Contemporary stock assessments include multiple sources of data, with applications illustrating that management quantities of concern (stock abundance, spawning stock biomass, recruitment, etc.) are often highly sensitive to ‘weighting’ protocols involved in statistical fitting of the data sets within the overall model. The weights given to data sets can be subjective and determined via ad hoc decisions (e.g., final lambdas), based on classical sampling error assumptions outside the model or estimated within the assessment model itself. The functional form (probability distribution) of the likelihood component, including robustification, can also influence final results and conclusions. Rigorous statistical methods are available to assist in determining appropriate weights for particular data sets, which ultimately, will allow inconsistencies in the integrated model to be better evaluated. However, these approaches are generally based on determining the effect of observation error solely, rather than identifying the magnitude of both process and observation errors in this estimation process.
The Center for the Advancement of Population Assessment Methodology (CAPAM) will host a technical workshop on Data conflict and weighting, likelihood functions, and process error in La Jolla, CA, USA, October 19-23, 2015. Please visit the workshop page here. Further details regarding registration, agenda and related information will be available in the future.