Applied Quantile, M-quantile and Expectile Regression Analysis (join a waiting list)
Date:
09/02/2017 - 10/02/2017
Organised by:
NCRM, University of Southampton
Presenter:
Professor Nikos Tzavidis
Level:
Intermediate (some prior knowledge)
Contact:
Jacqui Thorp
Training and Capacity Building Coordinator
NCRM, University of Southampton
Email: jmh6@soton.ac.uk
Tel: 02380 594069
Map:
View in Google Maps (SO17 1BJ)
Venue:
Building 39, University of Southampton, Highfield, Southampton, Hants
Description:
The course will cover a range of modern statistical modelling tools that considerably extend the scope of popular modelling approaches such as linear regression, generalised linear models and multilevel modelling. We introduce quantile and more generally M-quantile and expectile regression for continuous outcomes and then extend these models to handle discrete outcomes and hierarchical (multilevel and longitudinal) data structures. We further demonstrate how grouping-induced heterogeneity in these data can be semi-parametrically characterized by using quantile and M-quantile regression. The use of these ideas in prediction will be discussed with particular applications in small area estimation. Applications of multilevel and longitudinal data analysis using data from psychology and medicine will also be presented alongside existing and new functions in R for implementing the models.
The course covers the fundamental statistical background necessary to understand the ideas underpinning the use of both quantile and M-quantile regression as a semiparametric robust alternative to standard and multilevel regression. Substantive examples of the application of these ideas will be presented, with the aim of providing practical guidance on their use in realistic applications. Topics include:
- Quantiles and M-quantiles
- Regression quantiles and regression M-quantiles
- Likelihood methods for quantile and M-quantile regression: The role of the Asymmetric Laplace Distribution and its generalizations
- Robust M-Estimation for Generalized Linear Models
- Quantile and M-quantile regression for counts and binary outcomes
- Using M-quantile regression to characterize hierarchical data structures
- Multilevel (random effects) models for hierarchical data
- Quantile regression with random effects models
- Modelling heterogeneity via M-quantile regression
- Prediction with M-quantiles and quantiles – Small Area Estimation
- Applications
- R software
By the end of the course participants will:
- Understand the use of quantile, M-quantile and expectile regression with continuous and discrete outcomes;
- Become familiar with some of the background theory;
- Apply the methods to real data;
- Be trained in the use of statistical software for implementing related methods.
The course will use the R software. – Prior familiarity with R would be an advantage although the course will cover some R basics.
This course is appropriate for any statistician, doctoral students and post-doctoral researchers with a good grounding in standard modelling tools, particularly the use of regression modelling and generalised linear models. Statisticians in National Statistical Institutes will benefit from the applications addressing small area estimation.
Cost:
The fee per teaching day is:
• £30 per day for UK/EU registered students
• £60 per day for staff at UK/EU academic institutions, UK/EU Research Councils researchers, UK/EU public sector staff and staff at UK/EU registered charity organisations and recognised UK/EU research institutions.
• £220 per day for all other participants.
All fees include event materials, lunch, morning and afternoon tea. They do not include travel and accommodation
Website and registration:
Region:
South West
Keywords:
Small Area Estimation, Regression Methods, Multilevel Modelling , Longitudinal Data Analysis
Related publications and presentations:
Small Area Estimation
Regression Methods
Multilevel Modelling
Longitudinal Data Analysis