Need to put asterisk on the top of ggplot barplot to flag the level of significance (pvalue)?

Question

I have a lm model results containing R2 and pvalue, and I plotted them in a bar plot. I have then facetted them using two discrete variables. I want to put * on the top of bars to flag statistical significance (pvlue <= 0.05), as shown on the bottom-left-most panel of the below image.

I have not found an insightful tutorial on how to do this.

Any way to do this, please?

Here is some code I used

> head(res_all_s2)
         WI aggre_per  Season yield_level   slope Intercept r.squared
1    R IDW2       Dec Season2   Region II   -7.06      6091      0.41
2    R IDW2       Dec Season2    Region I   -7.29      6280      0.40
3    GDD AS       OND Season2   Region II   14.23    -18270      0.34
4    GDD AS       Nov Season2   Region II   36.84    -14760      0.33
5 SPI1 IDW2       Dec Season2   Region II -405.10      5358      0.31
6 SPI1 IDW2       Dec Season2    Region I -421.70      5523      0.32
  adj.r.squared fstatistic.value pval pearson
1          0.36             9.58 0.01   -0.64
2          0.36             9.49 0.01   -0.64
3          0.29             7.09 0.02    0.58
4          0.28             6.97 0.02    0.58
5          0.26             6.40 0.02   -0.56
6          0.27             6.51 0.02   -0.56

> # significance (pval <= 0.05)
> signif_reg <- res_all_s2 %>% filter(pval <= 0.05)
> head(signif_reg)
         WI aggre_per  Season yield_level   slope Intercept r.squared
1    R IDW2       Dec Season2   Region II   -7.06      6091      0.41
2    R IDW2       Dec Season2    Region I   -7.29      6280      0.40
3    GDD AS       OND Season2   Region II   14.23    -18270      0.34
4    GDD AS       Nov Season2   Region II   36.84    -14760      0.33
5 SPI1 IDW2       Dec Season2   Region II -405.10      5358      0.31
6 SPI1 IDW2       Dec Season2    Region I -421.70      5523      0.32
  adj.r.squared fstatistic.value pval pearson
1          0.36             9.58 0.01   -0.64
2          0.36             9.49 0.01   -0.64
3          0.29             7.09 0.02    0.58
4          0.28             6.97 0.02    0.58
5          0.26             6.40 0.02   -0.56
6          0.27             6.51 0.02   -0.56
> 
> # Plot R2
> 
> r <- res_all_s2 %>%  ggplot(aes(x=aggre_per,
+                                 y=r.squared )) +
+   geom_bar(stat="identity", width=0.8) +
+   facet_grid(yield_level ~ WI,
+              scales = "free_y",
+              switch =  "y") +
+   scale_y_continuous(limits = c(0, 1))  +
+   xlab("Aggregation period") +
+   ylab(expression(paste("R-squared"))) +
+   theme_bw() +
+   theme(axis.title = element_text(size = 12),  # all titles
+         axis.text = element_text(colour = "black"),
+         axis.text.x = element_text(angle = 90, vjust = 0.5,
+                                    hjust = 1, color = "black"),
+         strip.text.y.left = element_text(angle = 0),
+         panel.border = element_rect(color = "black",
+                                     size = .5))
> r

And, here is the link to my res_all_s2 dataset https://1drv.ms/u/s!Ajl_vaNPXhANgckJeqDKA0fzfFEbhg?e=VfoFaB

Answer 1

Technically, you can always add an appropriate geom with its independent dataset (that would be your data filtered to exclude pval > .05):

df_filtered <- res_all_s2 %>% filter(...)

## ggplot(...) +
      geom_point(data = df_filtered, pch = 8)
      ## pch = point character, no. 8 = asterisk

or

## ... +
      geom_text(data = df_filtered, aes(label = '*'), nudge_y = .05)
      ## nudge_y = vertical offset

or color only significant columns:

## ... +
   geom_col(aes(fill = c('grey','red')[1 + pval <= .05]))

So, yes, technically that's feasible. But before throwing the results of 13 x 7 x 5 = 455 linear models at your audience, please consider the issues of p-hacking, the benefits of multivariate analysis and the viewers' ressources ;-)