1. Let X 1 , X 2 ,, X n be a random sample from the Gamma( , ) distribution f (x | , ) (a) (b) (c) (d) 1 ( ) x x 1e , x 0 , 0 , 0 . ( 8 %) Find the method of moment estimates of and . ( 7 %) Find the MLE of , assuming is known. ( 7 %) Giving 0 , find the Cramer-Rao lower bound of estimates of . ( 8 %) Giving 0 , find the UMVUE of . 2. Suppose that X 1 , X 2 ,, X n are iid ~ B ( 2, p ) , p (0,1) . Let ( p) 2 p(1 p) . n (a) ( 5 %) Show that T X i is a sufficient statistic for p . i 1 1, if X 1 1 (b) ( 5 %) Let Y . Show that Y is an unbiased estimate of ( p ) . 0 , if X 1 1 (c) (10%) Find the UMVUE W of ( p ) . 3. Let X 1 , X 2 , , X n be a random sample from a Poisson ( ) , 0 , distribution. Consider testing H 0 : 1 vs H1 : 3 . (a) (10%) Find a UMP level test, 0 1. (b) ( 7 %) For n 3 , the test rejects H 0 , if X 1 X 2 X 3 5 . Find the power function ( ) of the test. (c) ( 8 %) For n 3 , the test rejects H 0 , if X 1 X 2 X 3 5 . Evaluate the size and the power of the test. 4. (10%) Let X 1 , X 2 , , X n be iid Poisson ( ) distribution, and let the prior distribution of be a Gamma( , ) distribution, 0 , 0 . posterior distribution of . Find the 5. Let X 1 , X 2 , , X n be a random sample from an exponential distribution with mean , 0 . n (a) ( 5 %) Show that T X i is a sufficient statistic n for . i 1 (b) ( 5 %) Show that the Poisson family has a monotone likelihood ratio, MLR. (c) ( 5 %) Find a UMP level test of H 0 : 0 1 vs H1 : 1 by the Karlin-Rubin Theorem shown below. [Definition] A family of pdfs or pmfs {g (t | ) | } has a monotone likelihood ratio, g (t | 2 ) MLR, if for every 2 1 , is a monotone function of t . g (t | 1 ) [Karlin-Rubin Theorem] Suppose that T is a sufficient statistic for and the pdfs or pmfs {g (t | ) | } has a non-decreasing monotone likelihood ratio. Consider testing H 0 : 0 vs H1 : 0 . A UMP level test rejects H 0 if and only if T t 0 , where P 0 (T t 0 ) . 1 數理統計期末考試試題答案 x ( 1) 1 x e dx and ( ) ( ) 0 1. (a) Since E ( X ) 1 x ( 2) 2 1 x e dx ( 1) 2 , ( ) ( ) 0 E( X 2 ) 1 Let m1 and m2 ( 1) 2 ~ m2 m12 1 m1 m2 m12 , ~ . m1 m2 m12 m12 ~ Furthermore, m1 X , m2 m12 The MME of .and are ~ 1 n 2 1 n (n 1) 2 X i X 2 ( X i2 X ) 2 S , n i 1 n i 1 n ~ (n 1) S 2 , nX (n 1) S 2 nX 2 n n x) [ (b) L( | , ~ 1 i 1 ( ) 1 xi xi x i e ] n 1 1 ( xi [( ) ]n i 1 i 1 )e n n x1 i 1 ln L( | , ~ x ) n ln ( ) n ln ( 1) ln xi i 1 Let n n 1 n x ˆ 1 ln L( | , ~ x) x 0 xi . i 2 i 1 n i 1 Furthermore, n 2 n n 2nx ln L( | , ~ x) ln xi 2 2 3 i 1 2 3 2 n 2nx nx 2nx n ln L( ˆ | , ~ x) 0, 2 ˆ 2 ˆ 3 x3 x2 X So, ˆ is the MLE of . 2 (c) E [ n 2 n n 2n n ln L( | , ~ x )] E ( Xi ) 2 2 3 i 1 2 3 2 2 1 CRLB = E [ 2 2 ln L( | , ~ x )] 2 n X is an unbiased estimate of , and , ˆ X 1 2 2 X is the UMVUE of . Var ( ) CRLB, ˆ 2 n n (d) Since E ( X ) 2 [Or] f ( x | , ) 1 I (0, ) ( x) x 1 x e 1 I (0, ) ( x) x 1 exp[ x( ( ) ( ) Given , { f ( x | )} is an exponential family in . 