1
$\begingroup$

I'm a beginner at hierarchical modeling and statistics in general and this is confusing me.

Question: If I draw a histogram of a posterior distribution from an MCMC chain, do I draw the prior as a histogram, too, or do I just paste the distribution into the posterior plot (and make sure it has the same scale)?

Background: I'm using WinBUGS/OpenBUGS to do parameter estimation of a memory decay model. This is my first try at hierarchical models and my first serious attempt at using statistics. Whenever, I want to plot the prior of one of my parameters, I usually use a function in R to get data e.g. dnorm(0:10,mean=5,sd=1) (if the prior is a normal distribution). I then use a histogram or a function like densityplot() to create a plot from the data. What I'm basically wondering is this: The output of dnorm(0:10,mean=5,sd=1) are ten values. Are these density values? Or are these probabilities? Which one should I use?

$\endgroup$
4
$\begingroup$

Your prior is a full distribution, not just 10 values. If I were using a $N(5,1)$ prior, I would probably plot it by evaluating the density function at a large number of values over small increments in the neighborhood of its mode, like this:

xv = seq(0,10,.01)
plot(xv, dnorm(xv, 5, 1), type='l')

Generally, I'd be more likely to plot the posterior using a kernel density estimate than as a histogram, but the histogram certainly has its place (especially if you aren't able to sample a lot of values from the posterior). In any event, there's very little excuse to plot a simple prior like yours using a histogram since we know its shape exactly.

Regarding this question:

The output of dnorm(0:10,mean=5,sd=1) are ten values. Are these density values? Or are these probabilities?

I refer you to this answer that does a good job explaining how to interpret a pdf: https://stats.stackexchange.com/a/4223/8451

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.