In this paper we explore the properties of the distribution of the mean of \(n = 40\) exponentials. The rate parameter \(\lambda\) is \(0.2\), and that means that mean of our exponentional distribution is \(\mu=1/\lambda\), i.e. \(5\) and the standard deviation is also \(\sigma=1/\lambda\), i.e. \(5\).
First, let’s simulate our data. We will have \(10000\times40\) matrix filled with exponentionally randomly generated values with rate parameter 0.2.
n<-40
nosim<-10000
lambda<-0.2
sim<-matrix(rexp(nosim*n,lambda),nosim)
Now, let’s create a vector with means of all the rows. It will be our sample from the distribution of the means.
means<-apply(sim,1,mean)
Central Limit Theorem tells us that this distribution should be normal with mean \(\mu = 1/\lambda = 5\) and standard deviation \(\sigma/\sqrt n = 5/\sqrt 40 = 0.79\). Since we were asked about variance, square that: \(\sigma^2/n = 25/40 = 0.625\). Let’s see what we’ve really got.
First, let’s look on mean of our distribution:
mean(means)
## [1] 5.004405
Seems pretty close to \(5\).
Now, variance:
sd(means)^2
## [1] 0.6396042
Not so far from \(0.625\).
Let’s compare the looks of our sample and normal distribution.
library(ggplot2)
g<-qplot((means-mean(means))/sd(means),geom="density")
g<-g+stat_function(fun = dnorm, size = 2)
g
Again, quite close.
We were also asked to evaluate the coverage of the confidence interval for \(1/\lambda: \bar X \pm 1.96 S/\sqrt{n}\)
left<-mean(means)-1.96/(lambda*sqrt(n))
right<-mean(means)+1.96/(lambda*sqrt(n))
sum(means>left & means < right)/length(means)
## [1] 0.9499
And we almost hit 95% confidence interval, as we should have.