Skip navigation


It is a little misleading to name only one theme in this course “Statistical modelling” because all statistical inference involves a statistical model of some description; for example, a CI and hypothesis test for a difference between two sample means uses a statistical model for the behaviour of the mean difference over repeated random sampling. We used the word statistical modelling to highlight the material in this theme because of the focus on regression models, which were developed to control for other variables (and so model relationships) and also for prediction purposes. And so we have the idea of selecting and choosing between models based on how well they fit the data – or put anotherway "modelling".


This theme starts with a short video that tries to set the scene by discussing the idea of a theoretical model and a regression model, and how a regression model can be used for the purpose of testing a theoretical model and explaining the relationships within it. It also highlights the use of graphs to think about models and patterns, in particular showing what confounding looks like graphically and how a multivariable model can be used to control for confounding, and what an interaction looks like graphically and how a regression model can test for, and estimate this interaction.


Lastly there is an e-lecture lasting about 60mins. This e-lecture discusses three types of regression model - linear regression, logistic regression and Cox proportional hazards regression. Each are used for a particular type of outcome, namely, continuous, binary, and time to an event. The objective is to help you understand the purpose behind these methods.