MAS465 Multivariate Data Analysis
|Semester 1, 2017/18||10 Credits|
|Lecturer:||Dr Frazer Jarvis||uses MOLE||Reading List|
|Aims||Outcomes||Teaching Methods||Assessment||Full Syllabus|
The analysis of multivariate data requires the extension of standard univariate statistical models and methods but also introduces new problems. Initial attention is given to data mining techniques such as summarising and displaying high dimensional data and to ways of reducing multivariate problems to more manageable univariate ones. This is followed by routine generalisations of standard distributions and statistical tests before consideration of new strategies for constructing hypothesis tests. Finally, problems specific to multivariate data such as discrimination and classification (use in medical diagnosis problems for example) are studied. Most of these methods can be implemented in standard computer packages.
Prerequisites: MAS223 (Statistical Inference and Modelling)
No other modules have this module as a prerequisite.
- Multivariate data summary: sample estimates of mean, covariance and variance
- Graphical displays: scatterplots, augmented plots, Andrews' plots, special techniques.
- Exploratory analysis and dimensionality reduction: principal component analysis, principal component and crimcoord displays, implementation in R.
- Multidimensional scaling: visualisation of similarity data.
- Linear discriminant analysis: visualisation of grouped data, linear discriminant analysis in R.
- Multivariate normal distribution: basic properties, confidence regions, simple hypothesis tests, statistical discriminant analysis.
- Single and two sample methods: Hotelling's T2 test, practical implementation in R.
- Construction of statistical hypothesis tests: the likelihood ratio method and the union-intersection principle, MANOVA, implementation in R.
- To illustrate extensions of univariate statistical methodology to multivariate data.
- To introduce students to some of the statistical methodologies which arise only in multivariate data.
- To introduce students to some of the computational techniques required for multivariate analysis available in standard statistical packages
- have some understanding of techniques of multivariate data summary and graphical display and of the principles of multivariate exploratory data analysis and dimensionality reduction;
- have some understanding of the construction of multivariate likelihood ratio tests and of the union-intersection principle in multivariate testing;
- be able to perform and interpret principal component analysis and linear discriminant analysis using a computer package;
- be able to understand the results of computer based multivariate analyses of one and two sample tests;
- be familiar with facilities offered by computer packages for multivariate analysis.
Lectures, problem solving
20 lectures, no tutorials
One formal 2 hour written examination. Format: 3 questions from 4 [75%]. Project [25%].
Multivariate data summary
- basic notation, sample estimates of mean, covariance and variance
- Scatterplots, augmented plots, Andrews plots, special techniques
- principal component analysis
- principal component and crimcoord displays
- Multidimensional scaling
- Linear discriminant analysis
- Cluster analysis and other data mining techniques
- multivariate normal distribution and confidence regions
- Hotelling's T2 test
- statistical discriminant analysis
- practical implementation in R
- the likelihood ratio method in multivariate data
- the union-intersection principle
|B||Chatfield and Collins||Introduction to Multivariate Analysis||519.53||Blackwells||Amazon|
|B||Cox||An introduction to multivariate data analysis||519.535||Blackwells||Amazon|
|B||Everitt||An R and S-PLUS companion to multivariate analysis||519.535||Blackwells||Amazon|
|B||James, Witten, Hastie, Tibshrani||An introduction to statistical learning||Blackwells||Amazon|
(A = essential, B = recommended, C = background.)
Most books on reading lists should also be available from the Blackwells shop at Jessop West.