Non-likelihood Based Statistical Modelling
7-9 September 2015
SCOPE
Likelihood forms the foundation of standard statistical inference, regardless of one’s allegiance to the Bayesian School or the Frequentist School. Its role has rarely been challenged. However, likelihoods are predicated on having at one’s disposal a true data-generating mechanism, commonly called a model. The question can be raised as to whether likelihood-based approaches are fit for purpose when the true model is unavailable. Moreover, the usually unavoidable observation error will add to the difficulties for these approaches. Since many statisticians accept that “all models are wrong” (-usually attributed to the late George Box), it is high time for statisticians to focus on developing statistical methods to deal with the practically most common situation, which is when we know that our model cannot be the true data-generating mechanism.
Format of the event
If you can remember the 1989-Edinburgh Workshop on Nonlinear Time Series, you will know the format, which incidentally has also been adopted by all the Oberwolfach workshops that I have attended since 1989. Basically, I'll look for 'volunteers' for Day 2 and Day 3 at the end of Day 1 and Day 2. This way allows people to respond to and interact with talks given the day(s) before. Speakers on Days 2 and 3 have a fairly free choice of time duration. Their talks can be as short as 15 minutes for example but they will never exceed 25 minutes. Additionally, we allow 5-10 minutes for discussion after each talk, as discussions form an important part of the workshop. In principle, one can volunteer for more than one talk, if time permits. No written contributions are expected but please do feel free to distribute them, if available.
Day 1 Programme
9:00 Welcome
9:30 David Cox
10:30 Tea and coffee break
11:00 Laurie Davies: Data analysis and approximate models (slides revised version 1 Sept)
12:00 Lunch
13:30 John Copas: Likelihood and wrong models
14:15 John Kent: Closed exponential families and composite likelihood
15:00 Tea and coffee break
15:30 Paul Fearnhead: Detecting changepoints in the presence of model error
16:15 Howell Tong: Is likelihood fit for purpose? (backround reading material)
17:00 Free discussion
18:00 Reception drinks and canapés
ORGANISERS
Howell Tong (Scientific Organiser)
Bärbel Finkenstädt (Local Organiser)
Correspondance should be sent to crism at warwick dot ac dot uk
PARTICIPANTS
- Wicher Bergsma (LSE, UK)
- Kung-Sik Chan (Iowa, USA)
- David Cox (Oxford, UK)
- John Copas (Warwick, UK)
- Laurie Davies (Duisburg-Essen, Germany)
- Mirela Domijan (Cambridge, UK)
- Paul Fearnhead (Lancaster, UK)
- Piotr Fryzlewicz (LSE, UK)
- Zhaoxing Gao (Hong Kong)
- Simone Giannerini (Bologna, Italy)
- John Kent (Leeds, UK)
- Tony Lawrance (Warwick, UK)
- Dong Li (Beijing, China)
- Guodong Li (Hong Kong)
- Wai-Keung Li (Hong Kong)
- Pak Hang Lo (Kaiserslautern, Germany)
- Alessandra Luati (Bologna, Italy)
- Kanchan Mukherjee (Lancaster, UK)
- David Rand (Warwick, UK)
- Granville Tunnicliffe-Wilson (Lancaster, UK)
- Qiwei Yao (LSE, UK)