Advanced Topics in Regression A: (STAA0010A)Back to Intranet events
|Date:||Five 3 hourly sessions during period Mondays 26/2-26/3|
|Time:||5:30PM to 8:30PM|
|Venue:||BA513 Hawthorn Campus|
Course: STAA0010A - Advanced Topics in Regression A:
Generalised Linear Model
Dates: Five 3 hourly sessions during period Mondays 26/2-26/3 (5:30pm – 8:30pm)
Assumed Knowledge: Multiple Linear Regression (e.g. STAA0005A)
Software used: SPSS
Maximum 20 students
Statistical techniques as listed below will be covered with an emphasis on the interpretation and reporting of these results.
A review of multiple linear regression with special attention to assumptions, unusual point identification and multicollinearity. Different regression techniques are introduced and tests for mediation and moderation are illustrated. A variety of methods for improving the fit of regression models are provided. Methods for weighted regression, nonlinear regression methods and the General Linear Model are then introduced, always assuming that residuals are independent and normally distributed.
When normal assumptions are no longer valid the Generalised Linear Model is used. These models are introduced and then we look particularly at categorical variables. Starting with Crosstab analyses we learn how to define residuals that help us to interpret relationships between categorical variables. Special measures of association are developed for particular types of categorical variables. Finally multi-order crosstab tables are introduced together with the loglinear analyses required to test for multi-way interaction effects.
Binary logistic regression is a special generalised linear model for binary response variables which uses a logistic link function. We learn how to interpret odds and odds ratios and then show how binary logistic regression is used in practice to fit models with more than one predictor variable. Univariate binary logistic regression models are first fitted using each predictor in turn with a multiple binary logistic regression model to follow. This allows us to test for mediation. Finally, ROC curves and the Hosmer-Lemeshow test are used to assess goodness of fit.
Ordinal logistic with an ordinal response variable are then introduced and tested for “parallel lines”. Nominal logistic regression does not assume parallel lines and can be used with categorical response variables which are not ordinal but have more than two categories, requiring the choice of a reference category.
Online quizzes will be available for self-assessment.