normal ump

Upload: anonymous-jshutp

Post on 08-Apr-2018

220 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/6/2019 Normal UMP

    1/2

    An UMP Test for the Mean of a Normal DistributionStat 305 Spring Semester 2006

    A random sample of size n is taken from a normal random variable X with unknownmean but with known variance 2. Formulate H0 and H1 as follows.

    H0: = 0H1: > 0

    We would like to devise a UMP test for the above set of hypotheses. To do this, let us rstdevise a test for the following set of simple hypotheses where

    1>

    0.

    H0: = 0H1: = 1

    We will apply the Neyman-Pearson Lemma to this set of simple hypotheses.Recall that the likelihood function evaluated at is

    fX(x1;:::;xn j ) = 1

    p2n

    exp

    1

    2

    nXi=1

    xi

    2!

    .

    Therefore, we will reject H0 if

    fX(x1;:::;xn j 0)fX(x1;:::;xn j 1)

    =

    1p2

    nexp

    1

    2

    nPi=1

    xi0

    2

    1p2

    nexp

    1

    2

    nPi=1

    xi1

    2 < 1k .

    Therefore, simplifying, we reject H0 if

    exp 1

    22

    nXi=1

    (xi 0)2 nXi=1

    (xi 1)2!!

    22

    ln1

    k

    .

    Expanding the sums, we have thatnXi=1

    x2i

    ! 2nxn0 + n20

    nXi=1

    x2i

    !+ 2nxn1 n21 > 22 ln

    1

    k

    1

  • 8/6/2019 Normal UMP

    2/2

    or that

    2nxn(1 0) > 22 ln

    1

    k

    n2

    0+ n2

    1.

    Therefore,

    xn >22 ln(1=k) n2

    0+ n2

    1

    2n(1

    0)

    = k.

    Thus, the critical region has the form xn > k. If we choose that () = 0,0 = (

    ; 0)

    = P(Xn > k j =

    0)

    = P

    Xn 0

    =p

    n>

    k 0=p

    n

    = P

    Z >

    k 0

    =p

    n

    .

    Therefore,

    k 0=pn

    = z0

    or, equivalently,

    k = 0

    + z0

    p

    n

    .

    Therefore, we have derived the form of the decision rule.

    :Decision Rule

    Reject H0 if Xn > 0 + z0 pn (or equivalently, z > z0)

    Note that the decision rule does not depend on 1. Therefore, any 1 > 0 would result inthe exactly the same critical region. Therefore, above decision rule constitutes an UMP testfor the mean.

    The power of the test can be easily calculated. At = 1

    > 0, the power is calculated

    by

    (; 1) = P

    Xn > 0 + z0

    p

    n

    j =

    1

    = P

    Xn 1

    =p

    n>

    0 1=p

    n+ z0

    = P

    Z >

    0

    1

    =p

    n+ z0

    = 1 0 1=pn + z0 .Therefore, () =

    0

    1

    =pn

    + z0

    . Notice that when

    1=

    0, the power is

    1

    0

    1

    =p

    n+ z0

    = 1 (z0) = 1 (1 0) = 0 = ()

    as expected of course.

    2