ggplot2 vs sm package density plot output (and statistical analysis)

Question

Consider the following data frame example

library('ggplot2')
library('sm')

    original<-c(1:100,1)
    a<-sample(original,100)
    b<-rep(1:4,25)
    lala<-data.frame(a,b)

My aim is to produce density plots for values in lala$a, according to each group (1,2,3,4) defined in lala$b.

For doing so in ggplot2 I could do the following

plotDensityggplot<-ggplot()+
  geom_density(data = lala, aes(a, colour=factor(b)))+
  theme_classic()
print(plotDensityggplot)

producing this:

However, when I plot the same data using the 'sm' package to make a formal comparison of the densities using the following code:

sm.density.compare(lala$a,as.numeric(lala$b),model = "equal")

The density curves extend beyond zero in the X-axis, despite there is no value below zero in lala$a

What's going on? - note that this affect the densities reported in the y-axis.

Is the p-value from the permutation test of equality obtained from sm.density.compare a reliable estimate? - thank you!

Answer 1

For what it's worth, you can (more or less) reproduce the sm output in ggplot by pre-computing densities with base R's density (I'm not familiar with sm but I imagine that sm.density calls base R's density at some point as well).

library(tidyverse)
lala %>%
    group_by(b) %>%
    summarise(tmp = list(map_dfc(c("x", "y"), ~density(a)[.x]))) %>%
    unnest() %>%
    ggplot(aes(x, y, colour = as.factor(b))) +
    geom_line()

I'm not sure how geom_density (or stat_density) tune kernel density estimation parameters, but you seem to have less control over them than in base R's density.