How to add the prior distribution to the mcmc_area chart

General
1

Hi. This is more like a R and ggplot question, and because I am not good at making graphs.

I finished running a Stan model using cmdstan, and fit is the output object.

I want to plot the mcmc_areas graph for a parameter along with its prior distribution, which is normal(0,10) in this example

>posterior1 <- fit$draws(format="df")
>mcmc_areas(posterior1, pars = c("b2"), prob = 0.9) +
  stat_function(fun = dnorm, args = list(mean = 0, sd = 10))

image
image1303×868 22.5 KB

This graph is not pretty. Is there a way to make it so?

2

The bayesplot outputs are just ggplot objects so you should be able to manipulate them the same way you would pretty much any other plot generated by that package.

3

Thanks for your reply. I am aware of the possibility and I even supplied my attempt in code. But the result is clearly not what one wants to see. I would appreciate it if you can provide any suggestion. Cheers.

4

In case it might be helpful, here’s an answer I wrote a while back that extracts prior and posterior draws and plots them on the same graph.

2 Likes
5

Thank you to your idea. I learned from your code and this is what I achieved for just one parameter.

posterior0 <- fit0$draws(format="df")

model_draws <-
  posterior0 %>%
  select(.chain, .iteration, .draw, starts_with("mu"))

S=500 #number of draws needed

plotting <- data.frame(mu=c(sample(model_draws$mu,S),
                rnorm(S,mean=0,sd=1)),
           group=c(rep('posterior',S),rep('prior',S)))

ggplot(data=plotting, aes(x=mu,colour=group)) +
  geom_density()

While this works, I would still prefer to see a way to add the prior distribution plot with mcmc_areas() or any of those functions in bayesplot package.