Add legend to ggplot histogram with different types of aesthetics

Question

I want to add a legend to one of my plots, but I have different aesthetics and I never created a legend so I find it very difficult to determine how to build it.

One of my aesthetics is a fill code, which I added manually as a vector. The other aesthetic is a vertical line that I added with geom_vline.

From the graph below, there are three characteristics that I want to add to the legend: 1) The bars with color dark blue, 2) The bars with color light blue and 3) The vertical line.

Does anyone have a suggestion for me on how to code this efficiently?

#df
df <- data.frame(Time_Diff <- runif(1000, 0, 200))


# Show median, IQR range and outliers
colors <- c(rep("blue",3), rep("paleturquoise2",38))
bp_overall <- ggplot(data = df, aes(Time_Diff)) 
bp_overall + 
  geom_histogram(binwidth = 5, fill = colors) + #create histogram
  ggtitle("Time Difference")  +
  xlab("Time in Days") +
  ylab("Amount") +
  geom_vline(xintercept = 3, linetype = "twodash", size = 1,      colour= "darkblue") + #show median
  scale_x_continuous(breaks = seq(0, 202, 10)) +
  theme_light() +
  theme(panel.grid.minor = element_blank(),
    panel.border = element_blank(), #remove all border lines
    axis.line.x = element_line(size = 0.5, linetype = "solid", colour = "black"), #add x-axis border line
    axis.line.y = element_line(size = 0.5, linetype = "solid", colour = "black")) + #add y-axis border line
  theme(plot.title = element_text(family = windowsFont("Verdana"),     color="black", size=14, hjust = 0.5)) +
  theme(axis.title = element_text(family = windowsFont("Verdana"), color="black", size=12)) 

After the suggestion of Djork I arrived to the following script, which works and I am happy with. The only thing I am trying to accomplish now is to get the Legend to be one whole (Histogram Legend and Line Legend are combined into a coherent whole). Anyone has a suggestion?

# reformat data
set.seed(1)
df <- data.frame(runif(1000, 0, 200))
colnames(df) <- "Time_Diff"

bp_overall + 
  geom_histogram(data = subset(df, Time_Diff <= 12.5), aes(x = Time_Diff, fill="BAR BLUE"), binwidth = 5) + # subset for blue data, where aes fill is fill group 1 label
  geom_histogram(data = subset(df, Time_Diff > 12.5), aes(x = Time_Diff, fill="BAR TURQUOISE"), binwidth = 5) + # subset for turquoise data, where aes fill is fill group 2 label
  scale_fill_manual("Histogram Legend", values=c("blue", "paleturquoise2")) + # manually assign histogram fill colors
  geom_vline(aes(xintercept = 3, colour="LINE DARK BLUE"), linetype="twodash", size = 1) + # where aes colour is vline label
  scale_colour_manual("Line Legend", values="darkblue") + #removed legend title
  scale_x_continuous(breaks = seq(0, 202, 10)) +
  ggtitle("Time Difference")  +
  xlab("Time in Days") +
  ylab("Amount") +
  theme_light() +
  theme(panel.grid.minor = element_blank(),
        panel.border = element_blank(), 
        axis.line.x = element_line(size = 0.5, linetype = "solid", colour = "black"), 
        axis.line.y = element_line(size = 0.5, linetype = "solid", colour = "black"),
        legend.position = c(0.95, 0.95),
        legend.justification = c("right", "top"),
        legend.box.just = ("right"))

Floccule

the easiest way to get this legend is to specify the characteristics inside aes(). Thus, you have to transform your data.frame including the statistics and a column with the colors. See here or here – Apparatus

@Jimbou I think I understand what you mean with transforming the dataframe. I have done this in my problem (see code below CONCEPT OF SOLUTION), but I can't figure out how to adjust the code in my ggplot figure – Floccule

Answer 1

I think @Jimbou's suggestion is preferable, but there is a work-around for creating legends artificially by assigning a character value to the geom_histogram aes fill value and geom_vline aes colour value, then setting colors in scale_fill_manual or scale_colour_manual.

However with this approach aes fill will only take one value (length 1) so you have to subset your df for the blue and turquoise values and plot a histogram for each, with the cutoff determined by your binwidth.

Here is the approach. Note your data needed reformatting.

