Probing past precision: Assessing bias in observer coverage

At-sea observers provide the most valuable source of information on discarded bycatch in commercial fisheries. At low observer coverage levels, fleets may be able to afford to change their fishing behaviour when an observer is on board (“observer effect”) to minimize bycatch of species or size classes that may draw negative attention from fisheries managers or environmental organizations. The resulting bias in the observer data may lead to underestimation of annual bycatch for the fishery. Assessment and quantification of bias in observer data are key elements to robust bycatch estimation for a fishery, but are rarely implemented. I will present the results of a multi-tiered approach to evaluating whether observer coverage in the Canadian pelagic longline fishery for swordfish and tuna is representative of the fishery as a whole. Univariate analyses can detect bias in observer coverage with respect to individual variables of interest. If sampling is representative at the resolution of these variables, but not on an unresolved scale or factor, more complex approaches, such as a multivariate analysis of catch per unit effort in observed versus unobserved trips, are required to assess bias.