Full length articleGet the biology right, or use size-composition data at your own risk

- Recruitment variation is needed to interpret abundance information from indices.
- Size-composition data contains recruitment information and cannot be ignored.
- These data can also wrongly influence estimates of absolute abundance.
- This influence is minimized when we “get the biology right” in our model assumptions.

Weighting of size-composition data (length or weight composition of the catches) can have a large influence on the results of contemporary integrated stock assessment models in the presence of model misspecification. Model misspecification leads to conflicting information among data sets, and the choice of data weighting will determine the results. Information content on absolute abundance and abundance trends contained in size-composition data is particularly susceptible to misspecification of the biological processes. Biological processes are often misspecified in assessment models for exploited fish stocks due to lack of information. The misspecification can be in a functional form (e.g., the growth curve) or in the values assumed for pre-specified parameters. Our application to bigeye tuna in the eastern Pacific Ocean shows how one needs to “get the biology right”, i.e. minimize model misspecification, to reduce the dependency of stock assessment results on the weighting of the various data components. The stock assessment results are sensitive to the conversion from processed weight to total weight, a common, but often overlooked, component of model specification, and to the asymptotic length of the growth curve. The results are also sensitive to the weighting of the composition data. Application of the Age-Structured Production Model diagnostic shows that recruitment variation must be taken into account to interpret the absolute abundance and trend information contained in a CPUE-based index of relative abundance. Unfortunately, recruitment cannot typically be estimated from the relative index of abundance alone, so composition data are needed. The abundance estimates from an age-structured production model with estimated recruitment deviates are too uncertain (i.e., have wide confidence intervals) to be of use for management advice. Therefore, there is a trade-off between using composition data to estimate recruitment and its influence on estimates of absolute abundance through a catch-curve type process. We conclude that (i) integrated analysis, the current approach for assessing fish stocks, is supported by our results; (ii) composition data are needed to estimate recruitment; and (iii) addressing key model misspecifications should be a major component of integrated analysis.