How to create a geom_line plot with reversed (second) y-axis barplot in ggplot?

Question

I am trying to plot a geom_line figure, combined with a barplot. The barplot will have a reversed y-axis. The final product is expected to be similar to the figure below. I am not sure how to combine the reversed barplot with the line plot.

Also, it would be great if we can have a legend for A, B, and C. Thanks for any help.

library(ggplot2)
Mydata = data.frame(ID=1:14, A=1:14, B=20:7, C=100:113)

ggplot(data = Mydata) + 
geom_line(aes(x=ID, y = A)) +
geom_line(aes(x=ID, y = B))

I have tried the following code. The second y-axis is now reversed from 90 to 130, however, the barplot is still not reversed.

ggplot(data = Mydata) + 
geom_line(aes(x=ID, y = A)) +
geom_line(aes(x=ID, y = B)) + 
geom_bar(stat="identity", aes(x=ID, y = (190-C)/5)) + 
scale_y_continuous(breaks=seq(0,30,5), sec.axis = sec_axis(trans=~190-.*5, breaks=seq(0,130,10)))

Answer 1

Update

If you wanted the appearance like the plot on the left instead of the plot on the right:

Then all you need is rev() in geom_col:

geom_col(aes(x = ID, y = rev(-(190-C)/5)))

Let me know if I've still misunderstood what you're looking for.

---

Original answer

If you want the bars to move down, they have to contain negative values.

Some of this you may already know.

When you add a second y-axis, it's really just a visual; none of your traces are actually following that axis. The 'trick' is the transformation.

If you want the bars to be above the lines as if there is a zero axis for the bars at the top and a zero axis on the bottom for the lines, some things need to change.

1. bars need to be negative to go down; that drives everything else

2. the lines need to be less than the values in the bars without losing the aspect ratio (for lack of a better term)

The easiest way to transform the y in the lines to ensure that the bars do not overlap is to offset both traces by negating the current y and offset that by the largest value that appears in the bar trace.

First, find that offset value:

offset = max((190 - Mydata$C)/5)  # how much to offset the other traces

Check out what we've got so far, with the y's negated and offset:

offset = max((190 - Mydata$C)/5)
ggplot(data = Mydata) + 
  geom_line(aes(x = ID, y = -A - offset)) +
  geom_line(aes(x = ID, y = -B - offset)) +
  geom_col(aes(x = ID, y = -(190-C)/5))

Next, we need to fix both y-axes. Starting with the y-axis on the left, we need this to look like we didn't negate everything. To do that, we need to find the actual negative value that reflects the perceived zero on the y-axis.

Since column B has the larger value (at 20), I used it to determine where the bottom is. Additionally, since the lowest value between A and B is 1, we're going to add -1 to get that zero line. To find the bottom or where we want the zero-line, find the minimum of negated column B - offset minus 1.

btm = min(-1 * Mydata$B - offset - 1) # because the min value is 1

Check out where we're at so far:

btm = min(-1 * Mydata$B - offset - 1) # because the min value of B is 1
offset = max((190 - Mydata$C)/5)
ggplot(data = Mydata) + 
  geom_line(aes(x = ID, y = -A - offset)) +
  geom_line(aes(x = ID, y = -B - offset)) +
  geom_col(aes(x = ID, y = -(190-C)/5)) +
  scale_y_continuous(breaks = seq(btm, by = 10, length.out = 5), 
                     labels = seq(0, 40, 10))

It's time to add the secondary y-axis.

I flipped the sign on the transformation and then looked at where the numbers landed on the secondary y.

btm = min(-1 * Mydata$B - offset - 1) # because the min value of B is 1
offset = max((190 - Mydata$C)/5)
ggplot(data = Mydata) + 
  geom_line(aes(x = ID, y = -A - offset)) + 
  geom_line(aes(x = ID, y = -B - offset)) +
  geom_col(aes(x = ID, y = -(190-C)/5)) +
  scale_y_continuous(breaks = seq(btm, by = 10, length.out = 5), 
                     labels = seq(0, 40, 10),
                     sec.axis = sec_axis(~190 + . * 5))  # changed sign!

