call write_data from within mceval

[k-doering-NOAA]

Imported from redmine, Issue #78529 Opened by @RickMethot on 2020-05-12 Status when imported: New

A suggestion from Teresa Amar is copied below: I have a question about SS and the MCMC process. In the ADMB mceval phase, can the expected values for the data be output, similar to how the bootstrap functionality works? That seems like it could be a straight-forward way to include (some) parameter uncertainty through the "true" data in the projections, although then one would need to do projections from 100 or more simulated populations for the subsampled MCMC parameter sets.

then: For the Gulf of Alaska walleye pollock MSE, we use the ADMB MCMC functionality in the operating model for the projections. After the operating model was conditioned, we ran the MCMC and saved 101 parameter vectors, 1 for the MLE, and 100 from the MCMC chain. https://github.com/amart/mse

Using the ADMB mceval_phase() functionality: https://github.com/amart/mse/blob/master/opmodel.tpl#L2214 At the time I had the idea of using the MCMC simulated population-specific expected historical data so that there would not be a discontinuity between the actual historical data and the data generated over the projection period for that population. However, we did not do this. I think that you could do this in Stock Synthesis, since a lot of this functionality already exists. I looked into doing this with SS in 2009 or 2010, but I didn't implement it. This functionality would give users another option for generating historical data with the bootstrap option in the starter file. For each MCMC simulated population, SS could output the expected data values given a MCMC parameter vector.