APPLIED STATISTICS WEEK 4
11-13 May Choosing and Fitting Regression Models for Normal and Non-Normal
Data
14-15 May Model Checking
For registration details or further information please contact Kellie
Watkins ([log in to unmask]), providing your address and/or fax number.
----------------------------------------------------------------------------
----------------------------
Choosing and Fitting Regression Models
This course is intended for scientists and statisticians who have a basic
knowledge of regression and need to become familiar with more advanced
regression modelling. Our objectives are to introduce some of these
methods and also to give participants the confidence to read the literature
on further methods that are not covered.
The course begins with a half-day review of standard regression modelling.
This sets the scene and introduces some of the software that is used later
in the course, in particular SAS, Genstat and S-Plus.
The first extension is the use of logistic regression to model binary data.
This session includes methods for model checking and will also consider
the use of exact rather than asymptotic methods, using the package
Logexact. This is followed by the analysis of ordered categorical data,
using ordinal logistic regression.
The next area is that of regression with many variables. Modern
instrumentation has led to prediction problems with very large numbers of
potential explanatory variables. A number of 'regularisation' techniques
which have been developed will be discussed. Topics covered include ridge
regression, principal components regression, partial least squares
regression and continuum regression.
The final area that is covered in some detail is that of non-linear
regression modelling. Examples include exponential curves and inverse
polynomials.
The last session of the course is a forum. This provides opportunities for
the resource persons to mention a a range of further topics and also for
participants to introduce topics on which they would like discussion. We
would welcome prior information on such topics, either when applicants
apply, or at the start of the course.
Model Checking
Model checking is an essential part of the process of staticial modelling.
If an inappropriate model is used, inferences about quantities such as the
magnitude of a treatment effect, or the association between certain
explanatory variables and a response variable, will be invalid.
In this course we will describe and illustrate modern techniques for
checking the adequacy of the general linear model and models used in the
analysis of binary, categorical and survival data. We will show how the
methods can be applied using standard statistical software such as SAS.
The course will include practical work based on the use of SAS.
The course will begin with a review of the process of statistical modelling
and the role of model checking. The general linear model will be outlined
and methods for model checking, using residuals and influence diagnostics,
described. Following brief reviews of the linear logistic model for binary
data, the proportional odds model for ordered categorical data and the
proportional hazards model for surival data, techniques for assessing the
adequacy of these models will be described. The course will include some
discussion on how to proceed when a fitted model is found to be
unsatisfactory.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|