Module Specification
An online version of this specification is available to prospective students at
http://www.lshtm.ac.uk/study/currentstudents/studentinformation/msc_module_handbook/section3_moduledescript/in
dex.html
GENERAL INFORMATION
Module name
Advanced Statistical Modelling
Module code
2450
Module Organiser
James Carpenter and Rhian Daniel
Contact email
James.Carpenter@lshtm.ac.uk, Rhian.Daniel@lshtm.ac.uk
Home Faculty
Faculty of Epidemiology & Population Health
Level
This module is at Level 7 (postgraduate Masters ‘M’ level) of the QAA
Framework for Higher Education Qualifications in England, Wales & Northern
Ireland (FHEQ).
Credit
LSHTM award 15 credits on successful completion of this module.
Accreditation
Not currently accredited by any other body.
Keywords
Statistics
AIMS, OBJECTIVES AND AUDIENCE
Overall aim
To introduce recent developments in statistical modelling, focussing in the first
part of the course on statistical models for drawing causal inferences, and in
the second part of the course on models for continuous and discrete dependent
data and spline models. In each part, the focus will be on (i) understanding the
nature of the scientific enquiry / practical problem that led to the
developments and for which they are appropriate; (ii) understanding the
concepts on which they are based and their relationship to each other and to
the approaches/techniques encountered earlier in the MSc Medical Statistics
course; (iii) developing the skills to use the methods creatively and
independently in practical problems and (iv) developing the capacity to review
critically their own and others' statistical modelling work in these settings.
Intended learning
outcomes
By the end of this module, students should, after Part I:



Appreciate that the mathematical and statistical language used thus far on
the MSc is insufficient for a formal study of causal effects;
Be familiar with, and be confident in using, the additional ingredients to our
statistical language necessary for studying causal effects, in particular the
use of potential outcomes;
Understand how verbal descriptions of assumptions such as “no
unmeasured confounding” can be mathematically expressed using this
extended language;
1





Understand how the “context” of a causal problem, traditionally only
verbally expressed, can also be mathematically encoded, eg using a causal
directed acyclic graph (DAG), and appreciate the benefits of doing so;
Have gained experience in drawing and interpreting causal DAGs for
particular practical problems;
Appreciate that traditional regression methods can be used for drawing
causal inferences, the assumptions under which such an approach is valid,
and the possible pitfalls in terms of unjustified extrapolation under the
parametric regression model;
Understand and be able to apply alternatives to traditional regression
methods that are based on modelling the exposure given confounders
(propensity score methods, inverse weighting and double robust methods)
instead of modelling the outcome given exposure and confounders, and
have an awareness of the strengths and limitations of this family of
approaches;
Understand the key concepts in causal mediation analysis, and be able to
apply some of the possible statistical modelling approaches in this context;
And, after part II:
 Recognise and understand the nature of the additional structure of
problems for which statistical techniques met earlier in the MSc course are
insufficient;
 Understand the basic concepts of models that can address these problems;
 Develop functional knowledge of modelling techniques that are appropriate
for such problems, specifically:
 Issues in the analysis of dependent non-normal data;
 Implications of non-linear link functions for parameter
interpretation;
 Use of subject-specific and marginal models;
 Use of Generalised Estimating Equations and empirical (sandwich)
estimates of variance;
 Use of numerical integration for inference about subject specific
models;
 Use of splines as fixed and random effects.
 Display this knowledge by using appropriate software to analyse data using
these techniques;
 Understand the way these techniques relate to each other in specific
contexts and be able to generalize from these contexts to new situations.
Target audience
This module is intended for students with an understanding of linear
regression models for hierarchical data and generalized linear models to the
level provided by the modules Generalized Linear Models (2462) and Analysis
of Hierarchical & Other Dependent Data (2465).
CONTENT
Session content
The module is expected to include sessions addressing the following topics
(though please note that these may be subject to change):
Part I:

Causal languages in statistics

Graphical models for statistics and causality
2

Causal inference using parametric statistical models: traditional regression

Causal inference using semiparametric statistical models: propensity
scores, inverse probability weighting and doubly robust estimation

Causal mediation analysis
Part II:

Revision of linear and generalized linear models, multivariate linear
models.

Modelling dependent non-normal data: subject specific and marginal
models.

