stat 613 exam 1debdeep/exam1-613-s18-soln.pdfstat 613 exam 1 february 21, 2018 name: uid: please...
TRANSCRIPT
STAT 613 Exam 1February 21, 2018
Name:
UID:
Please sign the following pledge and read all instructions carefully before starting the
exam.
Pledge: I have neither given nor received any unauthorized aid in completing this exam, and Ihave conducted myself within the guidelines of the University Honor Code.
Signature:
INSTRUCTIONS:
• This is an closed-book, closed-notes exam. However, a formula page is provided at the back.
• Total time is 75 minutes (3:00 P.M to 4:15 P.M.)
• Show all work, clearly and in order, if you want to receive full credit. When you use yourcalculator, explain all relevant mathematics. I reserve the right to take o↵ points if I cannotsee how you arrived at your answer (even if your final answer is correct).
• Circle or otherwise indicate your final answers.
• Answer all the questions in the space provided. You may attach additional sheets
if necessary.
• This test has 4 problems and is worth 80 points. It is your responsibility to make sure thatyou have all of the problems.
• Good luck!
Prob. No. Max Points Earned Pts.
1 20
2 20
3 20
4 20
TOTAL:
Question 1. (20 pts.) Let X1, X2, . . . , Xn be i.i.d discrete uniform distribution on {1, 2, . . . , N}where N is an unknown parameter taking values in the set of all positive integers, i.e. {1, 2, . . .}.Recall that P (X1 = j | N) = 1/N for all j = 1, . . . , N .
a) (8 points) Find a minimal su�cient statistic for N . No need to actually show that it is minimalsu�cient.
b) (12 points) Is the statistic obtained in a) complete? Explain your answer.
a) to ( a ,. - . ixnln ) = ntn 1[×a, en ]
.
Hence Xcn ,is an Mss
b)
Tosho=×cn)isC.
for N 71 , Re 1,2 ,.
. . in
PN ( Xa , =k) = Pn ( Xcn ,E K ) - PN ( Xcn )
Eh ' )
= an . #"
1 ifN =L
,k=1
Henn. Pnlxurr ) = { ⇐,n . rent it III.In
Start with g : { 1,2 ,.
. - . } → R such that
En { g(×a , ) } =° tn=i , 3 . --
SetN= ⇒ g ( i ) = 0
Sane ⇒ gG÷ +gc 2) ( I - In )=o ⇒ g (2) =o
Setty ⇒ gg , p(×a , =D + 9k )P( xcn )- 2) +9C 3)
PCXLN)=3 )
⇒ g (3) = 0 since PCXAD = 3) 40 . =o
Continuing like this gck ) = O t K =L , 4 . --
⇒ PN ( g Cxcn , ) =o ) = it N 31
⇒ xcn ) is CSS
Question 2. (20 pts.) Suppose we take one observation, X, from the discrete distribution, withprobability mass function given in the following Table where ✓ 2 [0, 1].
Table 1: Probability mass function of X
x �2 �1 0 1 2P (X = x | ✓) (1� ✓)/4 ✓/12 1/2 (3� ✓)/12 ✓/4
a) (4 points) Find a real-valued statistic T (X) with E✓[T (X)] = ✓ (T is an unbiased estimator of✓, but might take values outside [0, 1]).
b) (8 points) Obtain the maximum likelihood estimator (MLE) ✓(X) of ✓ and show that it is notnecessarily unique.
c) (8 points) Is any choice of MLE unbiased?
a) Let TGD = 12 if ×= -1
= 0 otherwise
to [ Tcx ) ] = 12 P(×= - , ) = 12 . 01,2=0
b) MLE depends onwhich X you
observe
of C- 2) =0
,€ ( t ) =i
,£ ( o ) is any
number in
[oD
EG ) =o,
I (a) =L
C)@ For MLE to be unbiased ,we must have
to [ Ecx ) ] = a t off ,i ]
⇒ o÷ + €62 + Eat °
@ ( → =a÷⇒ 8¥)
=
2¥⇒
→ contradiction Sinn
estimator Can't defend on
parameter .
