geom_abline does not seem to respect groups in facet_grid [ggplot2]

Question

Just trying to understand how geom_abline works with facets in ggplot.

I have a dataset of student test scores. These are in a data table dt with 4 columns:

student: unique student ID
cohort:  grouping factor for students (A, B, … H)
subject: subject of the test (English, Math, Science)
score:   the test score for that student in that subject

The goal is to compare cohorts. The following snippet creates a sample dataset.

library(data.table)
## cohorts: list of cohorts with number of students in each
cohorts <- data.table(name=toupper(letters[1:8]),size=as.numeric(c(8,25,16,30,10,27,13,32)))
## base: assign students to cohorts
base    <- data.table(student=c(1:sum(cohorts$size)),cohort=rep(cohorts$name,cohorts$size))
## scores for each subject
english <- data.table(base,subject="English", score=rnorm(nrow(base), mean=45, sd=50))
math    <- data.table(base,subject="Math",    score=rnorm(nrow(base), mean=55, sd=25))
science <- data.table(base,subject="Science", score=rnorm(nrow(base), mean=70, sd=25))
## combine
dt      <- rbind(english,math,science)
## clip scores to (0,100)
dt$score<- (dt$score>=0) * dt$score
dt$score<- (dt$score<=100)*dt$score + (dt$score>100)*100

The following displays mean score by cohort with 95% CL, facetted by subject, and includes a (blue, dashed) reference line (using geom_abline).

library(ggplot2)
library(Hmisc)
ggp <- ggplot(dt,aes(x=cohort, y=score)) + ylim(0,100)
ggp <- ggp + stat_summary(fun.data="mean_cl_normal")
ggp <- ggp + geom_abline(aes(slope=0,intercept=mean(score)),color="blue",linetype="dashed")
ggp <- ggp + facet_grid(subject~.)
ggp

The problem is that the reference line (from geom_abline) is the same in all facets (= the grand average score for all students and all subjects). So stat_summary seems to respect the grouping implied in facet_grid (eg, by subject), but abline does not. Can anyone explain why?

NB: I realize this problem can be solved by creating a separate table of group means and using that as the data source in geom_abline (below), but why is this necessary?

means <- dt[,list(mean.score=mean(score)),by="subject"]
ggp <- ggplot(dt,aes(x=cohort, y=score)) + ylim(0,100)
ggp <- ggp + stat_summary(fun.data="mean_cl_normal")
ggp <- ggp + geom_abline(data=means, aes(slope=0,intercept=mean.score),color="blue",linetype="dashed")
ggp <- ggp + facet_grid(subject~.)
ggp

Answer 1

This should do what you want. The stat_* functions use different collections of data for each facet. I think any expressions in the aes of the geom_* functions are intended to be used for the transformation of each y-value.

ggplot(dt,aes(x=cohort, y=score)) +
       stat_summary(fun.data="mean_cl_normal") + 
       stat_smooth(formula=y~1,aes(group=1),method="lm",se=FALSE) +
       facet_grid(subject~.) + ylim(0,100)

Answer 2

As golbasche mentioned, I would have probably done something more like this:

dt <- dt[,avg_score := mean(score),by = subject]

ggplot(dt,aes(x=cohort, y=score)) + 
    facet_grid(subject~.) + 
    stat_summary(fun.data="mean_cl_normal") +
    geom_hline(aes(yintercept = avg_score),color = "blue",linetype = "dashed") + 
    ylim(0,100)