The University of Edinburgh

College of Science and Engineering

School of Mathematics

Mathematics 3 Honours
MATH 10095 Statistical Methodology

Friday l4’ December 2018

2.3Opm-4.3Opm

Chairman of Examiners — Dr A Olde Daalhuis

External Examiner — Profcssor J McColl

Attempt ALL questions

Calculators and other electronic aids.

A scientific calculator is permitted in this examination.

ft must not be a graphical calculator.
It must not be able to communicate with any other device.

This examination will be marked anonymously.

MATH10095 Statistical Methodology 1

(1) The temperature (Y in ‘C) of the core of a novel coitiputing device for mining hitcoins
is being studied before the unit is connnercialised; it is assuiiicd that (tic teniperat tire
follows a Normal distribution

1 1 1 (q
IQu; ii)
—

(c known)
= — —

with a2 asst.tnied to he known.

(a) Let ‘ .i be a random .saitiple from this Noritial distrilnitioii. Show that
the log—likelihood for p is

= u— log2 —
—

where c is a constant. independent of t and u2. [2 marks]

(b) Derive expressions for the score and obtain the MLE.
[2 marks]
(c) Derive the test statistics for the score and Wahi tests for resting H0 = ito

against H] :/i # /4o What would he t lie (list rihut ion under H0 for large samples?
[4 marks]
(d) Suppose a2 61.2 and that n = 530 measurements have been itiade resulting iii
=
992DQ
Obtain a 9.5Y confidence interval for the mean teiiiperature using
the Wald statistic.
HINT: Recall that the 95 quantile of a chi—square raiidoiii variable with one
degree of freedom is 3.81. [4 marks]

[Please turn over]

MATH10095 Statistical Methodology 2

(2) A Marketing entrepreneur has been assuming that the number of it.enis in a shopper’s
basket at a supermarket checkout hue is given by a Poisson distribution

fz(z;A)= / (z=0,1,2,...).

After consulting a Statistician. they realized it would be more realistic to consider

a model that takes into account that nobody stands on the line without an item
to buy. and thus it would he not possible to observe zeros. The Statistician thus
reconnnended a zero—truncated model based on

= ii I Z >0.

(a) Show that the probability function of the observed (zero—truncated Poisson)
distribution is given by

A” e
fy(y; A) (y = 1.2....).
= —
e)

HINT: Note that -

= P(Z =qandZ >0)

fyQq;A) = P(Z = y [ Z >0)

[3 marks]
(1) Show that the expected number of items in a shopper’s basket is E(Y) = A/( 1 —
e A)
[5 marks]
(c) Let Y ), he independent observations from the zero—truncated Poisson
model. Obtain the Fisher information 1(A), and show that, for a fixed A, Fisher
information increases wit Ii the sample size (n). [7 marks]
(d) Write the recursive equation of the Fisher scoring method to (letenhlille the
maximum likelihood esthna.tor. Suppose Y ,...,), fy respectively denote
the number of items ill the basket of each of the ri = 2123 shopper’s, with
= 3663. Use A0 = Y to initialise the method, and obtain the first
iteration of the Fisher scoring method.
[5 marks]

[Please turn overj

MATH 1009.5 Statistical Methodology 3

(3) Suppose we want to study how the teiiipcrat nrc (1 in tQ) of a niachitie increases.
as a function of how long (X, > 0 in minutes) the macmile has been operating. It is
believed that the following model, known as regression I lirougli the origin, protides
a sensible approximation to the relation of interest,

= ]X1 +so I = 1,...,n, (1)

lid
with r .c (0. aj.
-

(a) Using the regression equal ion iii (1) just dv why the tat hi of sample averages
I is a natural estimator for 3. for situations where a is close to () zero.
[lINT: Note I hat when a 0 it follows that :3X1 [5 marks]
(h) Show that ,3 = ?/X is unbiased and obtain its variance.
[3 marks)
(c) Suppose i = 62 and .T = 180. Estimate what the average temperature should
be for an operating time of .z: = 90 minutes.
[5 marks]
(d) Obtain the mnaxinnun likelihood estimator for ,3 and show that it coincides with
lie least squares est iniat or. [5 marks)

[End of Paperl