Next, we need to determine where our perceived zero is at the top on the secondary y-axis. To calculate this, we need the max of column C, when calculated as it is in the plot and transformed as the axis is transformed, then altered by the largest value of C, The largest value of B, and the offset already calculated.

top = max(((190 + (-(190 - Mydata$C)/5) * 5) + offset - btm + max(Mydata$B)))

Note that in this call to create this value:

The Y in the bars is:

-(190 - Mydata$C)/5

The Y in the bars transformed by the sec axis is:

190 + (-(190 - Mydata$C)/5) * 5

And there you go! You can adjust the labels now that it's ready however you see fit. (I'm guessing you have something in mind.)

btm = min(-1 * Mydata$B - offset - 1) # because the min value of B is 1
offset = max((190 - Mydata$C)/5)
top = max(((190 + (-(190 - Mydata$C)/5) * 5) + offset - btm + max(Mydata$B))) 

ggplot(data = Mydata) + 
  geom_line(aes(x = ID, y = -A - offset)) +
  geom_line(aes(x = ID, y = -B - offset)) +
  geom_col(aes(x = ID, y = -(190-C)/5)) +
  scale_y_continuous(breaks = seq(btm, by = 10, length.out = 5), 
                     labels = seq(0, 40, 10),
                     sec.axis = sec_axis(~190 + . * 5,  # changed sign!
                                         breaks = seq(top - 20 * 4, top, by = 20),
                                         labels = seq(100, 0, by = -25)))

Answer 2

You might try base graphics, where this is quite straightforward.

## base plane with the lines
par(mar=c(5, 4, 4, 3) + .1)
matplot(Mydata[2:3], type='o', pch=c(0, 4), col=1, lty=1,
        ylim=c(min(Mydata[2:3]), max(Mydata[2:3]) + 10),
        yaxt='n', xlim=c(.5, nrow(Mydata) + .5),
        xlab='xlab', ylab='ylab', main='main')
axis(2, axTicks(2), axTicks(2), tck=.02, las=2)  ## make nicer west axis
ymax <- par()$usr[4]  ## store north coord
xseq <- seq_len(nrow(Mydata))  ## store x axis
fac <- 1/15  ## define reduction factor for upper barplot
## upper barplot using `rect`
rect(xseq - .25, ymax, xseq + .25, ymax - Mydata$C*fac, 
     col='#139FFF', border=NA)
axis(3, xseq, tck=.02, labels=FALSE)  ## north x axis
ats <- seq.int(0, max(Mydata$C), by=50)  ## east ats
axis(4, y - ats*fac, ats, tck=.02, las=2)  ## east axis
## legend
legend('bottomright', legend=c('foo', 'bar', 'baz'), lty=1,
       lwd=c(1, 1, 5), pch=c(0, 4, NA), col=c(1, 1, '#139FFF'),
       bty='n', horiz=TRUE)

---

Data:

Mydata <- structure(list(ID = 1:14, A = 1:14, B = 20:7, C = 100:113), class = "data.frame", row.names = c(NA, 
-14L))

Answer 3

Hadley Wickham answers here (https://stackoverflow.com/a/3101876/19594737) that seperate y axis scales aren't supported by ggplot, because he believes they're fundamentally flawed.

That said, using the gridExtra package, you can get pretty close. Basically you have to make the two charts separately, then attach them in a grid.

Here's the output...

...and here's the input:

library(tidyverse)
library(gridExtra)

Mydata = data.frame(ID=1:14, A=1:14, B=20:7, C=100:113)

lines<-Mydata%>%
  select(-C)%>%
  gather(key=Lines,value=vals,-ID)

bars<-Mydata%>%
  select(ID,C)


b<-ggplot()+
  geom_bar(data=bars,  aes(ID,C),stat="identity",fill='blue')+
  scale_y_reverse()+
  theme_minimal()+
  theme(axis.ticks.x = element_blank(),
        axis.text.x = element_blank())+
  labs(x="",y="Y Axis Label")+
  scale_x_continuous(limits = c(0,15))

l<-ggplot()+
    geom_line(data=lines,aes(ID,vals,color=Lines))+
  theme_minimal()+
  theme(legend.position="bottom")+
  scale_x_continuous(limits = c(0,15))+
  labs(y="Second Y axis Label")

grid.arrange(b,l,nrow=2)