using "overall" groups after facet_wrap (gepolt)-CodePudding

I want to add a bar (with significance) to my ggplot. I know the code for adding the bar (stat_pvalue_manual()) but my problem is of more general nature. Is it possible to use "overall" groups after a facet_wrap?

My code is the following:

library(data.table)
library(ggplot2)
library(dplyr)
data(mtcars)
data <-mtcars[,c(1,2)]
data <- data[c(1:30),]
setDT(data, keep.rownames = TRUE)[]
data <- as.data.frame(data[,c(1,2)])
data$group <- c(1,1,1,1,1,1,1,1,1,1,2,2,2,2,2,2,2,2,2,2,3,3,3,3,3,3,3,3,3,3)
data$group2 <- c(1,1,1,1,1,1,1,1,1,1,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2)
data$mpg2 <- data$mpg * 0.6345
data$mp3 <- data$mpg * 0.7564
data$group <- as.factor(data$group)
data$group2 <- as.factor(data$group2)


long_data <- melt(setDT(data), id.vars = c("rn", "group", "group2"), variable.name = "Timepoint")

p <- ggplot(long_data, aes(x = Timepoint, y = value, fill = group2))  
  geom_bar(stat = "summary", fun = "mean", position = "dodge", na.rm = T, alpha=0.35)  
  geom_point(data = long_data %>% filter(group=="1") , aes(x = Timepoint, y = value, colour =  group),  size = 0.9, shape = 16) 
  geom_point(data = long_data %>% filter(group=="2") ,aes(x = Timepoint, y = value, colour =  group),  size = 0.9, shape = 16) 
  geom_point(data = long_data %>% filter(group=="3") ,aes(x = Timepoint, y = value, colour =  group),  size = 0.9, shape = 17) 
  geom_line(data = long_data, aes(x= Timepoint, y = value, group = rn),  alpha = .3, size = 0.4)  
  facet_wrap(~group2)



p

I want to "use" both groups and add a bar between group 1 and 2 (which are separated by the facet_wrap). I attach a picture of what the plot gives me and manually draw in red what I would want.

CodePudding user response：

One way round this problem is to overlay two plots on top of each other using patchwork. One plot will have to have only a geom_signif layer, the other is your existing plot:

library(ggpubr)
library(patchwork)

p1 <- ggplot(long_data, aes(x = Timepoint, y = value, colour = group))  
  geom_bar(aes(fill = group2), color = NA, stat = "summary", fun = "mean", 
           position = "dodge", na.rm = TRUE, alpha = 0.35)  
  geom_point(data = filter(long_data, group == "1"), size = 0.9, shape = 16)  
  geom_point(data = filter(long_data, group == "2"), size = 0.9, shape = 16)  
  geom_point(data = filter(long_data, group == "3"), size = 0.9, shape = 17)  
  geom_line(data = long_data, aes(group = rn), color = 'black',
            alpha = 0.3, size = 0.4)  
  facet_wrap(~group2)  
  ylim(c(0, 40))

p2 <- ggplot(long_data, aes(group2, value))  
  geom_bar(stat = 'summary', fun = 'mean', fill = NA)  
  geom_signif(comparisons = list(c('1', '2')), test = 't.test',
              map_signif_level = function(p) sprintf("p = %.2g", p),
              extend_line = 0.025, y_position = 37, vjust = -0.2)  
  ylim(c(0, 40))  
  theme_void()

Now we draw p1 and put p2 directly on top of it:

layout <- c(
  area(t = 1, l = 1, b = 2, r = 2),
  area(t = 1, l = 1, b = 2, r = 2)
)

p1   p2   plot_layout(design = layout)

CodePudding user response：

As an alternative to the significance bar, perhaps it would be easier and just as clear to add bars for the overall group means in the two facets, such as geom_crossbar(aes(group = group2), color = "black"). Then the overlap of the error bars can signal the size of the differences across panels.