Registered: 1452795542 Posts: 17
Reply with quote #1
Emeritus Faculty, Security Executive Council and author of "Measures and Metrics in Corporate Security", presented the following chart at a security conference and again in a webinar as an "excellent" example of how to illustrate value to senior management. This is easily one of the worst graphs I've ever seen. Here the x-axis is running perpendicular to one y-axis (to be expected) and parallel to the other y-axis. Consider for a moment what you can discern from the data points of the line graph ... it is impossible to know for sure which data point in the line graph goes with which category of alarm. Below is what the graph would have looked like if he had left both y-axi perpendicular to the x-axis (as is logical practice) ... However, there are other poor design elements to be considered here. First, the use of a line graph is improper here as it suggests a relationship between the categories that doesn't exist. Also, given the data being presented here, I feel the use of a dual axis at all is unnecessary. How would you present this data more effectively?
__________________ "In the realm of data visualization, we face many challenges that are worth pursuing. Creating an effective radial gauge is not one of them. It is a fool’s errand. Do I strike you as a fool?" - Stephen Few
Registered: 1348995178 Posts: 186
Reply with quote #2
Difficult, indeed, to find any sense in original chart. It looks like due some formatting bugs two different charts just overlapped.
Dual axis charts were discussed in this forum on several occasions. Without a common unit of measure which makes sense for both variables, is no point in displaying them in one chart. Line chart in top of a bar chart makes nosense either.
How would I display this data more effective? Is no answer for this question. We display data with a purpose, don't we? What did the author intended to show?
1. If his intention was to prove a strong correlation between variables, a scatter plot with a regression line showing the linearity of correlation would do just fine.
2. If his intention was to compare individual values, two side by side horizontal bar charts, each with its own quantitative scale, sharing the same categorical axis, sorted in descending order of first quantitative measure would be my first choice.
3. If his intention was to show the ratio between variables a two clustered bar chart can show that (but it makes not much sense, having different unit of measures), my choice would be three side by side horizontal bar charts showing ratio, variable_1, variable_2 ordered by ratio...
It is no "best way" of displaying certain data. They are best practices in displaying data for certain purpose. It is true that for this trivial example of two quantitative and one categorical variables the amount of possible visual representations is quite small and we can design all of them (can we?) and choose relevant representations. I personally prefer to understand data, define an objective for my data investigation and design that accordingly.