray scale (implying less overall importance by shifting them into the background). The stacked chart is interesting but my eye sees the 2nd stack 30% as being higher up (and comparable) than the 66% bar in the 1st stack (very weird), whereas in the line graph i noticed the more pronounced plunge from 66 to 30. I lack the ability to explain this.

  4. Thanks, everyone, for your comments! Joel, I agree with your recommendation to put Microsoft on the bottom in the stacked bar version - this would get rid of the phenomenon James mentions where the 30% in the middle and right appear higher than the 66% at the left. Rene, in the case of lines, I agree that getting rid of the bottom two unless there's a story to focus on there.

    Jorge Cameos did a nice remake with bars separated vertically here:

  5. I like the line chart better, but not by a whole lot. I'd recommend using solid lines between dates with hard values and dotted/dashed lines between dates with forecasted values. Since only the 2008 values are known, it might be best to use dotted/dashed lines for this entire graphic. That said, I'm not sure how much I'd like it.

  6. Thanks for the fun exercise Cole.

    I experimented with side-by-side bar charts. It loses the sense of 100%, but emphasizes the change over time and that it goes from 1 to 3 players. On the other sheet in the spreadsheet I tried keeping the stacked charts, but using color on all three to emphasize the shift.

  7. Hmm. Good question. I might suggest a standard bar graph --not stacked-- since it makes it harder to tell who is growing if Microsoft is not. For example: What is the Google share in 2012 as compared to the 2016 projections?

  8. Here's a mock-up of Joel's suggestion above, with some extra formatting changes that may or may not make it better :)

    Great post - I think I'd still go with the line chart. I liked Andy Kriebel's coverage of this topic (although he compared line charts with stacked area as opposed to stacked bars):


  9. Cole, and Community of data visualizers. Thank you for your insights! It's fascinating just being on the sidelines watching all of you problem solve. Thank you, Cole, for sharing and explaining so clearly, through this blog post and so many others, how you untangle what your eyes see and reconfigure the information in a much improved format.

  10. I like Ben Jones's version of the stacked bar chart, with the largest share (Microsoft) on the bottom, which I think has the additional advantage over Cole's redesign in that it moves the legend from the top of the chart right next to the data.

    I also like Jorge's redesign--what I like to call a "broken stacked column chart"--but I would have put the x-axis labels at the bottom of the chart and would have sorted the rows by size of share in 2008.

  11. I would be curious to try a series of three pie charts. I wonder if there is a way to visually draw lines between the charts to trace Microsoft's ever shrinking slice of the pie. sorry I am the lame dude throwing an idea out without doing it, I am on a mobile phone : )