add factor levels that are not in use

Question

I've a simple question that I cannot solve: I want to plot a data.frame (a month) with factors, where sometimes levels are missing. R attributes then only the existing levels, so my plots differ if there are one, two ore more levels present.

Here an example:

    library(ggplot2)
    library(reshape2)

f             <- factor(c("Free", "Work"))
mon           <- as.data.frame(matrix(as.factor(rep(f[2], times = 8)), nrow = 4)) 
colnames(mon) <- c("A", "B")

mt    <- t(as.matrix(rev(data.frame(as.matrix(mon))))) #  change order of y
m     <- melt(mt)

col   <- c("azure",  "orange")

ggplot(m, aes(x = Var2, y = Var1, fill = value)) +
  geom_tile(colour="grey10") +
  scale_fill_manual(values = col, labels = f, name = NULL) +
  theme(panel.background = element_rect(fill = "white"), axis.ticks = element_blank()) +
  theme(axis.title.x = element_blank(), axis.title.y = element_blank()) 

As one can see, I attribute the second element of 2 factors, "Work" to the elements, but it plots "Free". What is disturbing, is that the factors of mon have only 1 level in place of 2 possible levels. It gives another plot if I attribute several levels to the mon:

mon   <- as.data.frame(matrix(as.factor(rep(c(f[1], f[2]), times = 4)), nrow = 4))

.. and re-running the plot obove. It is also not possible to assign another level, even if it was a choice from originally 2 levels:

mon[1,1] <- f[1]

I tried a lot with levels, relevel, order etc. without success. Does anyone have an idea?

Answer 1

Matrices can't hold factors. When you put a factor in a matrix, it gets coerced to character, and the unused levels are lost. as.data.frame(matrix(...))) is a bad habit for this (and other class conversion) reasons.

Here's a way to replicate your data transformations as near as I can follow them without losing factor levels:

f <- factor(c("Free", "Work"))
x= rep(f[2], 4)
mon <- data.frame(A = x, B = x)
str(mon)
# 'data.frame': 4 obs. of  2 variables:
#  $ A: Factor w/ 2 levels "Free","Work": 2 2 2 2
#  $ B: Factor w/ 2 levels "Free","Work": 2 2 2 2
## looks good

# What is y? What's the point?
#mt    <- t(as.matrix(rev(data.frame(as.matrix(mon))))) #  change order of y

mon$id = 1:nrow(mon)
m     <- reshape2::melt(mon, id.vars = "id", factorsAsStrings = FALSE)

levels(m$value)
# [1] "Free" "Work"
## looks good

Now, when we get to plotting, specify drop = FALSE in the scale to include unused levels in the legend. (Use the default drop = TRUE if you don't want the unused levels showing up.) Since the levels are already there, we don't need to customize the labels.

col   <- c("azure",  "orange")

ggplot(m, aes(x = id, y = variable, fill = value)) +
  geom_tile(colour="grey10") +
  scale_fill_manual(values = col, name = NULL, drop = FALSE) +
  theme(panel.background = element_rect(fill = "white"), axis.ticks = element_blank()) +
  theme(axis.title.x = element_blank(), axis.title.y = element_blank()) 

If you want to be extra safe with the color scale, you can add names to the values vector before putting it in the scale:

names(col) = levels(f)

---

Another way to get the data would be to not worry about the levels during transformation, and re-factor with appropriate levels at the end:

# your original code:
f             <- factor(c("Free", "Work"))
mon           <- as.data.frame(matrix(as.factor(rep(f[2], times = 8)), nrow = 4)) 
colnames(mon) <- c("A", "B")

mt    <- t(as.matrix(rev(data.frame(as.matrix(mon))))) #  change order of y
m     <- melt(mt)

# add this at the end
m$value = factor(m$value, levels = levels(f))

# check that it looks good:
str(m$value)
# Factor w/ 2 levels "Free","Work": 2 2 2 2 2 2 2 2