Most Powerful Test.

π Υ ( θ 0 )=P [ reject H 0 ∣ H 0 is true ]= alpha

P [ reject H 0 ∣ H 0 is false ]= Beta

Neyman Pearson Lemma Theoem

Let X 1 , … , X n be a random sample from some density f (x ; θ) where θ is one of the two known
values of θ0 or θ1 and let 0< α <1 be fixed.

Let k ¿ be a positive constant and C¿ be a subset of X which satisfy:

¿
(i) Pθ [( X 1 , … , X n )∈ C ]=α
0

L(θ0 ; x 1 , … , x n ) L0 ¿ ¿
(ii) λ= = ≤ k if ( x 1 , … , x n ) ∈ C
L(θ1 ; x 1 , … , x n ) L1
L1 ¿ ¿
(iii) and λ= ≥ k if ( x 1 , … , x n ) ∈ C
L0

Then the test Υ ¿corresponding to the critical region C ¿ is the most powerful test of size
¿
α of H 0 :θ=θ 0 versus H 1 :θ=θ1 . C is the complement of C ¿ ; that is, C ¿= X−C ¿
PROOF
 ¿ ¿
Suppose that k and C satisfying conditions (i) and (ii) exist. If there is no other test of size α or
less then it is the most powerful test . Let Υ be another test of size α or less and let C be its
corresponding critical region. We have Pθ [( X 1 , … , X n )∈ C]≤ α . We need to prove that
0

π Υ∗¿(θ )≥π 1 Υ (θ 1 )¿ .

For any subset of the critical regions R from the sample space X , we abbreviate as

[∏ ( ]
n
∫ ⋯R ∫ f x i ;θ j ) d x i
i=1

H 0 :f =f 0 , H 1 : f =f 1 , where f 0 and f 1 are continuous densities. Then among all tests of size less than
or equal to α , the test with smallest probability of the a type II error is given by

{
C= x :
f 1(x)
f 0(x) }
> k where K is chosen such that

α =P ( reject H 0 ∣ H 0 )

¿ P( X ∈C ∣ H 0 )=∫ f 0 (x )d x .
C

The given C specifies a likelihood ratio test with size α .

Let β=P ( X ∉ C ∣ f 1 )=∫ f 1 ( x ) d x .
C

¿
Let C be the critical region of any other test with size less than or equal to α .
Let α ¿ =P ( X ∈C ¿ ∣f 0 ) , β ¿ =P ( X ∉C ¿ ∣f 1) .

¿
We want to show β ≤ β .
We know α ¿ ≤ α , ie ∫ f 0 (x )d x ≤∫ f 0 (x )d x .
¿
C C

Also, on C we have f 1 ( x ) > k f 0 ( x ) , while on C we have f 1 ( x ) ≤ k f 0 ( x )

∫
¿
f 1( x ) d x ≥ k ∫
¿
f0(x)d x
C ∩C C ∩C

and

∫ ¿
f 1( x ) d x ≤ k ∫
' ¿
f 0 (x ) d x .
C ∩C C ∩C
 β−β ¿ =∫ f 1 ( x ) d x −∫ f 1 ( x ) d x
¿
C C

β−β ¿ = ∫ ¿
f 1 ( x ) d x− ∫
¿
f 1 ( x) d x
C ∩C C ∩C

β−β ¿ = ∫ ¿
f 1( x ) d x + ∫ ¿
f 1 ( x ) d x− ∫
¿
f 1 ( x ) d x− ∫ f 1( x ) d x
C∩C C ∩C C ∩C Ć ∩ Ć
¿

≤k ∫ ¿
f 0 ( x ) d x−k ∫ ¿
f 0( x ) d x
C ∩C C ∩C

¿
β− β =¿ k ∫ ¿
f 0 ( x )d x+ ∫ ¿
f 0( x)d x }
C ∩C C∩ C

β−β ¿ =¿−k
{∫
C¿ ∩ C
f 0 ( x ) d x+ ∫
CC ∩C ¿
f 0( x ) d x
}
¿ ¿
β−β ¿ k ( α −α )
Example 3.

Let x 1 , x 2 , … x nconstitute a sample from the density with pdf as f ( x ; θ)=θ e−θx

Find the most Powerful test of size α .

Solution.

Null hypothesis H 0 :θ=θ 0

Alternative hypothesis H 1 :θ=θ 1

Likelihood function for the exponential distribution:

n −θ0 ∑ x
Under Null Hypothesis L0=θ 0 e
n −θ 1 ∑ x
Under Alternative Hypothesis L1=θ1 e

Then for the Critical Region,

n −θ 0 ∑ x
L0 θ 0 e
= ≤k
L1 θ1n e−θ ∑ x 1

( )
n
L0 θ0 e−θ Σ x 0

¿ ⋅ ≤k
L1 θ1 e−θ Σ x 1

¿
L1 θ 1
e
( )
L0 θ0 n −θ Σ x+θ Σ x
≤k 0 1
 ( )
n
L0 θ0 ∑ x (θ −θ )
= e ≤k 1 0

L1 θ1

Taking Logarithm on both sides of the inequality

[( ) ]
n
L0 θ
=log 0 e ∑ x (θ −θ ) ≤ log k
1 0

L1 θ1

∑ x ( θ1−θ0 ) ≤ log k −nlog

( ) θ0
θ1

Divide both sides of the inequality by ( θ1 −θ0 )

∑ x≥
log k −nlog
( ) θ0
θ1
( θ 1−θ0 )
Divide both sides of the inequality by

x≥
log k −nlog ( ) θ0
θ1
=k
˙

n ( θ 1−θ0 )
˙
x≥k