Waterfall Charts

My daughter Anne graduated from college in May. To commemorate the occasion, I had my mother’s engagement ring (which I inherited many years ago) made into a necklace for her.

Annie graduated with a degree in Elementary Education from St. Joseph’s University in Philadelphia, and because my mother had also been an elementary school teacher (and a big fan of the Jesuits), it all felt very right.

Having the ring made into a necklace got me to pull out photos of my parents’ wedding. My mom had sewn her own dress (she is the third from the left in the photo below). The ceremony was on a brutally hot day (or so I was told); afterwards, like thousands of other young couples of their generation, my folks honeymooned in Niagara Falls, New York.

wedding photo

Of course thinking about the Falls made me think about waterfalls, and that made me reflect on the frequent misuse of waterfall charts to display data.

Yes, it’s true: you never want to spend any time inside of my head-ever.

I have spent quite a bit of time considering waterfall charts; even so, I find them rather tricky to understand. I suspect this is in large part because there are hardly any examples of well-designed waterfall charts. It is also entirely possible that I simply don’t like them. They seem a bit gimmicky to me, and most of the data they display would be easier to understand if it were presented in a different and (usually) simpler type of graph.

Originally created to show how totals (especially financial balances) change over time, waterfall charts are now often misused to show merely a vertical stacked bar with the different categories of data spaced and staggered along it.

The following example of Patient Days by Nursing Unit is a perfect example of this misuse of a waterfall chart:

waterfall chart

This chart is simply four sets of values that add up to the total bar on the far right of the chart. One problem of displaying the data in this way (from left to right) is that it gives the impression that the data are changing through time, that there is some sort of movement to it (as in a waterfall) when in fact the data aren’t trended; there is no movement or change through time. Rather, this chart simply displays the number of patient days, stratified by nursing unit, and in total.

The other problem with displaying the data this way is the suggestion of a hierarchy in the data or the Nursing Units, when there is none. The data aren’t ranked, and there is no basis for assuming significant or important differences in the Units.

This chart provides yet another perfect example of how a very simple, plain-vanilla bar graph can tell a story more directly and clearly than a more complicated structure, without implying something untrue or unreal, such as a shift or progression in the data. It displays simply part-to-whole relations, no more.

Below are the same data reworked into a simple bar chart. It easily ranks the information by nursing unit, and clearly displays the percentage each unit contributed to the whole. We can also include the number of days per nursing unit and show the total days at the end of the column. All it took to include these last two figures was to edit the data labels, then customize and insert a horizontal label to display the total number of patient days.

bar chart

I will admit that waterfall charts are, at first blush, rather seductive. But when you carefully consider them in light of what you are trying to convey, you are likely to find far more appropriate ways to display your data.

Now a final and more serious thought about what started me down this path. It is so terrific to come from a long line of teachers, and to now see my daughter poised to continue the legacy and-let’s be honest-provide me with more stories and ideas.

Leave a Reply