Box plots are sometimes used to represent a range in a single value, but I am curious if a similar presentation would be useful in representing an overlaid column chart. By using the same color as the background for the minimum height column, the other columns can be made to float.

For a two-value overlaid column, the impact seems strong, making it very clear which item is larger and by how much.

Ann vs. Dean sales

With three values, clarity seems to noticeably decrease. Even adding a marker to indicate the missing column may not make the interpretation clear, though this presentation does seem to dramatize difference.

Ann, Dean, Ed

A comparison to a consistently intermediate value might be practical, but it then seems necessary to clarify that the lower bar descends to the value. This also seems to assume that the deficit/surplus quantity is significant.

Ann and Dean versus Median

With five values, the distortion effects seem significant since the comparisons are much less simple.

Five-Way Comparison

Even so there might be some datasets and purposes where a floating column chart with more than three values provides appropriately greater emphasis than a dot plot without confusion or deception.

The Questions

Under what circumstances — if any — would such floating column charts be useful versus dot plots or more traditional (overlaid or adjacent) column charts?

How can the legend, fill choices, and other design elements make clear the overlaid nature, especially when comparing a larger number of values?

Table of Values
        1Q14  2Q14  3Q14  4Q14
Ann      20    18    17    25
Bob      17    20    19    21
Celia    25    35    32    38
Dean     15    22    23    17
Ed       12    15    15    19
  • 1
    Floating columns are used in Waterfall charts.
    – teylyn
    May 25, 2015 at 20:36
  • I like it. But this does Ed no justice. It's somewhat confusing that Ed sells everything below the light green arrow. Why not just sorted overlaid bars, as the image seems to deviate from? Just being 'artistic?' The "SalesPerson" legend should be below or above all the data. Perhaps? It's a very compact presentation. Very aggressive. May 27, 2015 at 6:41
  • @JediCommymullah The deviation from overlaid bars is intended to emphasize the difference (it seems to work very well when comparing only two items, similar to the waterfall charts teylyn mentioned). By using the floating form, there may also be less deception if the displayed range is reduced (similar to dot plots but with emphasis of the difference) without requiring (variable base) relative values. Its use might be limited to two-way comparisons even if other items are included (e.g., like the median example but with only two fill colors), though if a relationship to a reference …
    – user43012
    May 27, 2015 at 12:19
  • (whose unit is consistent but whose value varies moderately) is significant and could be more clearly communicated, then multiple items might be compared as such avoids the implicit minimum. E.g., comparing hotel average star rankings in various categories to the reviewers recommendation or the cheapest room. If the reference is constant, then it can be used as a baseline and there is no need for floating. E.g., computer performance on several benchmarks normalized to the reference.
    – user43012
    May 27, 2015 at 12:19
  • In thinking about this, if a trapezoid shape was used relative to a meaningful reference, then the overlaid nature would be clearer and the offset nature would be more clearly communicated. One problem is when two values are equal. This is most clearly problematic with more than values where the minimum value's category is implicit; dividing the column into equal area vertical fills might work for other cases.
    – user43012
    May 27, 2015 at 12:20

1 Answer 1


Paul, I was thinking of something like this... enter image description here
...in that even figures can be well represented. It's basically your floating example with the bars removed, to which I've appended a column. More info: median or average, can be added. It appears like a spectrum analyzer.

Box plots have a more constrained representation of data, and it's difficult to compare two sets of data within one block. But they are attractive.
enter image description here
Here is depicted a single box plot. When visualized as real estate, you realize there is a limited amount of progression that can be achieved with it. You can fill it with dots, stars, or horizontal bars to indicate data within its minimum and maximum, but you would be duplicating what you would normally accomplish outside of it with some other sort of plot. So the box is a container, which works great at representing a single set of data as a tri-partite. With the implication that each salesperson is a distinct set of data, it's probably difficult to squeeze all of them into that mold. Of course, this doesn't answer your question.

When and how can a floating column chart be useful?

Obviously, a lot of similar data is not going to look right in this scenario. All the bars terminate at the same point. So, by drawing on your footnotes, we can say, anytime you have a graph

"whose unit is consistent but whose value varies moderately"

It's useful in portraying sports performance. For instance, in an auto magazine to represent auto performance in competitive comparison. The idea is to exceed the bar of one's lesser opponent. To survive, to surpass, to win election. A very competitive and dissonant idiom.
You might find it useful sometime. Good luck.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.