ggplot2 assigning aesthetics falsely

Question

I am creating plots with ggplot2 and for some reason the function is acting really weirdly.

I have a dataframe df, and I want to visualize several columns.

Any dataframe seems to work fine. I have generated this dummy dataframe.

df <- data.frame(Date = seq.Date(as.Date.character("2019-01-01"), by = 1, length.out = 10), 
                 Value = rnorm(10), 
                 Foo = rnorm(10))

So what I do is

  library(ggplot2)
  gg <- ggplot(df, aes(x = Date)) + geom_line(aes(y = Value, color = "Value", linetype = "Value"))
  gg <- gg + geom_line(aes(y = Foo, color = "SomeWord", linetype = "SomeWord"))
  gg <- gg + scale_color_manual(name="Legend", 
      breaks=c("Value", "SomeWord"), values=c("steelblue", "firebrick")) + 
    scale_linetype_manual(name="Legend", 
      breaks=c("Value", "SomeWord"), values=c("solid", "twodash"))
  gg

Normally, ggplot2 would now correctly assign color steelblue and linetype solid to the column Value, while assigning firebrick and twodash to the Foo column, which I assigned the name SomeWord to. However, depending on what I choose for the name, ggplot assigns the colours and linetypes in the wrong way. For example, using "Test1" as a name seems to work just fine, but "Einschritt" causes ggplot2 to throw my entire ruleset out of the window.

I have tried googling this, but have not found any clue on why ggplot seems not to accept some names, while others are just fine. I would also like to use Hyphens in the color and linetype reference name which I assume might be a problem.

Edit: As an example, I have just tried replicating this on my dummy data frame. Using the code posted above, when I use the following names, the linetype and colour are matched falsely:

• "Value" for column "Value", anything for column Foo.
• "Ein-Schritt-Prognose" for column "Value", anything for column Foo.
• "SomeWord" for column "Value", anything for column Foo.

However, when I switch to something like:

• "ABD" for column Value, anything for column Foo.

then they are matched correctly.

Answer 1

"Ein-Schritt-Prognose" dose not work as colnames. Please see my comment below your question. In ggplot2, colnames are not quoted, thus using hyphen in colname will make it look like Ein - Schritt - Prognose (an expression). Use hyphen with caution in R.

Answer 2

First of all, just to make it clear: hyphens have nothing to do with this.

The issue is that the breaks argument is not used to define the data-to-aesthetic mapping in scales, at all. breaks just controls which data values appear on the legend, and in which order. Nothing else.

Here’s a demonstration (simplified to only colours; the concepts are the same):

library(ggplot2)

set.seed(42)

mydf <- data.frame(
  Date = seq.Date(as.Date.character("2019-01-01"), by = 1, length.out = 10),
  Value = rnorm(10), Foo = rnorm(10)
)

p <- ggplot(mydf, aes(x = Date)) +
  geom_line(aes(y = Value, color = "Value")) +
  geom_line(aes(y = Foo, color = "SomeWord"))

p1 <- p + scale_color_manual(
  breaks = c("Value", "SomeWord"),
  values = c("steelblue", "firebrick")
)

p2 <- p + scale_color_manual(
  breaks = c("SomeWord", "Value"),
  values = c("steelblue", "firebrick")
)

egg::ggarrange(p1, p2)

As you can see, the aesthetic mapping remains the same: "Value" is still red, and "SomeWord" is still blue; only the order of the legend has changed. If you want to control the data-to-aesthetic mapping, you have two options:

First, as metioned by @markus in the comments, you can set names for the vector given as the values argument:

p + scale_color_manual(
  values = c("Value" = "steelblue", "SomeWord" = "firebrick")
)

Alternatively (although not recommended), you can rely on aesthetics being mapped in the order of the limits:

p + scale_color_manual(
  limits = c("Value", "SomeWord"),
  values = c("steelblue", "firebrick")
)

(Note that here the legend order changed, too: this is because if not given, breaks is set to limits.)

By default, the limits are sorted in alphabetical order, which is the cause of the behaviour that you saw: V comes after S, which is why (if you don’t set the limits) "Value" gets matched with the second colour, and "SomeWord" with the first.

And as to how limits differs from breaks: limits controls what data values are mapped. If we have a data value that’s not inlucded in limits, the mapped aesthetic is set to NA:

p + scale_color_manual(
  limits = c("Value"),
  values = c("steelblue", "firebrick")
)
#> Warning: Removed 10 rows containing missing values (geom_path).

Whereas if you leave a value out from breaks, all the data are still mapped, but the omitted value is not shown on the legend:

p + scale_color_manual(
  breaks = c("Value"),
  values = c("steelblue", "firebrick")
)

^{Created on 2019-07-04 by the reprex package (v0.3.0)}

Answer 3

As @HongboZhu correctly said, the problems are the hyphens. Now, your real problem is that you want to use hyphens in the legend. There are plenty of ways to change the legend labels. One way is within your scale_x_manual function.

Note I have slightly shortened your code and also changed the name of your data frame to mydf. df is a baseR function and is not recommended (although very frequently used) as an example name on SO.

mydf <- data.frame(Date = seq.Date(as.Date.character("2019-01-01"), by = 1, length.out = 10),Value = rnorm(10), Foo = rnorm(10))

library(ggplot2)
ggplot(mydf, aes(x = Date)) + geom_line(aes(y = Value, color = "Value", linetype = "Value")) +
  geom_line(aes(y = Foo, color = "SomeWord", linetype = "SomeWord")) +
  scale_color_manual(breaks=c("Value", "SomeWord"), values=c("steelblue", "firebrick"), label = c('value','Ein-SChritt-Prognose')) + 
  scale_linetype_manual(name="Legend", breaks=c("Value", "SomeWord"), values=c("solid", "twodash"))

^{Created on 2019-07-04 by the reprex package (v0.2.1)}