Level: Professtional (P)

This two-day course is designed to give participants an introduction to the theory and application of multilevel regression models for clustered or hierarchical data using the statistical software (participants can choose between R, Stata and MLwiN). We focus on two-level linear and logistic regression models for clustered cross-sectional (individuals nested within groups) and longitudinal (repeated measures nested within individuals) data. The course consists of a 2:1 mix of lectures and “hands on” software practical sessions. Each new modelling development is illustrated with real applications to social, behavioural and medical science data sets.

Learning Outcomes

• An appreciation of the implications of ignoring clustered cross-sectional and longitudinal data structures in statistical modelling

• An understanding of a range of multilevel regression models for analysing clustered and longitudinal data

• Practice at estimating two-level linear and logistic regression models in statistical software, and at interpreting and plotting the results
   

Topics Covered

• Overview of multilevel modelling
• Variance-component models
• Random-intercept models with covariates
• Between - and wiithin - effects of level-1 covriates​
• Random-coefficient models
• Growth-curve models
• Three-level models
• Review of single-level logistic regression
• ​Two-level logistics regression

Target Audience

Applied quantitative researchers with an interest in the analysis of clustered cross-sectional and longitudinal data.

Assumed Knowledge

The course will not assume a high level of statistical knowledge, but participants will be expected to have a good understanding and experience of applying and interpreting conventional linear and logistic regression models. A set of on-line training materials to enable participants to refresh their knowledge of these topics are available from the Centre for Multilevel Modelling (CMM) website in Bristol (http://www.cmm.bristol.ac.uk/learning-training/course.shtml).

Delegates will need to install in advance their choice of statistical software (MLwiN, R or Stata) on their laptop and bring this to the course.

We assume participants intending to use R or Stata are familiar with these software. We assume participants intending to use MLwiN are new to the software. Those participants can download MLwiN from http://www.bristol.ac.uk/cmm/software/mlwin/download/

Delegate Feedback

“The course was organised and the topics were well-structured.”

“Good balance of practical & theory, knowledgeable & helpful speakers."

“Excellent course.  Very challenging but pushed me to understand what I needed – I had a breakthrough.  Thank you!”

Professor George Leckie

George is a Professor of Social Statistics and Co-Director of the Centre for Multilevel Modelling at the School of Education, University of Bristol, UK.

His methodological interests are in the development, application and dissemination of multilevel and related models to analyse educational and other data. His substantive interests focus on design, analysis, and communication issues surrounding school performance measures and league tables, especially the use of value-added models for estimating school effects on student achievement for accountability and choice purposes.

He was awarded the Frances Wood Medal in the Royal Statistical Society's 2017 honours for his contributions to Social Statistics over the past 10 years, especially his school league table research.

George has taught over 50 multilevel modelling short courses across UK, Europe, Australia and US.

Professor William J Browne

Bill is a professor of statistics at the University of Bristol. His research spans the area of statistical modelling, from the development of statistical methods to fit realistically complex statistical models to describe real-life problems, through the implementation of those models in statistical software to the application of the methods in several application areas.

Much of his computational work has been in the development of statistical software within the Centre for Multilevel Modelling and in particular in the development of the MLwiN and StatJR packages.

His methodological work has been with regard MCMC methods for fitting these multilevel models and also for tailoring MCMC algorithms for specific other models.

He has  worked on applying multilevel and other statistical models to problems as diverse as assessing welfare in chickens, modelling herd breakdowns with TB in cattle, investigating nesting behaviour of great tits and the impact of class sizes on primary school children.