Data visualization improvements for this graphic

I need to visualize two related sets of data, a performance differential and a performance ratio. The image below is my first thought. The dot and wiskers represent the average differential with 95% CI and corresponds to the left y-axis. The barplot represent the ratio and corresponds to the right y-axis.

The feature I like in this graph is that it make clear the importance of considering both the absolute (differential) and relative (ratio) performance in comparisons (for instance, Deer vs. U. americanus).

However, the obvious deficiency here is that I am lacking a good way to also include the CI for the ratio statistic. Another deficiency is that the differential and CI for 'carnivore', 'U.americanus', and 'C.latrans' are hard to see. I can't just log transform the data because the lower confidence value is negative. Taking the absolute value or translating then transforming the data makes the relationship look awry.

Is there a better way to visualize this type of data that still highlights that relationship?

Edit: A little more background on these data... I am studying a specific behavior in different species and species groups. I have a theoretical number of animals I expect to present this behavior in my treatment and then records of actual observations of the behavior in the treatment. To get the differential, I subtracted the the observed rate from the expected rate and then bootstrapped the CI. The sign of the value indicates if the treatment had a positive or negative effect on the behavior. However, this absolute value doesn't take into account the total number of animals present (e.g. 101-100 and 2-1 have the same difference of 1 but much different relative performance). So I also would like to show the ratio. I've also considered using percent change from the expected value which would be more convenient than ratios in that it would also be centered at zero.

• Could you describe the data you're plotting here a bit more? How does a ratio of performances end up being negative? – dsaxton Nov 28 '16 at 3:27
• @dsaxton I've added a little more background to the nature of the data in the original post. I never said that the ratios were negative, but the differentials are sometimes negative. – et is Nov 28 '16 at 4:03
• Ah, the tick marks look like negative signs so I would probably remove these. – dsaxton Nov 28 '16 at 4:35
• What is the range of the performance values? Can you include the raw data for the above graph? – xan Nov 28 '16 at 14:48

One way to look at it is that two continuous variables can be plotted against each other in a scatterplot. Here you want to represent an interval and a central value for one of the 2 variables, so what you will represent is that interval along the relevant axis. In ASCII art this would look like this:

Var A
|
|       [-x-]
|
|  [---X---]
|
+---------------- Var B


The name of the animal can then be encoded as a color (if you have as little as 7 animals like in your example), or as a label plotted on the graph. If the latter, if many points or labels are visually partially overlapping you may have either rely on a dynamic representation (e.g., tooltips or side selection for an animal of interest) or placing the labels outside of crowded areas, with a line showing the points they are associated with.

This is more of an opinion than an answer, but I think having one axis measuring two different things is generally confusing and not a very good practice. For instance, there's no way to know by looking at the plot which scale corresponds to the bar plots and which to the points with error bars. And the viewer is left wondering which metric he / she should be paying attention to.

I would either choose the metric that's more relevant to what you're trying to convey and plot only that one, or if you feel both need to be presented then consider creating two separate plots, possibly stacked vertically.

• I definitely, agree; it is confusing. I'm hoping someone has a clever idea to elucidate the graphic while still retaining the relationship between these metrics. The relationship between the two is almost, if not more, important than the statistics of each for this study. – et is Nov 28 '16 at 4:18