The Part-to-Whole Healthcare Data Puzzle

When I was a kid, my grandmother was always working on a jigsaw puzzle of one sort or another and I would always work on it with her while she and I visited. I was pretty good at the puzzles, but it was my Uncle Ed who always managed to find and fit way more pieces than anyone else.

I knew to find all of the outside edges first, that was pretty easy, and then I would pore over the picture on the box searching for the parts of the picture, using the colors as my guide to try and figure out how the remaining pieces might fit.

But not Uncle Ed–he never looked at the picture on the box–ever. He just fit the puzzle pieces together one after another, after another–and he always found more pieces than I did.

How did he do that? I literally begged to know–he was just too good–he was up to NO good.

Finally, my grandmother ratted him out–it seems that my Uncle Ed was color blind and so he only looked at the shapes of the puzzle pieces, which enabled him to include or exclude the pieces that could potentially fit together a lot faster than I could.

He saw the parts of the whole based on shapes alone.

Once my grandmother let me in on my Uncle’s secret I started to use it to my advantage. By looking at the shape of a puzzle piece first–like he did–I could determine if it should be included or excluded, quickly reducing the pool of possible puzzle pieces for the section I was working on. Only then would I check to see if the color matched. By adding the first step and then coupling it with my advantage of being able to see the colors of the puzzle I could (and did) beat my Uncle Ed at his own game.

To say that I was a competitive child may be a slight understatement–and yes, a foreshadowing of things to come.

The simple fact was–I learned a few rules and techniques that changed how I approached puzzles–and I received early training in how to think about and consider part-to-whole relationships.

Displaying part-to-whole (proportional relationships) in healthcare data is perhaps one of the more difficult challenges that people face, but if you know a few of the rules and techniques for displaying this type data, you will become proficient at it (just like I did at beating Uncle Ed at the puzzle game).

When faced with displaying part-to-whole relationships in your data you may be tempted to use a pie chart or a stacked bar chart with bright colors. Just say NO to the pie chart and consider other options that may be preferable to a stacked bar chart.

Begin with the appropriate medium to display your data–not the color of your graph. What encoding will best display the shape of the data–the story in the data?

Consider the following three examples:

1. Values and proportional relationships can be more easily interpreted and understood when displayed as a bar graph.

PSA Bar Chart

The pie chart below displays the same data as the bar graph above, but it is much harder to compare the sizes of the slices than it is the lengths of the bars–especially for the values that are close in size–and the colors add no information at all–they are purely decorative. Additionally, the bar chart allows for ranking values–not so much with a pie chart.

PSA Pie Chart

2. Bar graphs also work well for comparing how multiple part-to-whole relationships differ. See how challenging it is to compare the total number of medication occurrences and their causes in the following three pie charts.

Med Occ Pie Charts

Now notice how easy it is to make these comparisons using the bar graphs below. Again, it is about the shape of the data and the information that is imparted to the viewer–you no longer have to worry about the different colors to figure out the data (and blue is so soothing).

Med Occ Bar Chart Revised

3. A stacked bar chart is fine for some displays but remember, it should only be used when you want to display and compare several part-to-whole relationships (for example, one for each country in the following WHO example).

WHO Med Stack Graph"

The problem with displaying the information in this manner is that it is difficult (if not impossible) to compare the segments of a stacked bar between countries, because they do not share a common baseline. The most the reader can gain is a general sense of how the whole is divided into its parts.

Instead, if you want the viewer of your graph to compare multiple wholes to one another along with the ability to compare the parts of each whole, consider using multiple graphs like the following:

WHO Med Bar Chart

Puzzles are challenging and fun–they provoke us to think and strategize about how to fit the pieces together to create a picture, and they are a great model for how we can approach the display of healthcare data. And best of all, when you know the “rules of the road” and the “tricks of the trade” you will find that in actual fact you will have a whole lot of fun with both of them.

And truly, if you haven’t put together a puzzle in a while, I encourage you to give it a go–you will be amazed at how fun it can be–especially now that you know my Uncle Ed’s secret.

Leave a Reply