Module Descriptors

Statistics

ST413: Statistical Modelling (5 ECTS)

An advanced course on statistical modelling, building on the regression modelling introduced in MA322/ST313, covering the theory and application of generalized linear models and their fitting in R

Taught in Semester(s) II. Examined in Semester(s) II.

Workload: 40 hours (24 Lecture hours, 6 Tutorial hours, 10 Lab hours).


Module Learning Outcomes. On successful completion of this module the learner should be able to:

  1. Formulate a normal model with continuous and categorical explanatory variables, fit such models in R, interpret model output, make inferences using both parameter estimates and ANOVA tables.
  2. Check adequacy of normal models using standard model diagnostics and modify models using transformations and link functions.
  3. Formulate and derive properties of generalized linear models (glms).
  4. Obtain maximum likelihood estimating equations for glms and derive the 5. Iterative Weighted Least Squares algorithm.
  5. Formulate and use (in R) binomial regression models to analyse proportion data with both continuous and categorical explanatory variables.
  6. Formulate and use (in R) Poisson regression models to analyse count and rate data.
  7. Analyse and test associations in multi-way table using log-linear models. Analyse censored survival data using parametric and non-parametric (Cox proportional hazards) models.


Indicative Content

This course provides an introduction to generalized linear models and their application. A key component of the course will be the use of R (open-source statistical software) for statistical model fitting. The topics covered include

  1. Revision of the general linear model. Introduction to model fitting in R.
  2. Generalized linear models (glms): exponential family, link function, variance function.
  3. Fitting glms: Iterative Weighted Least Squares; analysis of deviance; inference; diagnostics.
  4. Binomial regression models for proportions; binary regression models; contingency tables.
  5. Log-linear models for count data; Poisson regression models; multinomial models; multi-way tables.
  6. Models for time to event data; exponential regression; survival data, censoring; Weibull model; Cox proportional hazards model.


Module Resources

  1. Statistical Modelling In R by Aitkin, Francis, Hinde & Darnell, Oxford.
  2. An Introduction to Generalized Linear Models by Dobson & Barnett, CRC Press


Back