You would have learnt about several hypothesis tests already in your degree, such as t-tests and F-tests (think ANOVA). These tests are designed for the situation where you have a single hypothesis you want to check and you know how many observations you’re going to collect before you collect your data.
Unfortunately, these assumptions can be problematic in real applications. For example, consider the situation where it’s time-consuming and expensive to collect data. Can you stop early if you’re getting very compelling evidence early on in your data collection? Can you collect more observations than you’d originally intended if the evidence looks compelling so far? Both of these ideas mean you’re peeking at the data before deciding what test you’re going to do and that will affect how trustworthy your results are. This project will investigate the impact of optional stopping and optional continuation procedures.
We will also investigate multiple hypothesis testing procedures for when people want to check multiple hypotheses at once (like Tukey’s multiple comparisons procedures).
The project will mostly involve learning about, coding up and comparing different methods. If time permits, we will investigate novel methods for assessing whether samples come from a hypothesised distribution, which has important applications in scalable Bayesian inference.
You can expect to:
- meet regularly with Dr South
- develop your knowledge of hypothesis testing procedures
- learn about the impacts of performing multiple hypothesis tests, doing optional stopping and doing optional continuation
- implement methods in R, run simulations and create figures to summarise the results
- be involved in the development of a new hypothesis testing procedure.
- code for various hypothesis testing procedures
- a comparison of different hypothesis testing procedures
- potential co-authorship of a journal article or conference paper.
Skills and experience
Some experience in programming (R preferred, followed by Matlab or Python)
You may be eligible to apply for a research scholarship.
Contact the supervisor for more information.