Code from: Enhancing data-limited assessments with random effects: A case study on Korea chub mackerel (Scomber japonicus)

In a state-space framework, temporal variations in fishery-dependent
processes can be modeled as random effects. This modeling flexibility
makes state-space models (SSMs) powerful tools for data-limited
assessments. Though SSMs enable the model-based inference of the
unobserved processes, their flexibility can lead to overfitting and
non-identifiability issues. To address these challenges, we developed a
suite of state-space length-based age-structured models and applied them
to the Korean chub mackerel (Scomber japonicus) stock. Our research
demonstrated that incorporating temporal variations in fishery-dependent
processes can rectify model mis-specification but may compromise
robustness, which can be diagnosed through a series of model checking
processes. To tackle non-identifiability, we used a non- degenerate
estimator, implementing a gamma distribution as a penalty for the standard
deviation parameters of observation errors. This penalty function enabled
the simultaneous estimation of both process and observation error
variances with minimal bias, a notably challenging task in SSMs. These
results highlight the importance of model checking and the effectiveness
of the penalized approach in estimating SSMs. Additionally, we discussed
novel assessment outcomes for the mackerel stock.