Statistical model selection: guidance for the researcher

Time is AEST

Seminar Abstract:

Researchers in medicine, epidemiology, psychology, biology, conservation, economics and many other disciplines seek the betterment of society through increased understanding in their disciplines. What statistical model selection should be used (if any) and what is the impact of this selection on the ability of the researcher to reach this goal?

The researcher is faced with the following questions. Which is better, the Akaike Information Criterion (AIC) or the Bayesian Information Criterion (BIC) for model selection? How many preliminary statistical tests of the model assumptions should one carry out before one proceeds confidently with the chosen model? Should one avoid statistical model selection altogether? What is the oracle property? What further research on statistical model selection is needed? Using as little technical work as possible, the speaker will invite the audience into a conversation about these questions.

Speaker Bio:

The speaker has over 20 years of experience applying and analyzing statistical model selection procedures. He has 24 publications in this field, including the paper “The coverage properties of confidence regions after model selection” in International Statistical Review, 2009.

Host institution: La Trobe University

Seminar Convener: Andriy Olenko

AGR Support: Darren Condon

How to participate in this seminar:

1.Book your University’s Access Grid Room, or a university or APAC etc. Access Grid Room that you are able to use, and

2. Send an email to the seminar convenor at La Trobe University (Andriy Olenko) with a cc to Darren Condon to advise that you will be attending the seminar.