MATH 335: Beta distribution

In R, we say

help(dbeta)

to get help about the pdf for the Beta distribution.

Notice that the help message gives help on several functions related to the Beta. This bonus information is how R creates help messages for any of its built in probability distributions. Click Here to see a list of available distributions.


beta.plot <- function(r,s) {
x <- seq(0,1,length=501)
y <- dbeta(x,r,s)
plot(x,y,type='l',main=paste("r =", r, ",", "s =", s))
}

Enter that function into R. Then run the following code to see a collection of beta pdf's.


m <- 4
n <- 4
par(mfrow=c(m,n))

r.vect <- c(.75, 2, 5, 100)
s.vect <- r.vect
for (i in 1:m) {
   for (j in 1:n) {
     r <- r.vect[i]
     s <- s.vect[j]
     beta.plot(r,s)
     }
     }
par(mfrow=c(1,1))

Thumbtacks : Empirically, I dropped a tack 50 times and got 35 "point-ups". So n=50 and k=35. The code below plots the appropriate posterior beta distribution using different beta priors.
k <- 35
n <- 50

r <- 1; s <- r  # uniform prior
par(mfrow=c(2,1))
beta.plot(r,s)
beta.plot(k+r,n-k+s)
qbeta(.025,k+r,n-k+s)  # lower limit of posterior interval
qbeta(.975,k+r,n-k+s)  # upper limit of posterior interval
(k+r)/(n+r+s) # posterior mean, aka "squared-error loss Bayes estimate

r <- 10; s <- r  # symmetric about .5; sd=.11
par(mfrow=c(2,1))
beta.plot(r,s)
beta.plot(k+r,n-k+s)
qbeta(.025,k+r,n-k+s)  # lower limit of posterior interval
qbeta(.975,k+r,n-k+s)  # upper limit of posterior interval
(k+r)/(n+r+s) # posterior mean, aka "squared-error loss Bayes estimate

r <- 100; s <- r  # symmetric about .5, sd=.035
par(mfrow=c(2,1))
beta.plot(r,s)
beta.plot(k+r,n-k+s)
qbeta(.025,k+r,n-k+s)  # lower limit of posterior interval
qbeta(.975,k+r,n-k+s)  # upper limit of posterior interval
(k+r)/(n+r+s) # posterior mean, aka "squared-error loss Bayes estimate

r <- 1000; s <- r  # symmetric about .5, sd=.011
par(mfrow=c(2,1))
beta.plot(r,s)
beta.plot(k+r,n-k+s)
qbeta(.025,k+r,n-k+s)  # lower limit of posterior interval
qbeta(.975,k+r,n-k+s)  # upper limit of posterior interval
(k+r)/(n+r+s) # posterior mean, aka "squared-error loss Bayes estimate


Here is a summary table of the results:


Beta Prior   Estimate    Lower Limit    Upper Limit

r=s=1           .692        .562           .809
r=s=10          .643        .528           .750
r=s=100         .540        .478           .601
r=s=1000        .505        .483           .527

Wald int	.700        .573           .827
Plus-4          .685        .561           .809