University of St Andrews

Key information

SCOTCAT credits : 15

ECTS credits : 7

Level : SCQF level 10

Semester: 2

Availability restrictions: Not automatically available to General Degree students

Planned timetable: 10.00 am Mon (odd weeks), Wed and Fri

This module aims to show how the methods of estimation and hypothesis testing met in 2000- and 3000-level Statistics modules can be justified and derived; to extend those methods to a wider variety of situations. The syllabus includes: sufficiency, comparison of point estimators; the Rao-Blackwell Theorem; minimum variance unbiased estimators; Fisher information and the Cramer-Rao lower bound; maximum likelihood estimation; theory of Generalized Linear Models; hypothesis-testing; confidence sets.

Relationship to other modules

Pre-requisite(s): Before taking this module you must pass MT3507

Anti-requisite(s): You cannot take this module if you take MT5701

Learning and teaching methods and delivery

Weekly contact: 2.5 lectures (weeks 1 - 10) and 0.5 tutorial (weeks 2 - 11).

Scheduled learning hours: 30

Guided independent study hours: 120

Assessment pattern

As used by St Andrews: 2-hour Written Examination = 100%

As defined by QAA
Written examinations : 100%
Practical examinations : 0%
Coursework: 0%

Re-assessment: Oral examination = 100%

Personnel

Module coordinator: Dr G Minas
Module teaching staff: Dr Nicolò Margaritella
Module coordinator email gm256@st-andrews.ac.uk

Intended learning outcomes

• Explain key ideas and notions of statistics such as bias, (minimal) sufficiency, efficiency, consistency, and uniformly most powerful test
• Apply important theorems of classical statistics to derive unbiased parameter estimators that attain minimum variance, and hypothesis tests that are uniformly most powerful
• Use likelihood methods for finding parameter estimators, for computing the available information about parameters, and for deriving hypothesis tests; and describe the properties of these methods
• Identify probability distributions that belong to the exponential family and derive statistics that are minimally sufficient, complete, and efficient, as well as uniformly most powerful tests, and generalised linear models for data that follow these probability distributions