We lost time in the module due to the industrial action that took place in February and March. As a result, we will not have time to cover everything that is in the notes. I have decided to remove Chapter 6 on the EM algorithm: there will no questions about Chapter 6 on the exam. However, it will still be useful for you to read section 6.1 from the notes on ignorable missing data mechanisms.
Note that Chapter 7 is still examinable. I have removed Chapter 6 as it is the most difficult mathematically - in previous years I have given students a lot of individual help on the EM algorithm, and as that will not be possible this year, this is the most sensible material to remove.
For the avoidance of doubt, here is the complete list of examinable topics for this year (2019-2020), listed by the chapter in which they are covered in the notes.
If you have any questions, please post them on the Blackboard discussion boards.
A complete list of video lectures for this year is given below.
The printed notes for this module are available here. Note that Chapter 6 is not being lectured or examined this year (due to the time lost from the industrial action).
There are a number of other handouts from the lectures:
Lectures 1-4 happened before interruptions from the industrial action or COVID-19 took place. You can view these on youtube. For the remaining lectures, I have put together videos from previous academic years. These are listed below. Please note that in some of these lectures I discuss assessed work that was not set this year.
Lecture 5 (16 March) part I and part II. I give a recap of matrix notation, and the first discussion of how to fit models to data using maximum likelihood.
Lecture 6 - 17 March part I and part II on restricted maximum likelihood (REML) - which is how mixed effects models are usually fitted to data.
Lecture 7 - 20 April available here, and is on best linear unbiased prediction (BLUP), and on some properties of multivariate Gaussians.
Lecture 8 - 21 April available here and is on split-plot models.
Lecture 9 part I and part II on split-plot models and random effect interactions.
Lecture 10 part I and part II on random effect interactions and diagnostic plots.
Lecture 12 part I and part II on bootstrap hypothesis testing and confidence intervals.
Lecture 14 available here on simple approaches to dealing with missing data. You only need to watch this lecture up to the 24 minute mark. I have removed Chapter 6 from the syllabus this year.
Lecture 15 part I and part II on multiple imputation. For the code, please see the Lecture19_ToyMultipleImputation.R script.
Lecture 16 available here on multiple imputation. For the code, please see the Lecture20_Mice_by_hand.R script.
For those who are interested, the full list of videos from previous years is available: 2018-19, 2016-17. Note that we have a slightly reduced syllabus this year due to the industrial action.
You will be able to complete sheet 1 after lecture 8, sheet 2 after lecture 13, and sheet 3 after lecture 20. The Exercises require the data in the following workspace: MAS473.RData
I strongly recommend you attempt the questions before looking at the solutions.
Some of the tasks given in the notes require R scipts. These can be found below:
The R code uses data available in the following workspace: MAS473.RData
The past exam papers are available here. The solutions are listed below.
Please note that the module was revamped in 2016-17, and a new part was written on missing data. Prior to 2016-17 the module was called MAS473, and contained the material on mixed effect models as well as some material on GLMs. Hence, for the exams prior to 2016-17 not all questions will be relevant to this module.
NOTE: There will be no questions on the EM algorithm on the 2019-2020 exam.
No additional past papers or solutions will be provided.