Summary statistics (mean, median, ...) in type_violin?

[zeileis]

Is there a sufficiently easy way to add summary statistics like mean or median to violin plots? If not, then maybe this could be added to type_violin() itself?

At first, I thought that type_summary() could be used for this but this is not fully straightforward because it

• does not yet support factor x variables, see Support for categorical variables in type_summary() #436
• handles by variables without juxtaposition

Examples:

In the case with numeric x the violin and the points are misaligned (see left panel) because type_violin always treats x as factor and type_summary always expects numeric:

tinyplot(len ~ dose, data = ToothGrowth, fill = 0.2,  type = "violin")
tinyplot_add(type = type_summary(median, type = "p"))

Using a factor x is currently not possible in type_summary:

tinyplot(len ~ factor(dose), data = ToothGrowth, fill = 0.2,  type = "violin")
tinyplot_add(type = type_summary(median, type = "p"))
## Error in Summary.factor(c(1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L,  : 
##   'range' not meaningful for factors

Only when the numeric values are 1, ..., k then violin and summary points are aligned (center panel):

tinyplot(len ~ as.numeric(as.factor(dose)), data = ToothGrowth, fill = 0.2,  type = "violin")
tinyplot_add(type = type_summary(median, type = "p"))

However, even in that case, by groups are handled differently with vs without juxtaposition (right panel):

tinyplot(len ~ as.numeric(as.factor(dose)) | supp, data = ToothGrowth, fill = 0.2,  type = "violin", legend = FALSE)
tinyplot_add(type = type_summary(median, type = "p"))

I guess this would indicate that adding some summary option to type_violin directly would be best?

Aside: Setting pch does not seem to be supported in type_summary?

[grantmcdermott]

I guess this would indicate that adding some summary option to type_violin directly would be best?

Hmmm, agree. "Raincloud" versions would be nice too, but that might have to be a separate, dedicated type.

We should fix some of those other issues that you've identified here too.

[zeileis]

OK, thanks for the feedback! Unfortunately, I won't have any time to look at this soon (first exams and then a couple of days off) but wanted to post this because I was asked about it today by one of my colleagues.

[grantmcdermott]

All good. I'm on vacation now too... Managed to sneak off that last submission with my laptop battery deep in the red (zero power coord = forced screen time constraint).