# reformat data
set.seed(1)
df <- data.frame(runif(1000, 0, 200))
colnames(df) <- "Time_Diff"


bp_overall <- ggplot(data = df) 
bp_overall +
  geom_histogram(data = subset(df, Time_Diff <= 12.5), aes(x = Time_Diff, fill="BAR BLUE"), binwidth = 5) + # subset for blue data, where aes fill is fill group 1 label
  geom_histogram(data = subset(df, Time_Diff > 12.5), aes(x = Time_Diff, fill="BAR TURQUOISE"), binwidth = 5) + # subset for turquoise data, where aes fill is fill group 2 label
  scale_fill_manual("Histogram Legend", values=c("blue", "paleturquoise2")) + # manually assign histogram fill colors
  geom_vline(aes(xintercept = 3, colour="LINE DARK BLUE"), linetype="twodash", size = 1) + # where aes colour is vline label
  scale_colour_manual("Line Legend", values="darkblue") + # manually assign vline colors
  scale_x_continuous(breaks = seq(0, 202, 10)) +
  ggtitle("Time Difference")  +
  xlab("Time in Days") +
  ylab("Amount") +
  theme_light() +
  theme(panel.grid.minor = element_blank(),
    panel.border = element_blank(), 
    axis.line.x = element_line(size = 0.5, linetype = "solid", colour = "black"), 
    axis.line.y = element_line(size = 0.5, linetype = "solid", colour = "black"))

Etc. add your remaining theme.

EDITED: To answer question on how to unify legend and decrease spacing between the two legend types

(1) Remove legend name for vline by setting to "" in scale_fill_manual, change histogram fill legend name to "Legend" in scale_colour_manual.

(2) Specify order in which legends should appear, fill first then colour using guides guide_legend.

(3) Remove the y-spacing between the two legend types by setting legend.spacing.y to 0, and remove margins on top and bottom using legend.margin in theme

bp_overall <- ggplot(data = df) 
bp_overall +
  geom_histogram(data = subset(df, Time_Diff <= 12.5), aes(x = Time_Diff, fill="BAR BLUE"), binwidth = 5) + 
  geom_histogram(data = subset(df, Time_Diff > 12.5), aes(x = Time_Diff, fill="BAR TURQUOISE"), binwidth = 5) + 
  scale_fill_manual(name="Legend", values=c("blue", "paleturquoise2")) +
  geom_vline(aes(xintercept = 3, colour="LINE DARK BLUE"), linetype="twodash", size = 1) + 
  scale_colour_manual(name="", values="darkblue") + 
  scale_x_continuous(breaks = seq(0, 202, 10)) +
  ggtitle("Time Difference")  +
  xlab("Time in Days") +
  ylab("Amount") +
  theme_light() +
  theme(panel.grid.minor = element_blank(),
    panel.border = element_blank(),
    axis.line.x = element_line(size = 0.5, linetype = "solid", colour = "black"),
    axis.line.y = element_line(size = 0.5, linetype = "solid", colour = "black"),
    legend.spacing.y = unit(0, "cm"),
    legend.margin=margin(t=0, r=0.5, b=0, l=0.5, unit="cm")) +
  guides(fill = guide_legend(order = 1), 
     colour = guide_legend(order = 2))

Minnesota

Thank you for your suggestion. Because you and Jimdou both indicate that there's a more neat way to do this. I added a little piece of code to transform the dataframe to make it suitable to take up in the ggplot, but I don't know how to take it from there. Can you check the new code in my problem? – Floccule

I tried to transform my code into yours, but I get the following error: 'Error: Aesthetics must be either length 1 or the same as the data (77): x, fill, binwidth'. What do you mean with reshaping the df? – Floccule

You will have to provide your data or a subset that reproduces the error for us to troubleshoot the error, but it is likely you are give aes something longer that 1 character which is what I used above. – Minnesota

Sorry, one more question :). It would be perfect to have just one legend (not a separate Histogram Legend and Line Legend). I get one step closer by removing the "Line Legend" tekst in the scale_colour_manual line of code (replace it with ""). But still, the legend for the line is too far below the rest of the legend indications. Do you know if there's a way in which I can make the "LINE DARK BLUE" variable be one of the "Histogram Legend" guys? (at least visually) – Floccule

Please see edited answer on workaround to unify the legends. If this works for you, please accept the answer to mark it close. – Minnesota

I understand what you are doing here, but unfortunately, I get the error 'legend.spacing.y' is not a valid theme element name – Floccule

Which version of ggplot2 are you running? It run the most current ggplot2_2.2.1, otherwise check your syntax. – Minnesota