How to create a density plot where density is over all groups?

[stefaneng]

Originally asked by @jeeyunhan

The default behavior for density plot is to plot independent densities where each group integrates to 1.
How can we create a density plot with multiple groups so that the integral of all groups is 1?

suppressPackageStartupMessages(library(BoutrosLab.plotting.general));
set.seed(13);
n1 <- 500;
n2 <- 1000;
my.data <- list(
  group1 = rnorm(n1, mean = 2),
  group2 = rnorm(n2, mean = 4)
  );

create.densityplot(
  my.data,
  col = default.colours(2),
  main = 'Independent densities'
  );

^{Created on 2023-03-09 with reprex v2.0.2}

[stefaneng]

Currently create.densityplot does not support this but we can create this plot using density function and create.scatterplot. We need to compute the density for each group and then scale by the number of data points in each groups.

suppressPackageStartupMessages(library(BoutrosLab.plotting.general));
set.seed(13);
n1 <- 500;
n2 <- 1000;

my.data <- list(
  group1 = rnorm(n1, mean = 2),
  group2 = rnorm(n2, mean = 4)
  );

grp1.density <- density(my.data$group1);
grp2.density <- density(my.data$group2);

grp1.df <- as.data.frame(grp1.density[c('x', 'y')]);
grp1.df$group <- 1;
# Scale by multiplying by the ratio of: (# of group 1 )/(total count)
grp1.df$scaled.y <- grp1.df$y * n1 / (n1 + n2);
grp2.df <- as.data.frame(grp2.density[c('x', 'y')]);
grp2.df$group <- 2;
# Scale by multiplying by the ratio of: (# of group 2)/(total count)
grp2.df$scaled.y <- grp2.df$y * n2 / (n1 + n2);

combined.df <- rbind(grp1.df, grp2.df);

# Recreate density plot
create.scatterplot(
  y ~ x,
  data = combined.df,
  groups = combined.df$group,
  col = default.colours(2),
  type = 'l'
  );

create.scatterplot(
  scaled.y ~ x,
  data = combined.df,
  groups = combined.df$group,
  col = default.colours(2),
  type = 'l'
  );

^{Created on 2023-03-09 with reprex v2.0.2}

[pboutros]

Yup fully agree with @stefaneng on the solution, I was going to offer the same strategy. I'd also put this as a feature we won't add in the near-term to create.densityplot() because it's a bit of a bespoke case, so will tag as wontfix for now (but glad the issue/solution are visible here!)