MATH5945: Categorical Data Analysis

- 3 x 15% assignment
- 60% final exam

No formal prerequisites stated (quite common for postgraduate courses). Assumed knowledge is pretty much any general level of statistics (for UNSW undergraduates, definitely MATH2801/2901). I recommend having some in-depth knowledge of statistical inference (for example, equivalent of MATH3811/3911) and/or statistical modelling techniques (for example, equivalent of MATH3821, especially GLMs) before coming to this course. Some overlaps are shared with MATH3851, but it is not a prerequisite.

This is one of many postgraduate statistics courses. From what I can see, it used to be on a 2-year offering, but has recently been moved up to every year offering.

The course is as its title says; the focus is on analysis of categorical data. Categorical data rises in tons of ways; some examples being whether you wore a helmet and/or involved in a crash, what kind of drug were you given as treatment for a sickness, breaking of ages into age groups (as opposed to the exact age itself) etc.. The course teaches a huge amount of techniques that statisticians use to incorporate categorical data into their studies.

This course definitely leans more on the applied statistics side, but some theory was also examinable. Don't expect a level 5 maths course to have absolutely zero algebra. But to those that want to focus on useful statistics skills, there is a ton of value. Applications were definitely the focus in my opinion.

SAS was used because Jake believes SAS is the superior option for categorical data. Having now done this course, I'm honestly not surprised. SAS gives a ton of output for categorical data, and it feels REALLY automated (no fidgeting around with the code). Of course, battling SAS can be a bit annoying. Jake assumes that you come into this course with statistical programming background of some sort (e.g. R), but doesn't assume you know SAS. There's a bit of guidance along the way.

Assignments weren't actually that hard (but I will give a piece of advice - when it says "estimate", give confidence interval estimates as well...). It was partly because guidance was subtly given through the lectures and labs. Occasionally it was possible to just copy code Jake provided, and appropriately adapt it to the problem at hand. The 5 page limits for the assignments should NOT be an issue.

Studying for the exam felt a lot like studying for MATH3821 again (only this time, there's no tutorials to worry about). It's pretty scary, because there's so much content out there, and we basically had no clue how it was gonna be examined. I spent a lot of days being really concerned for this paper, and it wasn't until I saw the actual paper I was like "oh wait this is friendly". Bless Jake for that; it's rare having stress pay off.

The final exam was open-notes (you're only using the course material; you shouldn't even need the internet). Carefully studying the notes helped a lot. It was very easy to look things up in the exam, because I knew where to look! If the exam is open-book again, that's a very valuable piece of advice I'd want to give.

1 x 3hr lecture, 1 x 1hr lab

4/5 (although surprisingly the final exam was a 2.5/5)

Yes during COVID period. Presumably yes otherwise, since Jake uses lecture slides.

Prof. Jake Olivier

A comprehensive 700+ lecture slides. Takes a long while to absorb everything. No tutorials (quite common for postgrad maths courses). One SAS lab every week. Not too many resources otherwise (I only ever used Google for SAS in this course.) No past papers.

4/5

None prescribed. Two reference textbooks are: Agresti A. (2012) An introduction to categorical data analysis, 3rd Edition. Wiley, and Dobson AJ, Barnett AG. (2008) An introduction to generalized linear models, 3rd edition. CRC Press. A SAS textbook was also referenced.

20T3

97 HD