1 )] n T X i is a sufficient statistic for . i 1 X T Since ˆ is an unbiased estimate of and a function of sufficient n X statistics T , by Rao-Blackwell Theorem, ˆ is the UMVUE of . n n 2 2. (a) f ( x1 , x 2 , , x n | p) f ( xi | p ) [ I{0,1,2} ( xi ) p xi (1 p) 2 xi ] x i 1 i 1 i n n 2 xi 2 p xi p i [ I{0,1,2} ( xi )( ) (1 p) 2 ] [ I{0,1,2} ( xi )]( ) 1 (1 p) 2n x x 1 p 1 p i 1 i i 1 i n n 2 p T ( ~x ) Let g (T ( ~ x ), p) ( ) (1 p) 2n and h( x) I{0,1,2} ( xi ) . By x 1 p i 1 i n factorization theorem, T X i is a sufficient statistic for p . i 1 2 2 p [Or] f ( x | p) I{0,1,2} ( x) p x (1 p) 2 x I{0,1,2} ( x)(1 p) 2 exp[ x ln( )] 1 p x x n { f ( x | p)} is an exponential family T X i is a sufficient statistic. i 1 2 (b) E (Y ) 1 P( X 1 1) 0 P( X 1 1) p1 (1 p) 2 1 2 p(1 p) , so Y is an 1 unbiased estimate of ( p ) . n (c) If X 1 , X 2 ,, X n , n N , are iid ~ B ( 2, p ) , then T X i ~ B (2n, p ) . i 1 n E (Y | T t ) P(Y 1 & T t ) P( X 1 1 & T t ) P(T t ) P(T t ) P( X 1 1 & X i t 1) i2 P(T t ) n 2n 2 t 1 p (1 p ) 2n t 1 2 p (1 p ) t 1 i2 P (T t ) 2n t p (1 p ) 2n t t 2(2n 2)! t!(2n t )! t (2n t ) , t 0,1,2,,2n . (t 1)!(2n t 1)! (2n)! n(2n 1) T (n T ) By Rao-Blackwell Theorem, W E (Y | T ) is the UMVUE of ( ) e . 2(2n 1) P ( X 1 1) P ( X i t 1) 3. (a) By Neyman-Pearson Lemma, a UMP level test rejects H 0 if and only if 3 f ( x1 , x2 ,, xn | 3) kf ( x1 , x2 ,, xn | 1) . n n 1x i 3 xi 3 e ] k [ e 1 ] 3 xi ke2n ( xi ) ln 3 2n ln k i 1 ( xi )! i 1 ( xi )! i 1 n 2 ln k c xi ln 3 i 1 n [ n Since X i ~ Poisson (n ) , a UMP level test rejects H 0 if and only if i 1 n X i c , where c is the smallest integer satisfying i 1 ni n i! e . i c 1 n [Or] T X i is sufficient for and T ~ Poisson (n ) . i 1 By the corollary of Neyman-Pearson Lemma, a UMP level test rejects H 0 if and only if g (t | 3) kg(t | 1) . n 3t 3 1t e k e 1 3t ke2 ( x1 ) ln 3 2 ln k t! t! i 1 (b) ( ) P ( X 1 X 2 X 3 5) 1 P ( X 1 X 2 X 3 4) (3 ) 0 (3 )1 (3 ) 2 (3 ) 3 (3 ) 4 3 , 0 ]e 0! 1! 2! 3! 4! 30 31 32 33 34 3 (c) The size of this test is (1) 1 [ ]e 0.1847 0! 1! 2! 3! 4! 9 0 91 9 2 93 9 4 9 The power of this test is (3) 1 [ ]e 0.9450 0! 1! 2! 3! 4! 1[ n 4. Since T X i is sufficient for and T ~ Poisson (n ) . i 1 1 (n ) t n 1 e ; and f ( ) fT | (t | ) e t! ( ) 1 ( n ) (n ) t n 1 nt , e 1e t 1e f (t , ) 0 t! ( ) t!( ) f T (t ) nt t 1 t!( ) 1 ( n ) e d 0 nt f (t , ) t!( ) f | t ( | t ) f T (t ) nt t 1 e nt t!( ) (n 1 (t )( t ) n 1 ) t 1 e (n 1 ) t (t )( ) t (t )( ) n 1 n 1 t!( ) ). The posterior distribution of is Gamma(t , n 1 4 , 0 1 i 1 1 n 5. (a) f ( x1 , x2 , , xn | ) f ( xi | ) ( e I (0, ) ( xi )) e I (0, ) ( xi ) n Let x n i 1 T (~ x) 1 g (T ( ~ x ), ) e n x i 1 n i 1 n and h( ~ x ) I (0, ) ( xi ) . By factorization theorem, i 1 n T X i is a sufficient statistic for . i 1 x 1 1 1 [Or] f ( x | ) e I (0, ) ( x) I (0, ) ( x) exp[ x( )] n { f ( x | )} is an exponential family. T X i is a sufficient statistic. i 1 X T Since ˆ is an unbiased estimate of and a function of sufficient n X statistics T , by Rao-Blackwell Theorem, ˆ is the UMVUE of . n (b) T X i ~ Poisson ( n ) g (t | ) i 1 (n ) t n , t 0,1,2, e t! (n2 ) t n 2 t e 2 n( 2 1 ) g (t | 2 ) t ! e g (t | 1 ) (n1 ) t n1 1 e t! g ( t | 2 ) If 2 1 2 1 is an increasing function of t , 1 g (t | 1 ) Hence {g (t | ) | 0} of T has MLR. n 1 i 1 (n) n (c) T X i ~ Gamma(n, ) g (t | ) 1 g (t | 2 ) g (t | 1 ) (n) 2n 1 (n)1n If 2 1 ( 1 2 t n 1 t e , t 0 t t n 1e 2 1 1 ( )t 1 e 2 1 , t 0 t 2 n t n 1e 1 1 1 ) 2 1 0 1 2 Hence {g (t | ) | 0} of T has an MLR. g (t | 2 ) is increasing in t . g (t | 1 ) n By Karlin-Rubin Theorem, the UMP size test rejecting H 0 if T X i c , where c i 1 n satisfies that P{ X i c | 1} ; i.e., i 1 c 5 1 ( n) x n 1 x e dx .