Question 3. (20 points) Let X1, X2, . . . , Xn be iid N(µ,�2) with unknown mean µ 2 R and knownvariance �2 > 0. Recall that X = (1/n)
Pni=1 Xi is a complete and su�cient statistic for µ.
a) (8 points) For fixed and known t 6= 0, find the uniformly minimum variance unbiased estimatorof etµ.
b) (12 points) Show that the variance of the uniformly minimum variance unbiased estimatorobtained in a) does not achieve the Cramer-Rao lower bound, but it is asymptotically e�cient.
a) In Mm , oyn ) e[ etx ] = etnt t÷I
⇒ E [ et'T - t÷T ] =etr
⇒ et× ' t 'Tis fmuiowof c Ss & unbiased for
etm,
So UMVUE for etm .
b)CRLB = Seenetm }
2
teenage 7¥,
= need
Fnwgf =" In Var ( etx - t÷ )then
= e. ¥ { ecentx ) -ETETD }
= e- t¥ { ezntt' ' IF
.
ezrtt tiny= eat { et¥ . , }
£252
µm var ( et 'T - II )=
et -1=
z- it -
n → A CRLBn , A t~÷
So UMWE is asymptotically efficient .
Question 4. (20 points) Suppose X1, . . . , Xn are independent, with Xi ⇠ N(✓i, 1), i = 1, . . . , n.In vector notation, we may write X ⇠ Nn(✓, In), where X = (X1, . . . , Xn)T is the observable,✓ = (✓1, . . . , ✓n)T is the unknown mean vector and In is the n⇥ n identity matrix.
a) (12 points) Find the MLE of = k✓k2, the squared norm of ✓ (prove that it is the MLE !).
b) (2 points) Show that the MLE of is biased.
c) (3 points) What happens to the bias as n ! 1? Explain the phenomenon.
d) (3 points) Find the Cramer-Rao lower bound for an unbiased estimator of .
a) l ( AID = . nzhog th - ÷ ( x - a)'
( x - o )
% KMD = +( x . a) =o ⇒ d '
- X
- E[off ,log A Cola ] = In .n .
d- . xisthe MLE . By invariant ,
⇒ zx ,
.2 is he MLE for
4
b) E ( Exp ) = [ oi2 in
= 4 + n
Bias = Ettzxiy - Y =n
c) |Bias| → a as n→& .
This is because #
of parameters dependon n .
- )
d) 04 = 20 .
CRLB =JYTICO ) 04
= 4110112
Table of Distributions
Distribution PMF/PDF and Support Expected Value Variance MGF
Bernoulli
Bern(p)
P (X = 1) = p
P (X = 0) = q = 1� p p pq q + pet
Binomial
Bin(n, p)
P (X = k) =�nk
�pkqn�k
k 2 {0, 1, 2, . . . n} np npq (q + pet)n
Geometric
Geom(p)
P (X = k) = qkp
k 2 {0, 1, 2, . . . } q/p q/p2 p1�qet
, qet < 1
Negative Binomial
NBin(r, p)
P (X = n) =�r+n�1
r�1
�prqn
n 2 {0, 1, 2, . . . } rq/p rq/p2 (p
1�qet)r, qet < 1
Hypergeometric
HGeom(w, b, n)
P (X = k) =⇣wk
⌘⇣b
n�k
⌘/⇣w+bn
⌘
k 2 {0, 1, 2, . . . , n} µ =nwb+w
⇣w+b�nw+b�1
⌘nµ
n (1� µn ) messy
Poisson
Pois(�)
P (X = k) = e���k
k!
k 2 {0, 1, 2, . . . } � � e�(et�1)
Uniform
Unif(a, b)
f(x) = 1b�a
x 2 (a, b) a+b2
(b�a)2
12etb�eta
t(b�a)
Normal
N (µ,�2)
f(x) = 1�p2⇡
e�(x � µ)2/(2�2)
x 2 (�1,1) µ �2 etµ+�2t2
2
Exponential
Expo(�)
f(x) = �e��x
x 2 (0,1)1�
1�2
���t , t < �
Gamma
Gamma(a,�)
f(x) = 1�(a) (�x)
ae��x 1x
x 2 (0,1)a�
a�2
⇣�
��t
⌘a, t < �
Beta
Beta(a, b)
f(x) = �(a+b)�(a)�(b)x
a�1(1� x)b�1
x 2 (0, 1) µ =a
a+bµ(1�µ)(a+b+1) messy
Log-Normal
LN (µ,�2)
1x�
p2⇡
e�(log x�µ)2/(2�2)
x 2 (0,1) ✓ = eµ+�2/2 ✓2(e�2 � 1) doesn’t exist
Chi-Square
�2n
12n/2�(n/2)
xn/2�1e�x/2
x 2 (0,1) n 2n (1� 2t)�n/2, t < 1/2
Student-ttn
�((n+1)/2)pn⇡�(n/2)
(1 + x2/n)�(n+1)/2
x 2 (�1,1) 0 if n > 1n
n�2 if n > 2 doesn’t exist