Scribd
Upload
55 views6 pages

Mathematical Statistics (MA212M) : Lecture Slides

This document discusses asymptotic confidence intervals (CIs), which provide an easy way to construct CIs when the sample size is sufficiently large. It describes using the central limit theorem and properties of maximum likelihood estimators to derive asymptotic CIs for the mean of a distribution-free population and the success probability of a Bernoulli distribution. Examples are given to illustrate how to calculate asymptotic CIs.

Uploaded by

Akshay Narasimha
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
55 views6 pages

Mathematical Statistics (MA212M) : Lecture Slides

This document discusses asymptotic confidence intervals (CIs), which provide an easy way to construct CIs when the sample size is sufficiently large. It describes using the central limit theorem and properties of maximum likelihood estimators to derive asymptotic CIs for the mean of a distribution-free population and the success probability of a Bernoulli distribution. Examples are given to illustrate how to calculate asymptotic CIs.

Uploaded by

Akshay Narasimha
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 6

Mathematical Statistics (MA212M)

Lecture Slides
Lecture 29
Asymptotic CI

In many cases it is very difficult to find pivot for a small sample.


For example it is difficult to find a pivot to construct CI for
successes probability of a Bernoulli distribution.
However, we may able to find CI quite easily if the sample size is
sufficiently large. This CI is called asymptotic confidence
interval.
We will use convergence in distribution (mainly CLT or large
sample distribution of MLE) and in probability (consistent
estimator).
Distribution Free Population Mean
Let X1 , X2 , . . . be i.i.d. random variables with mean µ and finite
variance σ 2 . Then using CLT
√ 
n Xn − µ L
−
→ Z ∼ N(0, 1).
σ
Now if n, sample √
size, is large, we can approximate the
n(X n −µ)
distribution of σ
using a standard normal distribution.
Hence
√  !
n Xn − µ
P −zα/2 ≤ ≤ zα/2 ≈ 1 − α.
σ
If σ is known and n is sufficiently large, we can use the last
statement to find an asymptotic CI for µ and it is given by
 
σ σ
X n − √ zα/2 , X n + √ zα/2 .
n n
Distribution Free Population Mean (Contd.)
If σ is unknown, we can proceed as follows:
Sn P
We know that σ
−
→ 1, and hence
√ 
n Xn − µ L
−
→ Z ∼ N(0, 1).
Sn
 √ 
n(X n −µ)
Hence P −zα/2 ≤ Sn
≤ zα/2 ≈ 1 − α.
An asymptotic CI for µ is given by
 
Sn Sn
X n − √ zα/2 , X n + √ zα/2 .
n n

Note that this method can be used for any distribution of


X1 , X2 , . . . , Xn , as long as the conditions of CLT hold true.
Hence it is called distribution free.
Using MLE

Let θbn be a consistent estimator of θ and


√ b L
n θn − θ − → N(0, b 2 (θ)), where b(θ) > 0 for all θ ∈ Θ.
Assume that b(·) is a continuous function.
√ b
b(θbn ) P n(θn −θ) L
Then b(θ) − → 1 and hence b(θb ) − → N(0, 1).
n

A 100(1 − α)% asymptotic CI for θ is given by


" #
b(θbn ) b(θbn )
θbn − √ zα/2 , θbn + √ zα/2 .
n n

Under some regularity conditions, we may use MLE of θ in place


of θbn .
Examples
i.i.d.
Example 142: X1 , X2 , . . . , Xn ∼ Bernaoulli(p), where p ∈ (0, 1).
We are interested to construct asymptotic CI for p.
√
P n(X n −p ) L
→ p and √
We know that pbn = X n − −
→ N(0, 1).
p(1−p)
p
Here b(p) = p(1 − p), which is a conditions function in p ∈ (0, 1).
√
n(X n −p ) L
Hence √ −
→ N(0, 1).
X n (1−X n )

A 100(1 − α)% asymptotic CI for p is


 s s 
X n − X n (1 − X n ) zα/2 , X n + X n (1 − X n ) zα/2  .
n n

You might also like