Use of contradictory data sources in stock assessments

• Published in: Fisheries research Vol. 11; no. 3; pp. 225 - 238
• Main Author: Richards, Laura J.
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier B.V 1991; Elsevier
• Subjects: Animal, plant and microbial ecology; aplicaciones del ordenador; application des ordinateurs; Applied ecology; Biological and medical sciences; computer applications; datos sobre pesca; donnees sur les peches; Exploitation and management of natural biological resources (hunting, fishing and exploited populations survey, etc.); fishery data; fishery resources; fishes; Fundamental and applied biological sciences. Psychology; Marine; modele; modelos; Ophiodon elongatus; peces; poisson animal; recursos pesqueros; ressource halieutique; Canada; British Columbia; Pacific Ocean; North America; America; Canada, British Columbia, Vancouver I; Population exploitation; Marine environment; Vertebrata; Fishery; Estimation; Pisces; Mathematical model; Population density
• ISSN: 0165-7836; 1872-6763
• DOI: 10.1016/0165-7836(91)90003-X

Model predictions can vary greatly when different data sources are used for input. Here, I develop the likelihood function for a composite model based on data from a lingcod ( Ophiodon elongatus) stock off Vancouver Island, British Columbia, Canada. The composite model includes two components, a fishery and a tag model. The fishery model incorporates historical data on commercial catch, effort, and the size-frequency distribution of the catch. The tag model summarizes a mark-recapture experiment on the same stock. Both component models predict natural and fishing mortality rates. However, as the fishery and tag components lead to statistically distinct estimates of lingcod mortality, I demonstrate that a composite model is unwarranted for the lingcod data. In addition, I show how the use of likelihood weights in composite models can be misleading. Moderate values of a likelihood weight result in parameter estimates that are exterior to the confidence regions for either component model. Finally, I suggest guidelines for evaluating composite models.