Lab4_sol

Embed Size (px)

Citation preview

  • 8/17/2019 Lab4_sol

    1/2

    ST 522-001: Statistical Theory II

    Solution to Lab Exercises - 4

    Prepared by Chen-Yen Lin

    Feb. 09, 2011

    6.24; (i) When   λ  = 0, random variable   X   degenerates to 0. It follows immediately thatthis family is not complete. (ii) When   λ   = 1, to show this family is not complete,we can let  g(0) =  −e  and  g(X ) = 1,   ∀x >   0. Then it can be shown that  Eg(X ) =

    g(0)e−1 + e−1∞

    x=1h(x)x!

      = −1 + 1 = 0.

    6.26: (a) If P 0 consists two distributions N (θ1, 1) and N (θ0, 1). By   Theorem 6.6.5 , a minimalsufficient statistic for  P   is

    T (x) = pθ1 pθ0

    =  e−

    nθ212

    e−nθ202

    exp

      ni=1

    xi(θ1 − θ0)

    which is equivalent to  X̄ (b) Similarly, consider a case in which P 0 consists only two distributions gamma(α, β 1)and gamma(α, β 0). By   Theorem 6.6.25 , a minimal sufficient statistic for  P 0   is

    T (x) = β α0β α1

    exp

    ni=1

    xi

     1

    β 0−

      1

    β 1

    which is equivalent to n

    i=1 X i.(c) Consider a case in which  P 0  consists only two distributions U(0,θ1) and U(0,θ0).By   Theorem 6.6.5 , the minimal sufficient statistic for  P 0   is

    T (X ) =θ−n1   I (X (n),∞)(θ1)

    θ−n0   I (X (n),∞)(θ0)

    which is X (n)

    6.31(a) (i) When  σ2 is known, it can be shown that  X̄   is complete sufficient statistic for  µ,whereas the distribution of (n −  1)S 2/σ2 does not depend on   µ  and therefore   σ2 isancillary for  µ. By   Basu Theorem ,  X̄  ⊥ S 2.(ii) When σ2 is unknown but fixed, first observe that  S 2 = (n− 1)−1

    ni=1(X i−

     X̄ )2 =(n −  1)−1

    ni=1(Z i  −

     Z̄ )2, where   Z i   =   X i  −  µ   is   N (0, σ2). Therefore,   S 2 is still an

    ancillary statistic. By Basu Theorem,  X̄  ⊥ S 2 for unknown µ  and fixed σ2. Since σ2 isarbitrary, this result can be extended to general unknown  µ  and σ2. (More discussioncan be found in Example 6.2.27.)

    1

  • 8/17/2019 Lab4_sol

    2/2

    6.31(c) (i) Given that X/Y   and Y  are independent, we have

    EX k =   E Y  X Y 

    k

    EX k =   EY kE 

    kEX k

    EY k  =   E 

    k

    (ii) Since   α   is known, it can be argued that   T   = n

    i=1 X i   is complete sufficient forβ . Moreover, since   X i  comes from a scale family with standard variable   Z i   whose

    distribution does not depend on  β . It follows that  X (i)

    n

    i=1 X iis ancillary for β . Therefore

    E (X (i)|T ) = E 

    X (i)T 

      T |T 

    = E 

    X (i)T   |T 

    E (T |T ) = E 

    X (i)T 

    T   =  E (X (i))

    ET   T 

    The last equality holds by using the result from (i).

    2