Crossed comparison of empirical distributions

Maxime sent another figure from the same manuscript as the previous one, comparing the same 12 groups against each other. While the previous figure presented inferential results (means and standard errors), this one is descriptive of the distribution of the viral load observed on the sampled individuals.

Once again, they wished to highlight the contrasts across groups that came up statistically significant out of their analysis.

(a) USUV viral load in Culex pipiens midguts, thorax and heads for EU2 and EU3 lineages at low and high bloodmeal titer. Viral load is expressed in genome copy/µL. For each condition, dotplot present the raw data, the black point the mean and the line the confidence interval. Only significant differences (***: p-value < 0.001) estimate by generalized linear mix models are represent on the plot.

Figure 1: Original figure and caption.

• I like the idea of using dot plots for this figure. They work like a histogram, but are more appropriate for data of limited size by representing the binned location of each individual in the data set.

  I only modified the stacking parameters so that points do not overlap and changed the colour of the border to match the fill.

  I also changed the centred arrangement by a unilateral stacking with different directions for each lineage, to emphasize contrast (same trick used in the first post).

• I suggest again switching the axes. It is easier to compare positions horizontally, and to read labels at the left hand side.

• As before, I also suggest modifying the representation of the groups, in order to facilitate the contrasts of interest.

  However, the mapping suggested in the previous figure with four subgroups within Tissue mapped to the \(y\)-axis results in a tall and narrow figure which is not natural to read.

  In this case, I would rather favour a presentation of Tissue over 3 panels (facets) arranged horizontally, with Boodmeal titer mapped to the \(y\)-axis.

• I am not going to insist as I am not in favour of emphasising the significant contrasts, and I am not going to link again to the related #Statstips #18, as that is not today’s job.

  Instead, I just de-emphasised the annotations a bit by using a light grey colour. As in the previous case study the re-arrangement of mappings allowed the highlighted contrasts to remain entirely within each panel.

• Finally, some rewording of the figure caption is needed to fix a few things.

(a) Distribution of observed USUV viral load in Culex pipiens by tissue, lineage and bloodmeal titer. Starred grey segments indicate significant differences across groups with \(p\)-value \(< 0.001\), according to the model <2.1>

Figure 2: Reworked figure and caption. Remember to replace <2.1> by the appropriate model name, or number.

While this figure is very similar to the one in the previous post, it is interesting how applying the same principles led to some different decisions (e.g. mapping tissue to panels rather than ordinates).

These mappings should be consistent across figures in the same manuscript, in the same way as you would keep the same colour codes for lineages.

Before

After

Figure 3: Side to side comparison

Code