Statistical Inference

Test Set 3

1. Let X 1 , X 2 ,..., X n be a random sample from a population with density function

x x2 
f ( x=
) exp −  , x > 0, θ > 0. Find FRC lower bound for the variance of an
θ  2θ 
unbiased estimator of θ . Hence derive a UMVUE for θ .

2. Let X 1 , X 2 ,..., X n be a random sample from a discrete population with mass function
1−θ 1 θ
P( X = −1) = , P( X = 0) = , P( X =1) = , 0 < θ < 1.
2 2 2
Find FRC lower bound for the variance of an unbiased estimator of θ . Show that the
1
variance of the unbiased estimator X + is more than or equal to this bound.
2
3. Let X 1 , X 2 ,..., X n be a random sample from a population with density function
f ( x) =θ (1 + x) − (1+θ ) , x > 0, θ > 0. Find FRC lower bound for the variance of an
unbiased estimator of 1/ θ . Hence derive a UMVUE for 1/ θ .

4. Let X 1 , X 2 ,..., X n be a random sample from a Pareto population with density

βα β
f X ( x=
) , x > α , α > 0, β > 2. Find a sufficient statistics when (i) α is known, (ii)
x β +1
when β is known and (iii) when both α , β are unknown.

5. Let X 1 , X 2 ,..., X n be a random sample from a Gamma ( p, λ ) population. Find a sufficient

statistics when (i) p is known, (ii) when λ is known and (iii) when both p, λ are
unknown.

6. Let X 1 , X 2 ,..., X n be a random sample from a Beta (λ , µ ) population. Find a sufficient

statistics when (i) µ is known, (ii) when λ is known and (iii) when both λ , µ are
unknown.

7. Let X 1 , X 2 ,..., X n be a random sample from a continuous population with density

function
θ
=
f ( x) , x > 0, θ > 0. Find a minimal sufficient statistic.
(1 + x)1+θ

8. Let X 1 , X 2 ,..., X n be a random sample from a double exponential population with the
1 −| x −θ |
density f= ( x) e , x ∈ , θ ∈ . Find a minimal sufficient statistic.
2
9. Let X 1 , X 2 ,..., X n be a random sample from a discrete uniform population with pmf
1
p=
( x) = , x 1,  , θ , where θ is a positive integer. Find a minimal sufficient statistic.
θ
10. Let X 1 , X 2 ,..., X n be a random sample from a geometric population with pmf
(1 − p ) x −1 p , x =
f ( x) = 1, 2,  , 0 < p < 1 . Find a minimal sufficient statistic.

11. Let X have a N (0, σ 2 ) distribution. Show that X is not complete, but X 2 is complete.

12. Let X 1 , X 2 ,..., X n be a random sample from an exponential population with the density
n
x) λ e − λ x , x > 0, λ > 0. Show that Y = ∑ X i is complete.
f (=
i =1

13. Let X 1 , X 2 ,..., X n be a random sample from an exponential population with the density
) e µ − x , x > µ , µ ∈ . Show that Y = X (1) is complete.
f ( x=

14. Let X 1 , X 2 ,..., X n be a random sample from a N (θ , θ 2 ) population. Show that ( X , S 2 )

is minimal sufficient but not complete.
 Hints and Solutions
2
x
1. log f ( x | θ ) = log x − log θ − .
2θ
2
 ∂ lo f g  X2 1 E ( X 4 ) E ( X 2 ) 1 8θ 2 2θ 1
2
1
E  = E  2 −  = − + 2 = − 3+ 2 = 2.
 ∂θ   2θ θ  4θ θ θ 4θ θ θ θ
4 3 4

θ 2
So FRC lower bound for the variance of an unbiased estimator of θ is .
n
1 θ2
Further T =
2n
∑ X i2 is unbiased for θ and Var (T ) =
n
. Hence T is UMVUE for θ .

1 1 1 + 4θ − 4θ 2
2. E ( X )= θ − . So T= X + is unbiased for θ . Var (T ) = .
2 2 4n
(θ − 1) −1 , x = −1
∂  ∂ log f 
2
 1
=log f ( x | θ ) = 0, x 0 So we get E   =
∂θ θ −1 ,  ∂θ  2θ (1 − θ )
 x = 1
2θ (1 − θ )
The FRC lower bound for the variance of an unbiased estimator of θ is .
n
2θ (1 − θ ) (1 − 2θ ) 2
It can be seen that Var (T ) − = ≥ 0.
n 4n
∂ 1
3. We have log f ( x | θ ) =− log(1 + x) . Further, E{log(1 + X )} = θ −1 , and
∂θ θ
 ∂ log f 
2
1
E{log(1 + X )} = 2
2θ . So E 
−2
 = 2 , and FRC lower bound for the variance of
 ∂θ  θ
θ2 1
an unbiased estimator of θ −1 is
n
.= Now T
n
∑ log(1 + X i ) is unbiased for θ −1 and
θ2
Var (T ) = .
n
4. Using Factorization Theorem, we get sufficient statistics in each case as below
n
 n

(i) ∏ X i ,(ii) X (1) (iii)  X (1) , ∏ X i 
i =1  i =1 

5. Using Factorization Theorem, we get sufficient statistics in each case as below

n n
 n n

(i) ∑ X i ,(ii) ∏ X i (iii)  ∑ X i , ∏ X i 
i =1 i =1  i =1 i =1 

6. Using Factorization Theorem, we get sufficient statistics in each case as below

n n
 n n

(i) ∏ X i ,(ii) ∏ (1 − X i ) (iii)  ∏ X i , ∏ (1 − X i ) 
i =1 i =1 =  i 1 =i 1 
n
7. Using Lehmann-Scheffe Theorem, a minimal sufficient statistic is ∏ (1 + X ) .
i =1
i

8. Using Lehmann-Scheffe Theorem, a minimal sufficient statistic is the order statistics

( X (1) ,  , X ( n ) ) .
9. Using Lehmann-Scheffe Theorem, a minimal sufficient statistic is the largest order
statistics X ( n ) .
n
10. Using Lehmann-Scheffe Theorem, a minimal sufficient statistic is ∑X
i =1
i .

11. Since E ( X ) = 0 for all σ > 0 , but but P( X= 0)= 1 for all σ > 0 . Hence X is not
1
complete. Let W = X 2 . The pdf= e − w/2σ , w > 0 .
2
of W is fW ( w)
σ 2π w
∞
Eσ g (W ) 0 for all σ > 0 ⇒ ∫ g ( w) w−1/2 e − w/2σ dw =
Now= 0 for all σ > 0 . Uniqueness
2

of the Laplace transform implies g ( w) = 0 a.e. Hence X 2 is complete.

n
12. Note that Y = ∑ X i has a Gamma (n, λ ) distribution. Now proceeding as in Problem 11,
i =1
it can be proved that Y is complete.

( y ) n e n ( µ − x ) , x > µ , µ ∈ .
13. Note that the density of Y = X (1) is given by fY=
Eg (Y ) 0 for all µ ∈ 
=
∞
⇒ ∫ n g ( y )e n ( µ − y ) dy =
0 for all µ ∈ 
µ
∞
⇒ ∫ g ( y )e − ny dy =0 for all µ ∈ 
µ

Using Lebesgue integration theory we conclude that g ( y ) = 0 a.e. Hence Y is complete.

14. Minimal sufficiency can be proved using Lehmann-Scheffe theorem. To see that
 n 
( X , S 2 ) is not complete, note that E S 2  0 for all θ > 0.
X 2 −=
 n +1 
 n 2
However, P  =
X 2 S= 0.
 n +1 