STAT0009 Introductory Notes
STAT0009 Introductory Notes
STAT0009 Introductory Notes
Contents
1 The law of large numbers 1
1.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2 Summary 4
1.1 Theory
You may recall that in probability theory, we are given the distribution of a non-negative integer-valued
random variable X and we define its mean to be
X
EX = xP(X = x).
x∈N
The justification of the use of the word mean or average is given by the following theorem. Recall that a
countable collection of random variables are independent if every finite subset of them is independent, and
they are identical if they have the same distribution. We say that a sequence of random variables (Xn )n∈Z+
converges almost-surely to a random variable X if there exists an event Ω0 with P(Ω0 ) = 1 such that for
every ω ∈ Ω0 , we have convergence in the usual pointwise sense: Xn (ω) → X(ω), as n → ∞.
Theorem 1.1 (Law of large numbers (almost-sure version). Let (Xn )n∈Z+ be a sequence of independent and
identically distributed (i.i.d.) random variables. If E|X1 | < ∞, then
almost-surely.
The almost-sure version of the law of large numbers has a somewhat difficult proof, but to prove convergence
in the mean-square and in probability is easy. We say that Xn converges to X in the mean-squared if
E|Xn − X|2 → 0
P(|Xn − X| > ) → 0.
1
Theorem 1.2 (Law of large numbers (convergence in mean-squared)). Let (Xn )n∈Z+ be a sequence of
i.i.d. random variables. If E|X1 |2 < ∞, then
in the mean-squared.
a2 P(X > a) ≤ EX 2 .
Exercise 1.1. Let g : [0, 1] → R be a continuous function. Let (Xi )i∈Z+ be independent random variables
that are uniformly distributed on the interval [0, 1]. Show that
n
1X
Z 1
g(Xi ) → g(x)dx.
n i=1 0
Recall that if Tn is a sequence of estimators for a parameter θ, then we say that Tn is consistent if Tn → θ
in probability.
Exercise 1.2. Show that the usual sample variance is an consistent estimator for the true variance. Hint:
use the fact that:
n
X
(n − 1)S 2 = Xi2 − n(X̄)2 .
i=1
## Application
The law of large numbers gives us another way to compute probabilities. If we want to compute say
p = P(Z > 1), one way to is to consider i.i.d. random variables with the same law as Z, and consider the
count
Sn = 1[X1 > 1] + · · · + 1[Xn > 1],
where 1[Xi > 1] = 1 if Xi > 1 and zero otherwise. Since the Xi are i.i.d. the Bernoulli random variables
1[Xi > 1] are also i.i.d. and thus the law of large numbers applies
n−1 Sn → p,
2
and thus the average for large values of n approximates p.
In R it is possible to simulate random variables, and thus it is possible to approximates probabilities. In
short, we can do most of the exercises of elementary probability by simulation.
Exercise 1.3 (The hat check problem). A professor completely forgets the names and faces of all n = 20
students in his class. He randomly hands back their midterms. Let pn be the probability that no one
receives their own test back. It is known that as n → ∞ we have pn → e−1 , and the approximation is quite
good for even n = 20. Demonstrate this fact with R.
Solution. Note that with R we can easily generate a random permutation of the 20 papers, as a permuation
of 20 numbers. Each random permuation generated by R is considered to be independent. We can check
whether a permuation is in-order by defining a function Hand; we repeat this experiment (10000 times) with
the replicate function and keep score.
## [1] 0.3751
1/exp(1)
## [1] 0.3678794
Exercise 1.4 (Boxes and balls). A box initially contains 8 blue balls and 4 red balls. Terry closes his
eyes and randomly picks a ball and then follows the following procedure: if it is a blue ball, he puts the
blue ball back in, and also adds another blue ball; if it is a red ball, then he removes the red ball from
the bin and in addition, removes another red ball. Suppose you did not see what Terry did and close your
eyes and pick a ball at random and it is red; then what is the probability that Terry picked a blue ball? Use R.
Solution. First, we code one iteration of this procedure, which reports the first ball, and the second ball. We
will use Bernoulli random variables, whereby zero will mean a blue ball, and one will mean a red ball, has
been drawn. Next, we play the game a large number of (independent) times.
We also, store as two separate vectors, the outcome of the first pick, and the second pick. We want to count
the number of times the second pick resulted in a red ball, and out of those times, the first pick was a blue
ball, and then take the ratio. We write a simple loop to accomplish this.
game <-function() {
x = rbinom(1,1, 4/12);
y = rbinom(1,1, 2/10);
if (x==0)
y = rbinom(1,1, 4/13);
c(x,y)
}
z = replicate(10000, game());
F = z[1,]
3
S = z[2,]
br=0
for (i in 1:10000)
if (S[i]==1 && F[i]==0)
br <- br+1
br
## [1] 2018
r=sum(S==1)
r
## [1] 2703
br/r
## [1] 0.7465779
Exercise 1.5 (Uniform random variables). Let (Ui )i∈Z+ be a sequence of independent random variables
that are uniformly distributed in [0, 1]. Let
Sn = X1 + · · · + Xn .
Let
T = inf{n ≥ 1 : Sn > 1}
so that T is the first time the sum is greater than 1. Use R to compute ET . You will not be disappointed.
Exercise 1.6 (Histograms). Carefully explain why plotting a (probability) histogram of some sample data
may give an approximation of the probability density function of some random variable.
Exercise 1.7 (Integration). Let f (x) = x2 . Compute
Z 2
f (x)dx
0
2 Summary
We reviewed the law of large numbers and gave some proofs of some basic results. We showed how the law of
large numbers can be used in conjugation with R to compute or estimate various probabilities arising from
elementary probability theory.