Bilkent University Spring 2018-19
Math255 Probability and Statistics
Final
28 May 2019
Name and Lastname:
Bilkent ID No:
Math 255 Section No:
Score (for instructor use)
(P1) (P2) (P3) Sum
Instructions
• The exam is 150 minutes. There are three problems. Closed book. You may use one two-sided
A4-size sheet of notes.
• Before the exam starts:
– Make sure you are taking the exam in the classroom your are assigned to.
– All electronic devices must be switched off and stored away.
– Remove all objects from your desk, including bags, coats, books, calculators, cell phones,
wallets. The only items allowed on your desk are one cheat-sheet, writing utensils, and
one bottle of water. No snacks no food!
– Once the exam starts, you will not be allowed to leave the room and come back for any
reason. But you may leave any time if you finish the exam early.
• During the exam:
– Place your ID visibly on your desk and keep it that way throughout the exam.
– Keep your exam sheets in front of you at all times. Exposing your answer sheets to
others shall be treated as cheating.
– Write legibly. If your answer cannot be read you will receive no credit.
– In each problem, show your work in the space provided for that problem and write your
final answer in the designated box.
– Double check all your results and pay attention to the following! You will receive no
credit if
∗ there is not sufficient justification for a correct answer,
∗ the answer does not appear in the designated box,
∗ a numerical answer (real number or a fraction) is not given where required,
∗ the answer is wrong for one reason or another, including trivial mistakes.
DO NOT USE THIS PAGE FOR SOLUTIONS.
Problem 1. [12 pts] Consider estimating the unknown parameter θ of a Poisson distribution based
on a sample x from the distribution. Use a Bayesian approach where θ has a priorfΘ (θ) = e−θ for
θ ≥ 0, and the observation process is defined by fX|Θ (x|θ) = θx e−θ /x! for x = 0, 1, 2, . . ..
(a) (6 pts) Compute the MAP estimate Θ̂MAP (x).
R∞
(b) (6 pts) Compute the LMS estimate Θ̂LMS (x). (You may find it useful that 0 un λe−λu du =
n!/λn .)
Show your calculations in full detail.
Θ̂MAP (x) = Θ̂LMS (x) =
Problem 2. [9 pts] Consider estimating the unknown parameter θ of a Poisson distribution
based on a sequence of independent samples x1 , x2 , . . . , xn from the distribution. Q
Use a classical
approach and model the observation as a sample form pX1 ,...,Xn (x1 , . . . , xn ; θ) = ni=1 pXi (xi ; θ)
where fXi (xi ; θ) = θxi e−θ /xi ! for xi = 0, 1, 2, . . ..
(a) (6 pts) Compute the ML estimate Θ̂ML (x1 , . . . , xn ). Show your calculation in full detail.
(b) (2 pts) Compute the bias of the ML estimator.
(c) (1 pt) Is the ML estimator consistent? Justify your answer by citing results from law of large
numbers if necessary.
Θ̂ML (x) = Bias =
Consistent (YES/NO)? Justify your answer here.
Problem 3. [9 pts] Suppose that X1 , X2 are i.i.d. Ber(p) random variables, where p is unknown;
i.e., pX1 ,X2 (x1 , x2 ) = pX1 (x1 )pX2 (x2 ) where pXi (1) = p and pXi (0) = 1 − p. Consider a binary
hypothesis testing problem with H0 : p = 1/2 and H1 : p 6= 1/2. A decision is to be made after
observing (X1 , X2 ).
(a) (6 pts) Consider tests for which the rejection region (of the null hypothesis H0 ) is of the form
R = {x1 + x2 ≥ γ}, where the possible values of γ are restricted to γ ∈ {0, 1, 2, 3}. (Thus, we
are considering four possible tests in all.) Compute the probabilities of type-1 error α and
type-2 error β as functions of 0 ≤ p ≤ 1 and γ ∈ {0, 1, 2, 3}.
(b) (3 pts) Is it possible to construct a test (in any manner whatsoever) such that α = 1/2 and
β = 1 − p? If so, describe how and show how to calculate the error probabilities for the test
you propose. If not, explain why no such test can exist.
α(p, γ) = β(p, γ) =
Answer for (b):
Solution of Problem 1 only.
Solution of Problem 2 only.
Solution of Problem 3 only.