A simulation approach to assessing bias in a fisheries self-sampling programme

The hierarchical structure and non-probabilistic sampling in fisher self-sampling programmes makes it difficult to evaluate biases in total catch estimates. While so, it is possible to evaluate bias in the reported component of catches, which can then be used to infer likely bias in total catches. We assessed bias in the reported component of catches for 18 species in the Barents Sea trawl and longline fisheries by simulating 2000 realizations of the Norwegian Reference Fleet sampling programme using the mandatory catch reporting system, then for each realization we estimated fleet-wide catches using simple design-based estimators and quantified bias. We then inserted variations (e.g. simple random and systematic sampling) at different levels of the sampling design (sampling frame, vessel, and operation) to identify important factors and trends affecting bias in reported catches. We found that whilst current sampling procedures for fishing operations were not biased, non-probabilistic vessel sampling resulted in bias for some species. However, we concluded this was typically within the bounds of expected variation from probabilistic sampling. Our results highlight the risk of applying these simple estimators to all species. We recommend that future estimates of total catches consider alternative estimators and more conservative estimates of uncertainty where necessary.