Generalized estimating equations and empirical (sandwich) variance
estimates.

Likelihood based analyses for subject-specific models.

Use of splines.
TEACHING, LEARNING AND ASSESSMENT
Study resources
provided or required
Copies of all the overheads used in handout format, practicals and their
solutions; datasets required for the practicals and assessment exercise.
Teaching and learning
methods
A combination of lectures and computer-based practicals.
Assessment details
Students will complete two assignments, one on each part of the course.
The first assignment will involve the application, interpretation and
comparison of different statistical approaches to addressing particular causal
questions using a dataset from an observational study.
The second assignment will involve the use, interpretation and comparison, of
several alternative statistical analyses for a given longitudinal data set from a
medical setting.
For students who are required to re-sit, or granted a deferral or new attempt,
the task will be the same overall structure and level as the original assessment,
but involving a different data set.
Assessment dates
Assessments will be due on Friday 20 May 2016. For students who are required
to re-sit, or who are granted a deferral or new attempt, the next assessment
deadline will be the standard School-recommended date in mid/late September
2016.
Language of study and
assessment
English (please see ‘English language requirements’ below regarding the
standard required for entry).
TIMING AND MODE OF STUDY
Duration
The module runs for 5 weeks at 2 days per week; this module runs between
Thursday morning and Friday afternoon.
Dates
For 2015-16, the module will start on Thursday 21 April 2016 and finish on
Friday 20 May 2016.
Timetable slot
The module runs in LSHTM timetable slot E.
3
Mode of Study
The module is taught face-to-face in London. Both full-time and part-time
students follow the same schedule. For full-time students, other LSHTM
modules are available in the other half of the week for the C and D slots.
Learning time
The notional learning time for the module totals 150 hours, consisting of:

Contact time ≈ 42.5 hours

Self-directed learning ≈ 57.5 hours

Assessment, review and revision ≈ 50 hours
APPLICATION, ADMISSION AND FEES
Pre-requisites
This module is intended for students with an understanding of linear
regression models, generalized linear models and linear regression models for
hierarchical data to the level provided by the courses Generalized Linear
Models (2462) and Analysis of Hierarchical & Other Dependent Data
(2465). Students require an understanding of matrix algebra and calculus (to
first year undergraduate mathematics level). The module Analysis of
Hierarchical & Other Dependent Data (2465) is strongly recommended for
prior study.
English language
requirements
A strong command of the English language is necessary to benefit from
studying the module. Applicants whose first language is not English or whose
prior university studies have not been conducted wholly in English must fulfil
LSHTM’s English language requirements, with an acceptable score in an
approved test taken in the two years prior to entry. Applicants may be asked to
take a test even if the standard conditions have been met.
Student numbers
Student numbers are typically 25 per year; numbers may be capped due to
limitations in facilities or staffing.
Student selection
Preference will be given to LSHTM MSc students, particularly those registered
for specific courses or who have taken specific prior modules, and LSHTM
research degree students. Other applicants meeting the entry criteria will
usually be offered a place in the order applications are received, until any cap
on numbers is reached. Applicants may be placed on a waiting list and given
priority the next time the module is run.
Full Registration (full participation) by LSHTM research degree students is
required for this module.
Fees
For registered LSHTM MSc students, fees for the module are included within
MSc fees (given on individual course prospectus pages).
If registering specifically for this module, as a stand-alone short course,
individual module fees will apply.
Tuition fees must be paid in full before commencing the module, or by any fee
deadline set by the Registry.
Scholarships
Scholarships are not available for individual modules. Some potential sources
of funding are detailed on the LSHTM website.
Admission deadlines
For 2015-16:

For registered LSHTM MSc students, the module choice deadline (for Term
2 and 3 modules) is Friday 20 November 2015.
4

If registering specifically for this module, applications may be made at any
time but, as places are limited, applications ahead of the MSc deadline are
strongly advised. All applications should be submitted at the latest 8 weeks
prior to the start of the module. Formal registration will take place on the
morning of the first day of the module.
ABOUT THIS DOCUMENT
This module specification applies for the academic year 2015-16
Last revised 13 August 2015 by Rhian Daniel
London School of Hygiene & Tropical Medicine, Keppel St., London WC1E 7HT.
www.lshtm.ac.uk
5