Why does ggplot size parameter not behave consistently?

[dylanjm]

A reprex to demonstrate my point. Setting geom_text(size = 12) results in a lot bigger of change than setting theme(axis.title = element_text(size = 12). It appears that the font size specified in theme() is close to what a 12pt font might resemble in a word editor. Is there a specific reason for this behavior?

library(tidyverse)

iris %>% 
  ggplot(aes(x = Sepal.Length, y = Sepal.Width)) + 
  geom_point() + 
  geom_text(data = group_by(iris, Species) %>% 
              filter(row_number(Sepal.Width) == 1),
            aes(label = Species),
            size = 12) + 
  theme(axis.title = element_text(size = 12))

[technocrat]

For no reason I'm unable to fathom, size in the context of theme and geom_text have diferrent interpretations

library(tidyverse)

iris %>% 
  ggplot(aes(x = Sepal.Length, y = Sepal.Width)) + 
  geom_point() + 
  geom_text(data = group_by(iris, Species) %>% 
              filter(row_number(Sepal.Width) == 1),
            aes(label = Species),
            size = 4) + 
  theme(axis.title = element_text(size = 12))

^{Created on 2019-01-15 by the reprex package (v0.2.1)}

(changed argument in geom_text from 12 to 4)

[rensa]

A solid discussion on how size works in ggplot2 works probably require a blog post (which I'd love to write, but this month is a difficult time for me, sorry!). But my understanding is that the most important difference here is that the size parameter in geoms:

(a) works on the data scale, not the plot scale, and
(b) uses different units depending on the shake being plotted (points, text, polygons, etc).

When I say that geoms' size works on the data scale, what I mean is that the plotting process translates data values (Sepal.Length and Sepal.Width) to "size on paper" values. Does that make sense?

(I'm by no means an expert here, mind!)

[neilcaithness]

SO throws up a good answer:

... . The unit of size in ggplot2 is mm, not pt, and I've just redefined .pt as 72 / 25.4 to hopefully make that more clear in the code – hadley Aug 6 '15 at 11:51

And I've now changed it to 72.27 / 25.4 since as pointed out in the comments, grid uses printers' pointers, not Adobe/big points. – hadley Aug 6 '15 at 11:56

ggplot2's magic number is point size / mm (i.e. 72.27 / 25.4)

So in your example:

library(ggplot2)
12 / .pt
#> [1] 4.217518

^{Created on 2019-01-15 by the reprex package (v0.2.1)}

[dylanjm]

All replies are really helpful but this one provides an answer and solution. Thanks for pointing this out!