2019/20 Taught Postgraduate Module Catalogue
EPIB5045M Modelling Strategies for Causal Inference with Observational Data
15 creditsClass Size: 25
Module manager: Professor Mark Gilthorpe
Email: m.s.gilthorpe@leeds.ac.uk
Taught: Semester 2 (Jan to Jun) View Timetable
Year running 2019/20
Pre-requisite qualifications
Academic entry requirementsA 1st degree in a quantitative or scientific subject area with substantial mathematical, statistical or numeracy components (at least 2:1). We also consider working experience (two years or more) of research in a quantitative subject area. Non-graduates who: have successfully completed three years of a UK medical degree; are normally ranked in the top 50% of the year 3 cohort; and wish to take the Health Data Analytics MSc as an intercalated programme, will also be accepted.
English language requirements
An overall score of 7.0 on IELTS (International English Language Testing System) with at least 6.0 in writing and no other skill below 6.5; from a TOEFL paper-based test the requirement is a minimum score of 600, with 4.5 in the Test of Written English (TWE); from a TOEFL computer-based test the requirement is a minimum score of 250, with 4.5 TWE; from a TOEFL Internet-based test the requirement is a minimum score of 100, with 25 in the "Writing Skills" score.
This module is mutually exclusive with
| EPIB5037M | Advanced modelling strategies |
Module replaces
EPIB5037M Advanced Modelling StrategiesThis module is not approved as an Elective
Module summary
The module is designed to give students an introduction to the common pitfalls and challenges in the statistical modelling of observational data, with emphasis on achieving robust causal inference. By the end of the module students will be able to critically appraise statistical models as presented in the literature, identifying modelling strategies that are potentially erroneous, understanding alternative strategies (if they exist) that avoid mathematical coupling, reversal paradox and inappropriate modelling of compositional data and composite variables.Objectives
The objectives of this module are to:- Describe common pitfalls and challenges in the statistical modelling of observational data.
- Explain a range of scenarios where standard modelling strategies present challenges; outline solutions to these challenges.
- Develop an understanding of underlying assumptions to various modelling strategies, leading to student comprehension of the various pitfalls in statistical modelling.
- Describe situations where statistical modelling strategies might be erroneous or misleading in delivering robust causal inference.
- Evaluate alternative modelling strategies, based on a comprehensive appreciation of the common pitfalls, to deliver robust causal inference of observational data.
- Promote an inherently questioning stance to modelling strategies of observational data.
- Formulate a sound rationale, with respect to undertaking the modelling of observational data, appropriate for the student’s future responsibilities in employment.
- Apply professional critical evaluation of evidence based on statistical modelling of observational data.
- Provide sign-posted links to skills introduced or developed by related modules to achieve meaningful and robust causal inference of observational data.
Learning outcomes
Critical evaluation of modelling strategies
By the end of this module the student should be able to:
- Understand common pitfalls in the standard regression modelling of observational data with respect to causal inference.
Identify appropriately directly observed confounders, unobserved confounders, proxy confounders, mediators and competing exposures.
- Understand the principles of mediation analysis and G-methods for causal inference in observational research.
- Identify modelling strategies that are potentially erroneous or misleading in specific contexts regarding causal inference.
- Formulate alternative strategies, if they exist, particularly avoiding mathematical coupling, reversal paradox and inappropriate modelling of compositional data and composite variables.
- Describe how to phrase observational data research questions to ensure valid statistical modelling strategies may be applied to yield meaningful and robust causal inference.
- Contrast the limitations of viable modelling strategies for various scenarios and complex observational data structures.
- Critically appraise statistical models as used in the literature for either prediction or causal inference, and recognised whether this was appropriate or inappropriate for the stated research objective.
Modelling skills
By the end of this module the student should be able to:
- Explain the definition and implications of causality for observational data.
- Describe how to establish which confounders and competing exposures are appropriate for different research questions in a modelling framework.
- Explain the principles of mediation analysis and G-methods in their application to causal inference for observational research.
