[MARMAM] A brief exchange on modelling small datasets (of vaquita)

Mike Lonergan mel5 at st-andrews.ac.uk
Tue Apr 24 04:58:47 PDT 2012


Marine Mammal Science has kindly published (currently Online Early) a comment I wrote:

Lonergan, M. 2012. Insufficient data are available to predict the success of protected areas for the vaquita (Phocoena sinus): A critique of Gerrodette and Rojas-Bracho (2011). Marine Mammal Science.
http://onlinelibrary.wiley.com/doi/10.1111/j.1748-7692.2011.00537.x/pdf

It was querying this paper:

Gerrodette, T., and L. Rojas-Bracho. 2011. Estimating the success of protected areas for the vaquita, Phocoena sinus. Marine Mammal Science 27:E101–E125.
http://onlinelibrary.wiley.com/doi/10.1111/j.1748-7692.2010.00449.x/pdf

And elicited this response (also currently Online Early):

Gerrodette, T., and L. Rojas-Bracho. 2012. Inference from limited data: A response to Lonergan. Marine Mammal Science.
http://onlinelibrary.wiley.com/doi/10.1111/j.1748-7692.2012.00569.x/pdf


The Comments have no abstracts, but my (probably biassed) summary of the exchange is:

I was surprised at a claim that vaquita abundance in 2008 could be estimated much more precisely from a model based on limited historic data than from a survey carried out in that year. I concluded that the model understated some sources of uncertainty, and listed examples.

Gerrodette and Rojas-Bracho responded to some of my detailed points, and said that their analysis "demonstrated the remarkable power of modern statistical methods".


Essentially we differ on which intuition to favour when unexpected results arise from a plausible representation of a small dataset. Simulation can't entirely answer that question since results should hold up if all their assumptions and priors are taken as correct but will fail if there are sufficiently gross violations - that then leaves the issue of how well the model structure approximates reality. I believe this to be a contained example of the more 
general practical problem of interpreting limited datasets and appropriately expressing both the strength of the evidence they provide and the assumptions that interpretations depend on. It matters because we mislead if we overstate results but waste information if we overemphasise uncertainty.

The pdfs are on the MMS website. I (mel5 at st-and.ac.uk) am happy to send my Comment to interested individuals; Tim Gerrodette (tim.gerrodette "at" noaa.gov) is listed as corresponding author on the other papers.

I would like to hear people's opinions of the wider issue.

Regards,

Mike.

-- 
Mike Lonergan
NERC Sea Mammal Research Unit,
Scottish Oceans Institute,
University of St Andrews
________________________________________________________________________________
"The University of St Andrews is a charity registered in Scotland : No SC013532"




More information about the MARMAM mailing list