Recently the topic of primary and secondary axis scales was discussed at the PTS Blog. The discussion starts with a reference to Stephen Few's March 2008 Visual Business Intelligence Newsletter titled Dual-Scaled Axis in Graphs - Are They Ever the Best Solution . Stephen starts his article by concluding that He then goes on to cite a series of examples. Finally, at the end of his article he states that "I certainly cannot conclude, once and for all, that graphs with dual-scaled axes are never useful; only that I cannot think of a situation that warrants them in light of other, better solutions. I invite you to propose viable exceptions, which I will welcome with open arms."

From a healthcare finance perspecitive, it has been my experience that these charts are great tools for those who know how to build them and for those audiences that know how to interpret the underlying data. Below are several comments I have concerning Stephen's article:

Comment #1 - It's not necessarily the chart with two axes that's bad, it's the data behind it. In the newsletter, most of Stephen's graphs illustrate the relationship between revenue and units sold. I would argue that is comparison is flawed from the beginning and, as such, this type of chart should not be used. My reasoning is that, although it seems like the data is should be related, when you drill down into the data you might find out that the correlation is not as close as you might have thought. Depending on the data:

• What defines a unit?
• Are the units the same unit each quarter or are there multiple types of units being charted as one?
• If there are multiple units, how are the changes in volume accounted for within the sales mix?
• Won't price increases distort the data over time?
• How does the chart account for decreases in purchasing power over time i.e inflation over time?
• If the data is measured by quarter and one of the quarters includes a leap year, how do you account for the extra day?

In my opinion, I think an argument can be made that a dual axis chart showing revenue and units sold is probably not the the best use of a dual-axes chart. That being said I do have to admit that I do use this type comparison at a top-level where the components of variance might be considered immaterial. If a dual-axes chart is to be used the audience should be educated on how to interpret the underlying data and how to recognize and address possible flaws before decisions a made.

Comment #2 - If you are going to present a dual-axis chart, the axes need to be proportional. That means doing the math for each axes to make sure it is in fact proportional to the other.

Comment #3 - As Stephen observed, line graphs are the best presentation for this type of chart.

Comment #4 - "I can’t think of a single case when there isn’t a better solution than a graph with a dual-scaled axis." In my line of work, we've found that these graphs are very useful for auditing calculations. For example, you are tasked with building a projection of revenue and discounts for the next year. In a healthcare environment, revenue and discounts both contain the same components of variance i.e. work days, fee increases, volume changes, payer mix changes, and changes in service mix or acuity. The calculations the build all of these components can get very complex. Rather than attemping to proof each calculation that builds the projection, this chart can quickly pick up areas of possible error.

In the example below, it's obvious that there's problem with June calculations because the lines are not in sync for that month.

Comparing the growth in same-type volumes (for example Radiology vs. Laboratory volumes) is another example.

To conclude, there may in fact be better charts for displaying certain types of data. But if the audience understands the relationships between the data I think there is a place for these charts. And like the example above shows, there's most definitely a place for these charts when performing audits.