MATH5945 Categorical Data Analysis

MATH5945 is a Honours and Postgraduate Coursework Mathematics course. See the course overview below.

Units of credit: 6

Prerequisites: Nil

Cycle of offering: Yearly in Semester 1

Graduate attributes: The course will enhance your research, inquiry and analytical thinking abilities.

More information: This recent course handout (pdf) contains information about course objectives, assessment, course materials and the syllabus. 

The Online Handbook entry contains up-to-date timetabling information.

If you are currently enrolled in MATH5945, you can log into UNSW Moodle for this course.

Course Overview

Data analysts often face a situation where the response outcomes are categories rather than being measured on the interval scale. The accompanying explanatory variables may also be categorical or be continuous. Such type of data is abundant in social sciences, in medical research, particularly in epidemiology and biostatistics, in market research and in other areas. Categorical data are often obtained as counts and presented in the form of contingency tables.

Studying relationships between categorical variables cannot be done using standard regression-type techniques based on the assumption of normality and requires specific methods and techniques.

The core methodology used is the methodology of the generalised linear models. Within this framework, we shall study log-linear models, logistic regression, Poisson regression, logit and probit models and analysis of categorised time-to-event data. Specific attention will be paid to the Generalized Likelihood Ratio testing methodology and its application for choosing the "most suitable" model within a hierarchical set of models.
The classical logistic regression models will be extended to cover polytomous responses. The latter are unordered categorical responses which in econometrics are called discrete choices. Computing features prominently in the course and the techniques will be illustrated with the SAS package.