Health Insurance Cost Data Visualization
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?
December 20th, 2007 at 1:53 am
Out of those, I’d go with the Google one. Despite the lack of formatting, it still seemed to render best. Those last two defaults aren’t too pretty. I don’t know why they don’t just take a few minutes and make the defaults nicer looking.
Personally, for a simple chart like this, I’d go straight to Adobe Illustrator, and modify away. For larger, more complex data, I use R, and if it has to be super pretty, I open the resulting pdf in Illustrator.
December 20th, 2007 at 1:18 pm
Nathan - I just wish that Google would add a few, what I would consider primary, formatting options. I agree that it renders well and is probably the best default chart out of all the tools.
I’m not too familiar with R and Illustrator, but I would have to think that the process you described is fine for a one-time chart, but not very repeatable. Would you agree?
One of these releases of Office will probably have a better default chart.
December 20th, 2007 at 3:36 pm
It’s repeatable in the sense that R is programmatic, so you can do it as many times that you want. When I bring something I do in R into Illustrator, that part isn’t “repeatable.” At this point, I’m usually customizing and adding context, which is going to be different for every chart.
December 21st, 2007 at 9:09 am
I really can’t imagine creating visualisations in anything that isn’t programmable (in my case, I use R). I have yet to work with a client who doesn’t discover some problems with their data that requires changes. I’ve written the ggplot2 package for R, which implement’s WIlkinson’s grammar of graphics (a great theoretical graphics book if you aren’t familiar with it). You can see some examples of what you can do with it at http://had.co.nz/ggplot2
December 21st, 2007 at 2:35 pm
This may be obvious, but I see two different uses of information visualization:
1) A setting (typically business) where a user needs to present data, which in most cases will be recurring (daily, monthly, quarterly, annually, etc). For this purpose I wouldn’t see Illustrator as a good option. For this instance, one would need to use a program that has a sufficient output format. Also, the place much of the software fall short.
2) A one-time presentation of data, which would commonly be found in magazines, newspapers, or possibly on the Internet. I believe this is where Nathan was coming from when he takes an object from R to Illustrator to fine-tune. I don’t see doing this for something that is recurring. I do see a huge benefit when personalizing data visualizations.
I also agree that data typically needs to be scrubbed for various reasons most typically, formatting. I have perused your site and think you have done an outstanding job as noted by the many awards and accolades you have received.
One limitation I often see in the business world is that there are very few people, like the two of you that can program with a package like R and present something that is extremely eye-appealing (as I’ve seen Nathan do). In my opinion, most people are good at one or the other, but rarely both.
How does R compare to Minitab or SPSS? I have a hunch that many readers will not be familiar with R.
December 21st, 2007 at 3:36 pm
I think you’ve missed an important use of visualisation: for exploration. When performing a (complex) data analysis I might produce on the order of hundreds of graphics, most of which I throw away. These need to be cheap to create - both in terms of thinking time (how go I go from the plot in my head to one on the screen) and computing time (once I’ve told the computer what I want, how do long does it take).
With respect to R vs Minitab and SPSS, both make it easier to produce canned graphics, but harder to produce custom graphics. SPSS also uses a system based on the grammar of graphics, but I’m not sure how well that is exposed to the user. The other important difference (if you’re not familiar with R) is that it’s open source and if you’re doing serious research into statistical computing, you’re probably doing it with R.
December 21st, 2007 at 3:51 pm
You make a great point with exploration. I truly believe that exploration is widely underutilized in the business world. I find very few people that just try different models or algorithms to see what they can learn. Much of the visualization is in the form of simple [backwards] reporting.
My goal for 2008 is to discuss more about what I call ‘Intelligent Business’ vs. Business Intelligence. Stay tuned!
January 15th, 2008 at 4:57 am
[...] 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 [...]