Representative sampling by fishers in a reference fleet programme

Fisheries catch sampling is often done using non-probabilistic methods to tackle the operationally challenging sampling environment, but where the sampler must judge whether the sample is representative, there is a risk of intentional or unintentional biases being introduced. In this study, I investigate potential biases in sampling by fishers in the Norwegian Reference Fleet. Biases in haul selection were identified by modelling the effect of available covariates on sampling probability using mandatory catch logbooks as a census dataset. The model detected small biases in haul selection, favouring larger catches which are more representative of typical catch sizes in the fishery, and larger variations in sampling probability during daylight, which is partly driven by the availability of the sampler because of shift patterns. Biases in sub-sampling of fish was investigated by comparing length frequency distributions with monitoring, control, and surveillance inspections, revealing marginally larger proportions of undersized fish in Norwegian Reference Fleet sampling. In conclusion, there is no evidence of either convenience sampling or intentional manipulation of sampling. Rather, deviations from random sampling were small and isolated to individual vessels, suggesting that targeted feedback would be effective for improving future sampling. • Fishers tended to sample hauls of typical catch sizes for that vessel. • Diurnal variations in haul sampling probability are not significant. • Fishers sub-sampled slightly larger proportions of small fish than inspectors. • No evidence of either convenience sampling or intentional manipulation of sampling. • Small deviations from random sampling isolated to individual vessels.