When overlaying two response variables on a single plot, you may want to give each variable its own Y-axis, especially if the scale or unit of the axis is not suited for both variables at the same time.

Adding a secondary Y-axis in ggplot is not complicated. The argument sec.axis = sec_axis() in scale_y_continuous() is indeed able to add one to the right side of any plot. Here we will see how this works with a simple example where one response variable (response1) is represented in the form of a bar plot, and another response variable (response2) is represented in the form of a line plot. Both response variables will eventually be plotted against the same predictor variable (ID). Here we first have a look at the response variables in two separate plots:

ggplot(df, aes(x = ID)) +                                                        # bar plot - response1 vs. ID
  geom_col(aes(y = response1), size = 1, color = "darkblue", fill = "white") +
  scale_x_continuous(breaks=seq(1, 12, 1))
ggplot(df, aes(x = ID)) +                                                        # line plot - response2 vs. ID
  geom_line(aes(y = response2), size = 1.5, color="red") +
  scale_x_continuous(breaks=seq(1, 12, 1))

For each of the plots above, ggplot defines the scale of the Y-axis that fits best, i.e. almost the highest amplitude possible, where all data points are included. Now, let’s put the two response variables in the same plot:

ggplot(df, aes(x = ID)) + 
  geom_col(aes(y = response1), size = 1, color = "darkblue", fill = "white") +   # bar plot - response1 vs. ID
  geom_line(aes(y = response2), size = 1.5, color="red")+                        # line plot - response2 vs. ID
  scale_x_continuous(breaks=seq(1, 12, 1))

As you realize here, ggplot has kept the scale of the Y-axis that fits best the series with the highest amplitude. The result is of course nice when considering the variable response1 (i.e. the bar plot), but that scale is not optimal for the line plot showing response2 as the line is flatened at the bottom of the chart.

Before trying to fit the secondary axis, we may try to improve the display of response2 so that it makes a better use of the plot area. To do so, we shall apply an arithmetic transformation to response2 so that we change the amplitude of the display of the line plot. Here, we can for instance multiply the whole series by 2 so that it “uses more space” in the present plot. To do so, we just have to add 2* in front of response2 in aes():

ggplot(df, aes(x = ID)) + 
  geom_col(aes(y = response1), size = 1, color = "darkblue", fill = "white") +
  geom_line(aes(y = 2*response2), size = 1.5, color="red") +
  scale_x_continuous(breaks=seq(1, 12, 1))

Now, the line plot has better proportions compared to the the bar plot. We may thus add the secondary Y-axis to the right using scale_y_continuous(sec.axis = sec_axis(~.)). However, if we do so, the secondary axis that will appear will be identical to the primary axis. Its scale won’t be fitted for the transformed variable response2. We thus need to apply an arithmetical transformation to the scale of the axis that will compensate for the transformation applied to the variable. Logically, since the variable was multiplied by 2, we need to compensate by dividing the scale by 2. To do so, we write ~./2 between the parentheses:

ggplot(df, aes(x = ID)) + 
  geom_col(aes(y = response1), size = 1, color = "darkblue", fill = "white") +
  geom_line(aes(y= 2*response2), size = 1.5, color="red") +
  scale_x_continuous(breaks=seq(1, 12, 1)) +
  scale_y_continuous(sec.axis = sec_axis(~./2))      # adds the secondary Y-axis

The resulting plot now has a secondary Y-axis which fits the variable response2. The only thing that is missing to finish the job is to add a title to the secondary axis. To do so, we add in scale_y_continuous() the argument name= followed by the text to display:

ggplot(df, aes(x = ID)) + 
  geom_col(aes(y = response1), size = 1, color = "darkblue", fill = "white") +
  geom_line(aes(y= 2*response2), size = 1.5, color="red") +
  scale_x_continuous(breaks=seq(1, 12, 1)) +
  scale_y_continuous(sec.axis = sec_axis(~./2, name = "response2"))

---