|Document Type:||Journal Article|
|Title:||Standardizing compositional data for stock assessment|
|Author:||James T. Thorson|
|Journal:||ICES Journal of Marine Science|
|Keywords:||composition-standardization models, sampling intensity, strata, Dirichlet-multinomial, down-weighting, integrated model, stock assessment,|
Population dynamics and stock assessment models frequently integrate abundance index and compositional (e.g., age, length, sex) data. Abundance indices are generally estimated from survey data using index standardization models, which provides estimates of standard errors for the abundance index while dealing with three confounding factors: (1) differences in sampling intensity spatially or over time; (2) non-independence of available data; and (3) the effect of covariates. However, compositional data are not generally processed using a standardization model, so effective sample size is not routinely estimated and and these three issues are unresolved. To rectify this absence, I propose a computationally simple ‘normal approximation’ method for standardizing compositional data, and compare this with design-based and Dirichel-multinomial methods for analyzing compositional data. Using simulated data from a population with multiple strata, heterogeneity within strata, differences in sampling intensity, and additional overdispersion, I show that the normal-approximation method provides unbiased estimates of abundance at age as well as estimates of effective sample size that are consistent with the imprecision of these estimates. The Dirichet-multinomial, by contrast, fails to account for known differences in sampling intensity, and hence provides biased estimates when sampling intensity is correlated with variation in abundance-at-age data. We end by discussing uses for ‘composition-standardization models’ and propose that future research develop methods to impute compositional data in strata with missing data.
|Theme:||Recovery, Rebuilding and Sustainability of Marine and Anadromous Species|
Describe the relationship among human activities and species stock status, recovery, rebuilding and sustainability.
Develop methods to use physiological and biological information to predict population-level processes.