Simulation analysis using stock synthesis

1
Simulation analysis using stock synthesis

• Maunder, Piner, and Lee

2
Objective

• Use the SS2/SS3 program as a simulator to test
  the ability to estimate key model parameters

3
SS2 bootstrap

• Built in bootstrap procedure to generate
  artificial data files
• Data files have randomly generated data based on
  the same characteristics of the real data
• Data file in the SS2 format and can be used
  directly to run the model
• Data files are in a single file one on top of the
  other and need to be separated
• Use data files to run bootstraps, and estimate
  uncertainty and bias
• Can also be used to generate data based on fixed
  parameter values
• (note in SS3 first data set is the actual and
  second is with no error)

4
Simulation algorithm for estimating M(based on
Ian Stewarts R code)

• Run the original model to generate the bootstrap
  files (M fixed)
• Set the number of bootstrap files to 500 in the
  starter.ss file
• Split out the data files into separate files, put
  them in separate directories with the model files
• Set the bootstrap files to 0
• Set M to a positive phase
• Set it to start from par file in the starter.ss
  file to make it run faster
• Run the assessment in each directory
• Do this by creating a batch file that moves
  through the directories and runs the model
• Read in the results from each run

5
Show R code
6
Methods

• Use current peer reviewed SS2 assessments
• Use the exact setup used in the assessment,
  including model assumptions, parameters
  estimated, and data
• Estimate any Ms that are different (e.g.
  old/young, male/female)

7
Species

• Arrowtooth flounder
• Black rockfish-North
• Black rockfish-South
• Chilly pepper rockfish
• Pacific cod
• Cowcod
• Dark blotch rockfish
• English sole
• Hake
• Longnose skate
• Blue rockfish
• Sablefish
• Yelloweye rockfish

8
Good estimates
9
Biased estimates
10
Estimating male and female
11
Estimating old and young
12
Indication of model miss-specification
13
Indication of statistical assumption violation
(incorrect sample size or residual patterns)
14
Modification for estimating steepness of the
stock recruitment relationship

• Run the original model
• Set the maximum phase to 0 in the starter file
  and use the par file (i.e. simulate data from
  fixed parameter values)
• Replace the recruitment deviates in the par file
  with a N(0,Rsd)
• Run one bootstrap
• Estimate the model parameters
• Repeat 3-5 500 times
• Summarize results

15
Results for Pacific cod

• Estimates of steepness equal to one
• Even if h0.5
• Only gets estimates lt 1 if h0.5 and Rsd 0.2
• Need to try on the other stocks

16
Other examples

• Estimating the growth parameters
• Does iterative reweighting improve the results
• Can observation error adjust for process error
• MSE

17
Simon, talk about Condor
18
Acknowledgements

• Stacey Millar and all the assessment authors