ggplot: filling color based on condition - r

Question

I want to plot two categorical variables (group, condition) and one numeric variable (value). In addition, I want to base the filling color on the significance of the values (significant bars should be grey, the rest white). With the following code, however, only some significant bars are colored in grey.
plot <- ggplot(dat, aes(group, value))+
geom_col(aes(fill = condition), position = position_dodge(0.8), width = .7, color= "black") +
scale_fill_manual(values = ifelse(dat$significance > .05, "white", "grey")) +
geom_linerange(aes(group = condition, ymin = ci_lower, ymax= ci_upper), position = position_dodge(0.8)) +
coord_flip(ylim =c(-.2,1))
plot
here is my data:
dat <- structure(list(group = c("friends", "parent", "esm", "friends", "parent", "esm"),
value = c(0.25, 0.44, 0.33, 0.47, 0.25, 0.32),
significance = c(0.08, 0, 0, 0, 0.01, 0),
condition = c("S1", "S1", "S1", "S2", "S2", "S2"),
trait = c("E", "E", "E", "E", "E", "E"),
ci_lower = c(0.52, 0.74, 0.53, 0.67, 0.44, 0.49),
ci_upper = c(-0.03, 0.14, 0.14, 0.27, 0.06, 0.15)),
row.names = c(1L,2L, 3L, 16L, 17L, 18L), class = "data.frame")

Answer 1

You can add an inline mutate to create a column to specify the color group based on significance. The key here is to use the group aesthetic so the bars can still be dodged and positioned correctly based on the condition variable.
dat %>%
mutate(sig = significance < .05) %>%
ggplot(aes(group, value, group = condition)) +
geom_col(
aes(fill = sig),
position = position_dodge(0.8),
color = "black",
width = .7
) +
scale_fill_manual(values = c("white", "grey")) +
geom_linerange(aes(ymin = ci_lower, ymax = ci_upper),
position = position_dodge(0.8)) +
coord_flip(ylim = c(-.2, 1))
Gives this plot:
However, I think you need another aesthetic to distinguish condition in addition to significance. Color is one option, but this is a nice place to use ggpattern which will be more obvious than the outline color and keep the B&W look.
Here's an example:
library(ggpattern)
dat %>%
mutate(sig = significance > .05) %>%
ggplot(aes(group, value, group = condition)) +
geom_col_pattern(
aes(fill = sig, pattern_angle = condition),
position = position_dodge(0.8),
pattern_fill = "black",
pattern_spacing = 0.025,
pattern = "stripe",
width = .7,
color = "black"
) +
scale_pattern_angle_discrete(range = c(45, 135)) +
scale_fill_manual(values = c("grey", "white")) +
geom_linerange(aes(ymin = ci_lower, ymax = ci_upper),
position = position_dodge(0.8)) +
coord_flip(ylim = c(-.2, 1))
Which gives this plot:
Finally, it's worth noting that the color of a bar is not usually used to denote significance of a statistical metric; a much more common convention would be to use asterisk to indicate relevant p value thresholds (e.g. ** p < 0.01) or letters to indicate membership in a grouped analysis such as an ANOVA. These can be easily implemented using the ggpubr package. That would leave fill color free to indicate the grouping by condition.

Answer 2

It can also be useful:
library(ggplot2)
#Code
ggplot(dat, aes(group, value))+
geom_col(aes(fill = interaction(condition,significance > .05)),
position = position_dodge(0.8), width = .7, color= "black") +
scale_fill_manual(values = c("grey","grey","white"),
breaks = c('S2.FALSE','S1.TRUE'),
labels=c('S2','S1')) +
geom_linerange(aes(group = condition, ymin = ci_lower, ymax= ci_upper), position = position_dodge(0.8)) +
coord_flip(ylim =c(-.2,1))+
labs(fill='Var')
Output: