ggplot: how to assign both color and shape for one factor, and also shape for another factor?

Question

I must code factor age with both color and shape. I know how to do that (see the plot and data/code below).

In addition, I also have to code factor day with shape.

Is it somehow possible to assign specified different shapes to two different factors?

Below is the legend I would love to achieve (I made an example in power point). The plot is not quite right, as only factor age is coded with color and shape.

df = data.frame(test = c(1,2,3, 1,2,3, 1,2,3,  1,2,3, 1,2,3, 1,2,3),
                age = c(1,1,1, 2,2,2, 3,3,3,   1,1,1, 2,2,2, 3,3,3),
                day = c(1,1,1,1,1,1,1,1,1, 2,2,2,2,2,2,2,2,2),
                result = c(1,2,2,1,1,2,2,1,0, 2,2,0,1,2,1,2,1,0))

df$test <- factor((df$test))
df$age <- factor((df$age))
df$day <- factor((df$day))

windows(width=4, height=3 )

df %>%
  ggplot( aes(x=test, y=result)) +
  geom_point(aes(color=age, shape=age, group=age), 
             position=position_jitterdodge(dodge.width=0.8,jitter.height=0.2, jitter.width=0), 
             size=2, stroke=0.8)  +
  scale_shape_manual(values=c(16, 15, 17), name="", labels=c("young","older","the oldest")) +
  scale_color_manual(name="", labels=c("young","older","the oldest"), values=c('#009E73','#56B4E9','#D55E00')) +
  
  theme_bw()+
  theme(panel.border = element_blank(), axis.ticks = element_blank(),
        legend.position=c(), legend.text=element_text(size=10, face="bold"), legend.title=element_text(size=10),
        panel.grid.major.x = element_blank() ,
        panel.grid.major.y = element_blank() ,
        plot.title = element_text(size=10, face = "bold"), axis.title=element_text(size=11),
        axis.text.y = element_text(size=9, angle = 45), 
        axis.text.x = element_text(size=9, angle = 90),
        plot.margin = unit(c(0.5,0.2,0,0), "cm")) +
  
  labs(title= "", x = "",y = "Test result") +
  scale_y_continuous(breaks=c(0,1,2), labels=c('good','better','the best')) +
  geom_vline(xintercept=c(0.5,1.5,2.5),color="grey90")+
  geom_hline(yintercept=-0.5, color="grey90")+
  expand_limits(x=3.9, y=c(0,2.35)) +
  scale_x_discrete(limits = c("1", "2", "3"),labels = c("test a", "test b", "test c")) + 
  coord_cartesian(clip = "off")

tjebo · Accepted Answer

You can use shapes on an interaction between age and day, and use color only one age. Then remove the color legend and color the shape legend manually with override.aes.

This comes close to what you want - labels can be changes, I've defined them when creating the factors.

how to make fancy legends

However, you want a quite fancy legend, so the easiest would be to build the legend yourself as a separate plot and combine to the main panel. ("Fake legend"). This requires some semi-hardcoding, but you're not shy to do this anyways given the manual definition of your shapes. See Part Two how to do this.

Part one

library(ggplot2)
df = data.frame(test = c(1,2,3, 1,2,3, 1,2,3,  1,2,3, 1,2,3, 1,2,3),
                age = c(1,1,1, 2,2,2, 3,3,3,   1,1,1, 2,2,2, 3,3,3),
                day = c(1,1,1,1,1,1,1,1,1, 2,2,2,2,2,2,2,2,2),
                result = c(1,2,2,1,1,2,2,1,0, 2,2,0,1,2,1,2,1,0))

df$test <- factor(df$test)
## note I'm changing this here already!! If you udnergo the effor tof changing to
## factor, define levels and labels here
df$age <- factor(df$age, labels = c("young", "older", "the oldest"))
df$day <- factor(df$day, labels = paste("Day", 1:2))

ggplot(df, aes(x=test, y=result)) +
  geom_jitter(aes(color=age, shape=interaction(day, age)),
              width = .1, height = .1) +
  ## you won't get around manually defining the shapes
  scale_shape_manual(values = c(0, 15, 1, 16, 2, 17)) +
  scale_color_manual(values = c('#009E73','#56B4E9','#D55E00')) +
  guides(color = "none", 
         shape = guide_legend(
          override.aes = list(color = rep(c('#009E73','#56B4E9','#D55E00'), each = 2)),
          ncol = 3))

Part two - the fake legend

library(ggplot2)
library(dplyr)
library(patchwork)

## df and factor creation as above !!!

p_panel <- 
ggplot(df, aes(x=test, y=result)) +
  geom_jitter(aes(color=age, shape=interaction(day, age)),
              width = .1, height = .1) +
  ## you won't get around manually defining the shapes
  scale_shape_manual(values = c(0, 15, 1, 16, 2, 17)) +
  scale_color_manual(values = c('#009E73','#56B4E9','#D55E00')) +
  ## for this solution, I'm removing the legend entirely
  theme(legend.position = "none") 

## make the data frame for the fake legend
## the y coordinates should be defined relative to the y values in your panel
y_coord <- c(.9, 1.1)
df_legend <- df %>% distinct(day, age) %>% 
  mutate(x = rep(1:3,2), y = rep(y_coord,each = 3))

## The legend plot is basically the same as the main plot, but without legend -
## because it IS the legend ... ;)
lab_size = 10*5/14
p_leg <- 
  ggplot(df_legend, aes(x=x, y=y)) +
  geom_point(aes(color=age, shape=interaction(day, age))) +
  ## I'm annotating in separate layers because it keeps it clearer (for me)
  annotate(geom = "text", x = unique(df_legend$x), y = max(y_coord)+.1, 
           size = lab_size, angle = 45, hjust = 0,
           label = c("young", "older", "the oldest")) +
  annotate(geom = "text", x = max(df_legend$x)+.2, y = y_coord, 
           label = paste("Day", 1:2), size = lab_size, hjust = 0) +
  scale_shape_manual(values = c(0, 15, 1, 16, 2, 17)) +
  scale_color_manual(values = c('#009E73','#56B4E9','#D55E00')) +
  theme_void() +
  theme(legend.position = "none",
        plot.margin = margin(r = .3,unit = "in")) +
  ## you need to turn clipping off and define the same y limits as your panel
  coord_cartesian(clip = "off", ylim = range(df$result))

## now combine them
p_panel + p_leg +
  plot_layout(widths = c(1,.2))

ggplot: how to assign both color and shape for one factor, and also shape for another factor?

Answers (2)

Related Questions