Pass data parameter to Comparisons in stat_compare_means

[emcnerny]

I would like to be able to use comparisons for a subset of the data passed to stat_compare_means. Right now you can do this to compare means against a reference group, but not for specified comparisons. For example, here I'm testing against the lowest dose only for the VC group:

df <- ToothGrowth
df$dose <- as.factor(df$dose)
df.VC <- df[df$supp == "VC",]
#Plot using ref.group
ggbarplot(
  df,
  "dose",
  "len",
  color = "supp",
  add = c("point", "mean_sd"),
  xlab = "dose",
  ylab = "len",
  position = position_dodge(0.72)
  ) + stat_compare_means(
    data = df.VC,
    method = "t.test",
    ref.group = "0.5",
    label = "p.format"
  )

Which produces:

But this does not work for comparisons (the data parameter is ignored, so it pools the supp groups):

#Plot using comparisons - should have same values, but ignores subsetted data (df.VC)
mycomparisons = list(c("0.5", "1"), c("0.5", "2"))
ggbarplot(
  df,
  "dose",
  "len",
  color = "supp",
  add = c("point", "mean_sd"),
  xlab = "dose",
  ylab = "len",
  position = position_dodge(0.72)
) + stat_compare_means(
  data = df.VC,
  method = "t.test",
  comparisons = mycomparisons
)

Which produces:

A separate but related issue, any attempt to pass a label type with comparisons doesn't have the expected result:

#Plot using comparisons - should have same values, but ignores data
mycomparisons = list(c("0.5", "1"), c("0.5", "2"))
ggbarplot(
  df,
  "dose",
  "len",
  color = "supp",
  add = c("point", "mean_sd"),
  xlab = "dose",
  ylab = "len",
  position = position_dodge(0.72)
) + stat_compare_means(
  data = df.VC,
  method = "t.test",
  comparisons = mycomparisons,
  label = "p.format"
)

[kassambara]

Thank you for this well written and reproducible bug report.

Fixed, now!

Install the latest developmental version and kindly confirm that it works.

[emcnerny]

Works great, thanks!

[kassambara]

so we can close this issue!

[emcnerny]

Would it be possible to make it so additional params can be passed through to geom_signif? It would make it possible to adjust xmin/xmax and y_position so the tips fall over the correct bars when using subsetted data like in this example.

Edit - NM, I realized that geom_signif doesn't allow both comparisons and x position to be set.

[ousabd]

Thanks for this great package kassambara.

I was wondering whether emcnerny's last proposition (adjusting the x position so the tips fall over dodged bars in subsetted data) was feasible ?

[kassambara]

To add manually p-values from a data frame, use the function stat_pvalue_manual() as described at
#99

[bznes]

This is a great package but I am having the same problem. I want to overlay p-values for a subset of comparisons ONLY to the treated data to a base plot that looks like this:

ggbarplot(d=d, x = "Group",y="response",color="Treatment",add = "mean_se",position = position_dodge(0.8))

When I add stat_compare_means for the treated plot I get this:
my_comparisons=list(c("C1","C2"), c("A","C2"), c("C2","B"), c("C2","D"), c("D","E"), c("D","F")) d.treated <- d[d$Treatment == "treated",] ggbarplot(d=d, x = "Group",y="response",color="Treatment",add = "mean_se",position = position_dodge(0.8))+ stat_compare_means(d.treated,comparisons=my_comparisons)

However, this is ignoring use of the treated group for stat_compare_means. When I run as facet I see the p-values I actually want displayed on the plot.

ggbarplot(d=d, x = "Group",y="response",color="Treatment",facet.by="Treatment",add = "mean_se",position = position_dodge(0.8))+ stat_compare_means(comparisons=my_comparisons)

Is it possible to get the data from the right panel on my first image? Per the above solution I updated all my packages and it still is not working. Can I overlay two ggbarplots like geom_line and geom_point in ggplot? I haven't been able to get that to work.

Thanks!

[bznes]

Problem solved...despite a recent download, sessionInfo() revealed I was running an older version of ggpubr package, updating to ggpubr_0.2 fixed my problem. Now my question is if you can recenter the comparison brackets over the group being compared (treated, not centered between the two treatment levels).