CSIRO Marine and Atmospheric Research
Past Seminars

Friday 20 February 2004, 11.30 am (Tas time)

CSIRO Auditorium, Hobart and via videoconference to Floreat and Cleveland

Tore Schweder
Professor, Economics Institute,
University of Oslo, Norway

Are complex empirical models deep in the water or can the uncertainty in results be netted in?

The quantification of uncertainty is central in empirical research, and is a thorny issue when complex models are used to integrate diverse data in stock assessment, ecology, oceanography and elsewhere. I will address this issue by first briefly reviewing the history of statistics: The initial paradigm was Lapalace’s inverse probability from 1774. Fisher revolted against this Bayesian approach in 1930, but the Bayesians have staged a counter revolution against his frequentist methods and are now on the offensive. The Bayesian’s posterior distribution representing information/uncertainty is indeed attractive and so is the rational updating of information. However it often needs subjective input to get going and results can be biased due to the ocean of parameters and non-linearities in the model relative to the information content in the data. I will present a frequentist approach based on confidence distributions and likelihoods in the spirit of Fisher and Neyman. The confidence distributions will be presented in the format of confidence nets (my term) obtained by simulating the data and thus uncovering potential bias and sampling variability in the objective function. I will illustrate this on bowhead whales off Alaska.

[Back to Seminars]

Legal Notice and Disclaimer | Copyright