Data conflict caused by model mis-specification of selectivity in an integrated stock assessment model and its potential effects on stock status estimation

Abstract

Integrated analysis is widely used for fisheries stock assessment. The analysis has several merits, including its prevention of the loss of information and its enabling of the evaluation of uncertainties that arise when integrating individual data, such as size composition data and abundance indices. However, the data integration itself causes one of the most difficult problems: conflicts among data sources. This paper provides an illustrative explanation of how data conflict can occur under model mis-specification on selectivity in a multi-fishery integrated stock assessment model through simulation with known parameters. We demonstrated that estimation bias can be caused under various scenarios of model mis-specifications on selectivity. Relative likelihood profiles decomposed into the likelihood components from different data sources revealed data conflicts among multiple datasets under the model mis-specifications, whereas the relative likelihood profiles displayed no severe data conflicts without any model mis-specifications. Data conflicts occurred because a single estimation model without correct selectivity assumptions cannot explain the observed multiple data sources simultaneously. We also found that almost all of the parameters estimated in the integrated model were potentially affected by the model mis-specification in any single component of selectivity for a specific fishery. If we accumulate empirical knowledge on the general patterns of conflicts and what types of data sources are robust against typical model mis-specifications, the relative likelihood profiles could be utilized more efficiently in diagnostics to identify potential model mis-specifications and to obtain more accurate estimates of the quantities of interest for stock management.