Is there a way in R's ggplot2 to incorporate the stat summary argument into the legend with geom_line?

Question

I have a dataset with SUD treatment rates for each state for each year from 2008 to 2017. Two states received an intervention and the rest did not. I am able to plot the treatment rates for each intervention state as a separate line and the non-intervention states (grouped as one line using the mean) on the same graph.

I would like to include the grouped line in the legend, but am having trouble doing so. So far only the intervention states are labeled. I've pasted a subset of the data (not exactly what the plot shows, but provides insight into the structure) and provided the ggplot code with the resulting figure. Any help would be appreciated. Thank you!

structure(list(statename = c("Alabama", "Alabama", "Alabama", 
"Alabama", "Alabama", "Alabama", "Alabama", "Alabama", "Alabama", 
"Alabama", "Colorado", "Colorado", "Colorado", "Colorado", "Colorado", 
"Colorado", "Colorado", "Colorado", "Colorado", "Colorado", "Iowa", 
"Iowa", "Iowa", "Iowa", "Iowa", "Iowa", "Iowa", "Iowa", "Iowa", 
"Iowa", "Washington", "Washington", "Washington", "Washington", 
"Washington", "Washington", "Washington", "Washington", "Washington", 
"Washington"), YEAR = c(2008L, 2009L, 2010L, 2011L, 2012L, 2013L, 
2014L, 2015L, 2016L, 2017L, 2008L, 2009L, 2010L, 2011L, 2012L, 
2013L, 2014L, 2015L, 2016L, 2017L, 2008L, 2009L, 2010L, 2011L, 
2012L, 2013L, 2014L, 2015L, 2016L, 2017L, 2008L, 2009L, 2010L, 
2011L, 2012L, 2013L, 2014L, 2015L, 2016L, 2017L), RML_ever_state = c(0L, 
0L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 
1L, 1L, 1L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 1L, 1L, 1L, 
1L, 1L, 1L, 1L, 1L, 1L, 1L), TotalAdolescent_noprior_Admissionrate = c(4.07959747971533, 
17.1095315089946, 28.605586683907, 11.488423601983, 14.9233623164375, 
13.0325736612464, 11.5686458431041, 15.2887154594351, 18.9275946907701, 
17.9164783752891, 28.9653195613967, 30.542173819128, 33.8611241088185, 
29.9657748758525, 25.214146698236, 22.847991066509, 21.975843495247, 
21.783383749025, 25.8868468603421, 23.6029880132029, 45.9863203727017, 
51.6710909784629, 61.713238062499, 48.3292305542656, 52.4339028740025, 
54.9353761700907, 50.2282357945714, 46.8358255273624, 63.8611763171622, 
63.7710085311979, 50.5294974023959, 53.0407358054702, 54.7486826331652, 
59.981622172154, 57.2193036593259, 54.5478493207391, 49.924220486418, 
41.9396870928129, 11.3163228651143, 0.258492875751707)), row.names = c(1L, 
2L, 3L, 4L, 5L, 6L, 7L, 8L, 9L, 10L, 31L, 32L, 33L, 34L, 35L, 
36L, 37L, 38L, 39L, 40L, 117L, 118L, 119L, 120L, 121L, 122L, 
123L, 124L, 125L, 126L, 387L, 388L, 389L, 390L, 391L, 392L, 393L, 
394L, 395L, 396L), class = "data.frame")

ggplot(vars1, aes(x = YEAR, y = TotalAdolescent_noprior_Admissionrate, group = statename))  +
  stat_summary(
    data = ~ subset(., RML_ever_state == 0),
    aes(group = -1),
    fun.data = mean_se,
    geom = "line", size = 3, colour = "green"
  ) +
  geom_line(
    data = ~ subset(., RML_ever_state == 1),
    aes(colour = statename)
  )   + scale_x_discrete( limits=c(2008,2009,2010,2011,2012,2013,2014,2015,2016,2017))

Answer 1

A nice solution for this can be realized by understanding how ggplot2 creates legends. We will utilize that to our advantage to force a legend out of stat_summary(). Here, we're going to effectively "hijack" another aesthetic, and then squish the legends together to make it look like it's all one legend.

There are two points to go over to help explain why the final answer is working.

Legend creation inside aes()

First, ggplot2 will create a legend for anything that needs to be discriminated when inside aes(). So, if you have an aesthetic like color, fill, or size inside aes(), you're going to start the process of having ggplot2 make a legend. This process is overwritten when you also define the aesthetic outside aes().

So this code generates a legend:

ggplot(df, aes(x,y)) + geom_point(aes(size=my_category))

But this one does not:

ggplot(df, aes(x,y)) + geom_point(aes(size=my_category), size=3)

So, this means that if you wanted to define the size in the first example above, you would have to take whatever ggplot2 does be default or specify the actual size by adding a scale_size_manual() command:

ggplot(df, aes(x,y)) + geom_point(aes(size=my_category)) + scale_size_manual(values=3)

Quoted stuff vs. column names inside aes()

You can also control the names of the items in your legend by what you send to aes(). Take the above example. If I put aes(size=my_category), this assumes that there is a column in df called my_category with values. If that column is a factor, the levels of the factor will be used as the legend item names and the particular "size" values in the final plot. If the column is a character vector, the legend will be created by kind of "forcing" the column into a factor and then items will be listed alphanumerically. If df$my_category is a number vector, you will get a continuous scale of size. The game changes a bit when you supply a character/string. Let's assume df$my_category is a character vector. When I send it this:

ggplot(df, aes(x,y)) + geom_point(aes(size="my_category")...)

I'll get everything the same size, and a legend with one item called "my_category". Interesting. Basically, you are supplying the full contents of the column in the second example, where every item is labeled as "my_category". Make sense? We can put this logic together to form your solution.

A Solution to Your Question

Now, I'm going to use this information to create your plot. To separate out that stat_summary() into a legend item, I need to pick an aesthetic. You could just specify that color= is inside aes() for stat_summary and use a character to label it correctly. That will work; however, if you do that all the legend keys (called "glyphs") will look the same as the line in stat_summary. That's fine if all the lines look the same, but your summary line is "fatter", so you want it to be different than the rest.

Here, I'll just create a legend using size, which will be a different legend from the color one used by geom_line(), and they can have their different formats without any issue.

Since I'm pulling size into aes() for stat_summary, I need to remove size outside of aes() in stat_summary and put that value inside a separate scale_size_manual() command. Second, I use guides() to control the order of the legends and some title stuff. Finally, I use theme elements to control the spacing between the two legends to push them a bit closer. The final result:

ggplot(vars1, aes(x = YEAR, y = TotalAdolescent_noprior_Admissionrate, group = statename))  +
  stat_summary(
    data = ~ subset(., RML_ever_state == 0),
    aes(group = -1, size='Summary'),
    fun.data = mean_se,
    geom = "line", colour = "green"
  ) +
  geom_line(
    data = ~ subset(., RML_ever_state == 1),
    aes(colour = statename)
  )   +
  scale_x_discrete( limits=c(2008,2009,2010,2011,2012,2013,2014,2015,2016,2017)) +
  scale_size_manual(values=3) +
  guides(
    color=guide_legend(title='Legend', order=1),
    size=guide_legend(title=NULL, order=2)
  ) +
  theme(
    legend.margin = margin(t=0,b=0),
    legend.spacing.y = unit(0,'pt'),
    legend.title = element_text(margin=margin(b=10))
  )