33

Suppose I have data with both an ordinal variable and a categorical variable:

set.seed(35)
df <- data.frame(Class = factor(rep(c(1,2),times = 80), labels = c("Math","Science")),
                 StudyTime = factor(sort(sample(1:4, 16, prob = c(0.25,0.3,0.3,0.15), replace = TRUE)),labels = c("<5","5-10","10-20",">20")),
                 Nerd = factor(sapply(rep(c(0.1,0.3,0.5,0.8),c(30,50,50,30)), function(x)sample(c("Nerd","NotNerd"),size = 1, prob = c(x,1-x))),levels = c("NotNerd","Nerd")))

One could use ggplot and geom_bar with x, fill and alpha (or color) aesthetic mappings to visualize the relationship between these variables. 

ggplot(data = df, aes(x = Class, fill = StudyTime, alpha = Nerd)) + 
  geom_bar(position = "dodge", color = "black") + 
  scale_alpha_manual(values = c(Nerd = 0.5, NotNerd = 1)) +
  scale_fill_manual(values = colorRampPalette(c("#0066CC","#FFFFFF","#FF8C00"))(4)) +
  labs(x = "Class", y = "Number of Students", alpha = "Nerd?") +
  theme(legend.key.height = unit(1, "cm"))

enter image description here

However, alpha and color are not ideal. A better alternative might be to apply a pattern such as stripes or a crosshatch.

The accepted answer to this question from over 10 years ago says to use colors, and the most upvoted answer (while clever) uses over 100 lines of code.

This question received some upvotes but no new answers.

Is there any better alternative to adding a pattern such as can be seen here?

enter image description here

Ian Campbell
  • 23,484
  • 14
  • 36
  • 57

1 Answers1

53

One approach is to use the ggpattern package written by Mike FC (no affiliation):

library(ggplot2)
#remotes::install_github("coolbutuseless/ggpattern")
library(ggpattern)
ggplot(data = df, aes(x = Class, fill = StudyTime, pattern = Nerd)) +
  geom_bar_pattern(position = position_dodge(preserve = "single"),
                   color = "black", 
                   pattern_fill = "black",
                   pattern_angle = 45,
                   pattern_density = 0.1,
                   pattern_spacing = 0.025,
                   pattern_key_scale_factor = 0.6) + 
  scale_fill_manual(values = colorRampPalette(c("#0066CC","#FFFFFF","#FF8C00"))(4)) +
  scale_pattern_manual(values = c(Nerd = "stripe", NotNerd = "none")) +
  labs(x = "Class", y = "Number of Students", pattern = "Nerd?") + 
  guides(pattern = guide_legend(override.aes = list(fill = "white")),
         fill = guide_legend(override.aes = list(pattern = "none")))

enter image description here

The package appears to support a number of common geometries. Here is an example of using geom_tile to combine a continuous variable with a categorical variable:

set.seed(40)
df2 <- data.frame(Row = rep(1:9,times=9), Column = rep(1:9,each=9),
                   Evaporation = runif(81,50,100),
                   TreeCover = sample(c("Yes", "No"), 81, prob = c(0.3,0.7), replace = TRUE))

ggplot(data=df2, aes(x=as.factor(Row), y=as.factor(Column),
                     pattern = TreeCover, fill= Evaporation)) +
  geom_tile_pattern(pattern_color = NA,
                    pattern_fill = "black",
                    pattern_angle = 45,
                    pattern_density = 0.5,
                    pattern_spacing = 0.025,
                    pattern_key_scale_factor = 1) +
  scale_pattern_manual(values = c(Yes = "circle", No = "none")) +
  scale_fill_gradient(low="#0066CC", high="#FF8C00") +
  coord_equal() + 
  labs(x = "Row",y = "Column") + 
  guides(pattern = guide_legend(override.aes = list(fill = "white")))

enter image description here

Ian Campbell
  • 23,484
  • 14
  • 36
  • 57
  • 2
    Interesting package, thanks for sharing. I personally find striped bar charts not really helpful, but the demonstration on tiles is a quite convincing use case for patterns, IMO :) +1 – tjebo Jun 27 '20 at 14:52
  • 3
    Hidden gem here: remove the pattern from the fill guide using `guides(fill=guide_legend(override.aes=list(pattern="none")))` – dnlbrky Jun 16 '21 at 15:44
  • 2
    Some journals require black and white figures and patterns instead of fill gradients between black and white, which makes something like geom_bar_pattern absolutely critical if you want to continue using ggplot2 for figures. – CrunchyTopping Aug 05 '21 at 13:44