 |
 |
Hi Guullaume, I wrote a surplus production model for Cusk which can be found here: http://www.seaturtle.org/pdf/DaviesTD_2011_CanJFishAquatSci.pdf What I found was DIC for this model was an extremely poor measure for model selection for my data and model. I spend quite a few months simulating data and testing trying to figure it out and just ended up tossing it. " We also discovered that DIC alone may not be a reliable metric for some Bayesian state-space model applications. An additional model we preliminarily tested, but is not described in the manuscript, was to test for a knife-edge change in catchability in the trawl survey in 1992 when the survey trawl index rapidly declined. According to DIC, this model had the most support using both the real and the simulated data, even though in the latter case we specified that a change in catchability was not responsible for the declines. This underscores the importance of using simulated data to investigate further the plausibility of the models." Without looking at your code & data it's hard to provide more advice (and actually right now I don't have time to do either - sorry!). My code is also in that paper which may possibly be of help. I hope that is some help to you. Trevor > Dear list, > > I'm currently developing a Schaeffer surplus production model (P[t] = > P[t-1] + r * P[t-1] *(1-P[t-1]) - C[t]/K ) for a redfish stock that > covers 2 NAFO divisions. > So far, the various model used (non Bayesian) have been pooling the > different catches and surveys available in the two divisions but I'm > trying to make the best use of the data available by looking at what is > happening in the 2 divisions (r is shared and there is one K for each > division). > > Something that is important too is that for one surveys covers both > division and it cannot be split. > > So after developing this "spatially" structured model I wanted to check > how it performed in comparison to a non "spatially" structure model. > So I checked the residual and how the posterior distributions looked > like in both model and looking in particular at the parameters r and K > that are of particular interest for me. > > in the non spatially structured model, the r parameters is not well > estimated at all (the posterior distributionis highly skewed and the > mode is close to 0 which is not what is expected and encountered in the > litterature for this species. > > in the spatially structured model, the r parameter is well estimated > (nice posterior distribution with a clear node) , all the residuals seem > distributed around 0 , so I thought ok that is great I have a keeper. > > > > However I looked at the DICs of the 2 models and I had some > "interesting" results" > > > > For the non-spatially structured model: > Dbar Dhat DIC pD > > > total 828.9 824.0 833.9 4.977 > > > > For the spatially structured model: > > Dbar Dhat DIC pD > > > total 1176.0 1179.0 1174.0 -2.163 > > > > So based on these DIC, the non spatially structured model is a better > fit to the data, however it provides more uncertain estimates of the > population abundance than the spatially structured model and it > struggles estimating some of the key parameters of the model. > > I guess my question is can these two models be compared through the DIC > since there are structurally two different models. > > > > If more information is needed, I'm happy sending the code along. > > Thanks in advance for your help > > > > Guillaume > > > > -- > > Guillaume Dauphin Ph.D. > > NSERC Post-Doctoral Fellow > > Recovery strategies for straddling stocks: 3LNO American > plaice(Hippoglossoides platessoides) and 3NO cod (Gadus morhua). > > Science Branch > Department of Fisheries and Oceans > Northwest Atlantic Fisheries Center > 80 East White Hills Road > PO Box 5667 St John's NL > CANADA A1C 5X1 > > > > Email: [log in to unmask] > > > > > ------------------------------------------------------------------- > This list is for discussion of modelling issues and the BUGS software. > For help with crashes and error messages, first mail [log in to unmask] > To mail the BUGS list, mail to [log in to unmask] > Before mailing, please check the archive at www.jiscmail.ac.uk/lists/bugs.html > Please do not mail attachments to the list. > To leave the BUGS list, send LEAVE BUGS to [log in to unmask] > If this fails, mail [log in to unmask], NOT the whole list > -- Trevor Davies Ph.D. Candidate Department of Biological Sciences 1355 Oxford Street Dalhousie University Halifax, Nova Scotia, Canada, B3H 4J1 ph: (902) 494 3910: fax (902) 494 3736 ------------------------------------------------------------------- This list is for discussion of modelling issues and the BUGS software. For help with crashes and error messages, first mail [log in to unmask] To mail the BUGS list, mail to [log in to unmask] Before mailing, please check the archive at www.jiscmail.ac.uk/lists/bugs.html Please do not mail attachments to the list. To leave the BUGS list, send LEAVE BUGS to [log in to unmask] If this fails, mail [log in to unmask], NOT the whole list