以几种方式在 R 和 ggplot2 中显示相对频率或密度直方图

Question

i would appreciate your help on the following please (i should add that i have sent an email also to ggplot2 mailing list, and have not heard from anyone yet).

we do have a dataframe with a FACTOR called EXP that has 3 LEVELS (DMSO, DMSO1, DMSO2)

 head(pp_ALL)
    VALUE  EXP
1 1639742 DMSO
2 1636822 DMSO
3 1634202 DMSO

shall i aim to overlay the relative frequency histograms, or the density histograms for the FACTOR LEVELS,

please would you let me know why the following 2 pieces of R code show very different results (in terms of the height of the density histograms, and interpretation):

ggplot(pp_ALL, aes(x=VALUE, colour=EXP)) + geom_density()

versus

ggplot(data=pp_ALL) +
       geom_histogram(mapping=aes(x=VALUE, y=..density.., colour=EXP),  bins=1000) 

thanks,

bogdan

Answer 1

Let's compare two examples below. With default settings on this data, they look pretty similar in shape and density. They do have slight differences based on how they work, with the density plot using a smoothing algorithm while the histogram uses discrete bins. This can sometimes make the histogram more easily interpretable (height = what share of obs in that bin, scaled by bin width), while the density plot might be more reflective of an underlying smooth distribution. But you'll note that here they're pretty similar.

diamonds %>%
  mutate(cut = factor(cut)) %>%
  ggplot() + 
  geom_density(aes(x = carat, color = cut)) +
  facet_wrap(~cut)

diamonds %>%
  mutate(cut = factor(cut)) %>%
  ggplot() + 
  geom_histogram(aes(x = carat, y = ..density.., fill = cut)) +
  facet_wrap(~cut)

With alternative settings, they could look very different. If we made the bandwidth on the density plot 1/10th as big, or used 10x as many bins for the histogram, it would make the spikes narrower and higher, while continuing to integrate in area to 1, just as before. I presume the main thing you are seeing is that your histogram has a much higher granularity (bins = 1000) than your density plot, so any spikes will be narrower and taller than in the density plot.

geom_density(aes(x = carat, color = cut), adjust = 1/10) + ...

geom_histogram(aes(x = carat, y =..density.., fill = cut), bins = 300) + ...

One other difference in behavior you might notice in your case with multiple factors is that the density plot uses lines without stacking, while the histogram stacks -- this will tend to make the histogram taller even if it has equivalent bandwidth settings, because it is stacking the densities for each factor.

Same as orig but w/o facet_wrap.

[Try geom_density(aes(x = carat, color = cut), position = "stack") to make it more like the histogram.]

Try geom_histogram(aes(x = carat, y =..density.., fill = cut), position = "dodge") to make it more like the density plot.]