Displaying proper colors in legend for multiple layers

Question

I am an absolute beginner and it's been about 2-3 days since I have started using ggplot2. So far, I have always used Excel for graphs. ggplot2 is really killing me, so I thought of posting my query here.

Last night, I discussed how we can plot geom_smooth() with another layer, say geom_point() This is discussed here: Scale for aesthetics used in the plot | ggplot2

In continuation to this, I thought of trying out multiple geom_smooth().

Here's what I did:

   ggplot(mpg, aes(displ, hwy)) +
     geom_point(aes(color = class)) +
     geom_smooth(method = "loess", se = FALSE, color = "black", aes(linetype = "loes")) +
     geom_smooth( method = "lm", se = FALSE, color = "red", aes(linetype = "lm",color = "green")) +
     labs(colour = "Method")

It's similar code to the previous one except that I have added another geom_smooth().

The output is:

I also looked at Format legend for multiple layers ggplot2 It seems I could manually override colors.

As we can see, the third layer still overrides the colors of the second layer (in the legend).

So, here's what I did:

   ggplot(mpg, aes(displ, hwy)) +
     geom_point(aes(color = class)) +
     geom_smooth(method = "loess", se = FALSE, color = "123", aes(linetype = "loes")) +
     geom_smooth( method = "lm", se = FALSE, color = "345", aes(linetype = "lm",color = "green")) +
    scale_colour_manual(values=c("coral", "chocolate", "cornsilk", "papayawhip", "blanchedalmond","red","black","yellow","pink")) +
     labs(colour = "Method") 

The third layer still overrides the colors of the second layer (in the legend). I'd appreciate your help.

I have two questions:

Question 1: Is there any fix for the questions I have posted above? I'd appreciate any thoughts. Is there any fix for this? I'd appreciate any thoughts.

Question 2: I noticed that sometimes people use aes(linetype = "lm") and other times they simply use (linetype = "lm") inside geom_smooth(). Why do we do this? I believe if we use aes(..) I don't have a clear hypothesis here, so I would avoid speculating. I'd appreciate your thoughts.

---

Update: My question is about the posted solution.

Can we not use any other shape for scatter plot ? The posted solution recommends changing the shape to size = 21, which is something I am a little uncomfortable.

I changed the code (in solution below) for other shape as below:

 huron <- data.frame(year = 1875:1972, level = as.numeric(LakeHuron))
   ggplot(mpg, aes(displ, hwy)) +
     # map geom_point class to 'fill'
     geom_point(shape=5, aes(color = class)) +
     # use color and linetype for geom_smooth
     geom_smooth(method = "loess", se = FALSE,
                 aes(linetype = "loess", color = 'loess')) +
     geom_smooth(method = "lm", se = FALSE, 
                 aes(linetype = "lm", color = "lm")) +
     # merge linetype and color legends by giving them the same name
     scale_linetype_discrete(name = "Method") +   
     scale_color_manual(name = "Method", values = c("red", "black","coral", "chocolate", "cornsilk", "papayawhip", "blanchedalmond","red","black"))

However, after running this code, we will see that the color for lm and loess has got reset to blue and legend for scatter plot is no more solid-type. I was able to change the shape, but not the color issue and legend issue. Any thoughts?

Answer 1

Use fill and a hollow shape for geom_point, and color for geom_smooth.

huron <- data.frame(year = 1875:1972, level = as.numeric(LakeHuron))
ggplot(mpg, aes(displ, hwy)) +
  # map geom_point class to 'fill'
  geom_point(shape=21, aes(fill = class), color = NA) +
  # use color and linetype for geom_smooth
  geom_smooth(method = "loess", se = FALSE,
              aes(linetype = "loess", color = 'loess')) +
  geom_smooth(method = "lm", se = FALSE, 
              aes(linetype = "lm", color = "lm")) +
  # merge linetype and color legends by giving them the same name
  scale_linetype_discrete(name = "Method") +   
  scale_color_manual(name = "Method", values = c('red', 'black'))

However, I would also point out that the different colors for the smooth lines is sort of distracting, if you want color information to serve to differentiate the point classes. I think it would be better to leave both smooth lines black -- linetype is enough to distinguish them