- Assimilate the pitfalls, strategies and caveats of dealing with complex data structures (e.g. hierarchical data, mathematically coupled data, ratio or composite variable data, compositional data).
- Formulate under what circumstances the adverse impacts of mathematical coupling in regression modelling can be overcome.
- Explain the challenges and limitations of ratio variables or other composite variable for frequently encountered observational research.
- Recognise that biological and statistical interactions are not the same and formulate when and how to use statistical interaction to infer underlying biological processes within a meaningful and robust causal inference framework.
Transferable skills
By the end of this module the student should be able to:
- Create appropriate statistical models of observational data for several contexts where causal inference is required.
- Undertake critical appraisal of the statistical modelling strategies of observational data by other researchers.
Skills outcomes
Fundamental knowledge and understanding of a range of statistical modelling strategies appropriate to observational data. This includes use of directed acyclic graphs, avoidance of regression to the mean, mathematical coupling, reversal paradox and the limitations of statistical interaction.
Syllabus
The module will be delivered by Professor Mark S Gilthorpe over 12 weeks, as a blend of face-to-face lectures, small group workshops, tutorials (for face-to-face feedback on assignments), use of interactive Shiny apps, online written material, online formative questions, answers and feedback, and online further reading materials.
The course will cover the following subjects:
Causality, directed acyclic graphs (DAGs) and their use within observational studies, adjustment for confounding, competing exposures and mediation analysis; regression to the mean; analysis of change, appropriate adjustment of baseline effects; the reversal paradox (Simpsons paradox, Lord’s paradox, and suppression); mathematical coupling; ratio variables and other compositional varaible; statistical interaction; G-methods as used for causal inference in observational research.
Teaching methods
| Delivery type | Number | Length hours | Student hours |
| Workshop | 7 | 2.50 | 17.50 |
| Group learning | 0 | 0.00 | 0.00 |
| Lecture | 11 | 1.00 | 11.00 |
| Tutorial | 3 | 2.00 | 6.00 |
| Independent online learning hours | 22.50 | ||
| Private study hours | 93.00 | ||
| Total Contact hours | 34.50 | ||
| Total hours (100hr per 10 credits) | 150.00 | ||
Private study
- Lectures deliver the theoretical basis of the issues covered; comprehensive, detailed lecture notes, along with multiple references, are provided ahead of each lecture online.- Students will be guided and encouraged to search for additional associated literature using material provided by Skills@Library (http://library.leeds.ac.uk/skills).
- Additional resources are uploaded to VLE, including podcasts and animated illustrations of the concepts introduced in lectures / workshops.
- Shiny apps (which are user-interactive) facilitate learning within the workshops or in own time, by allowing students to explore the consequences of various modelling assumptions.
Opportunities for Formative Feedback
This will be done in a number of ways:- Individual online formative assessments, delivered using the VLE, will help students assess their learning throughout.
- Students will be set formative group problems to work through, the outcome of which will be presented in the workshops and peer-assessed under tutor guidance.
Methods of assessment
Coursework
| Assessment type | Notes | % of formal assessment |
| Essay or Dissertation | Short reports – max 800 words | 33.30 |
| Essay or Dissertation | Short reports – max 800 words | 33.40 |
| Total percentage (Assessment Coursework) | 66.70 | |
Normally resits will be assessed by the same methodology as the first attempt, unless otherwise stated
Exams
| Exam type | Exam duration | % of formal assessment |
| Unseen exam | 2 hr 00 mins | 33.30 |
| Total percentage (Assessment Exams) | 33.30 | |
Normally resits will be assessed by the same methodology as the first attempt, unless otherwise stated
Reading list
The reading list is available from the Library websiteLast updated: 30/04/2019
Browse Other Catalogues
- Undergraduate module catalogue
- Taught Postgraduate module catalogue
- Undergraduate programme catalogue
- Taught Postgraduate programme catalogue
Errors, omissions, failed links etc should be notified to the Catalogue Team.PROD