Incorporating survey variance in sequential population analysis

No abstracts are to be cited without prior reference to the author.

We derive some basic statistics that describe the variability of a survey index derived from stratified random sampling for several Northwest Atlantic fish stocks. The variability is expressed as a function of population abundance and is based on a Negative Binomial (NB) distribution assumption for trawl catches. Diagnostics that support this assumption are presented. However, maximum likelihood estimates of the NB over-dispersion parameter based on a stratumeffects model can have severe bias, and an alternative estimator is shown to give much better results. We also show how the survey variance component can be incorporated into stock assessment models like ADAPT or XSA. Interestingly, this results in an estimation procedure that is more similar to the implicit and intuitive weighting that many fisheries scientists use to track cohorts in survey data by focusing on the ages that tend to be caught well, whereas ADAPT or XSA tend to give higher weight to ages not well caught.