Violin plots appear like they are not violins
envest opened this issue · 5 comments
Exploration: can we adjust violin plot line thickness such that it does not obscure the shape of the distributions
@jaclyn-taroni I made the outlines less thick to highlight the shapes of each violin... Thoughts? I can propagate this change to other plots.
Compared to the original
That does make them look more like violin plots in my opinion. Do they look enough like violin plots in every case to read as violin plots? I'm not sure.
Summary of slack discussion with @jaclyn-taroni
- Most important things to communicate: median values and horizontal shape (flat good, slanted less good)
- Variability between repeats: violin shape uninformative, try a confidence interval around median
- Jackie referenced this SO conversation
The notches in a box plot correspond to a ROUGH 95% confidence interval around median: median +/- 1.58*IQR/sqrt(n)
where IQR is inter-quartile range, and n is the number of observations (ideally >= 10, which we have)
Here is another draft, showing error bars for 95% CI around median. It also has reduced weight of the panel grid lines.
