Recruitment Bias Adjustment

Forums

RSS (Opens New Window)

Laura Lee, modified 4 Years ago.

Recruitment Bias Adjustment

Youngling Posts: 95 Join Date: 3/31/16 Recent Posts

Hey all,

Sorry for all the questions of late...feeling like I am hitting a brick wall after every turn.

So, my latest model run is producing the following warning:

Main recdev biasadj is >2 times ratio of rmse to sigmaR

This warning appears both before and after adjusting the years chosen for bias adjustment and the maximum bias adjustment parameter (adjusted using values from output of r4ss based on Methot and Taylor 2011). In my base run, sigmaR is fixed at 0.74. Out of curiosity, I allowed the model to estimate sigmaR and the estimated value was 0.073, which suggests to me there may not be much recruitment variability information in the data. When I do let the model estimate sigmaR, I don't receive the warning shown above. Alternatively, with the fixed sigmaR, I ran the model in a series of runs by decreasing the maximum bias adjustment parameter until I no longer received the warning and had to reduce down to 0.1 before the warning disappeared.

I would appreciate any recommendations on how to proceed. I was also wondering if I should be adjusting the lambdas for the recruitment deviations (I thought I saw that suggested somewhere)?

Thanks again for all your time and patience!

Cheers,

Laura

Chantel Wetzel, modified 4 Years ago.

RE: Recruitment Bias Adjustment

Youngling Posts: 32 Join Date: 10/7/15 Recent Posts

Hi Laura,

This is a good question and there are some useful resources in r4ss to assist with this problem. As the warning states the input RMSE to sigmaR ratio is out of step with the bias adjustment. While it is generally not recommended to estimate the sigmaR parameter within SS, your estimated runs are hinting at the solution to this problem. The SS_output function within r4ss prints some useful output to the R terminal window that allows the user to diagnose if there input sigmaR aligns with the estimated variance of the recruitment deviations. If you look at the $sigma_R_info object in the list there is column called "alternative_sigma_R". This column shows an alternative sigmaR value that may wan to be used to better align with the estimated variance of the recruitment deviations. I ran the model you sent yesterday and after a couple of run iterations the "alternative_sigma_R" value was much lower than the initial 0.74 value indicating that the model is estimated very little variation in the recruitment deviations by year. If you update the fixed sigmaR value based on the "alternative_sigma_R" this should eliminate this waring.

Also, given that very little recruitment variation is being estimated you may want to explore a deterministic model as a sensitivity to understand if the additional recruitment deviation parameters are providing a meaningful improved fit to the data resulting in changes in the derived quantities (spawning biomass, fraction unfished).

Chantel

Laura Lee, modified 4 Years ago.

RE: Recruitment Bias Adjustment

Youngling Posts: 95 Join Date: 3/31/16 Recent Posts

Thanks so much, Chantel!