Integration and Simulation of Meta-Analysis Estimates of Steepness for West Coast Groundfish Stocks

Steepness estimation is a classic estimation problem in fisheries management. Biology and meta-analysis studies have shown there is undoubtedly a stock recruitment relationship for a fish stock. But uncertain observational data, no controlled experiments, environmental factors, estimation error, difficulty collecting data on survival and fecundity, among other things have rendered it difficult to estimate steepness at a high precision. This presents a cause for concern since steepness directly affects the stock’s productivity and recruitment at low stock sizes. Meta-analysis methods have been presented as a way to constrain this uncertainty.

However, meta-analysis methods are not perfect. The added benefit of higher precision is compensated by bias due to shrinkage towards the mass. The classic Stein’s estimator will uniformly outperform maximum likelihood estimates at an aggregate, but individually it will perform worse. An example of one assumption (exchangeability) being violated is when stock assessments for the West Coast groundfish fix steepness values for stocks that lack data contrast (synonymously, have high steepness) to estimate the latent series of stock and recruitment. 

Two methods, the Bayesian hierarchical method and an empirical likelihood method will compete against each other in an array of simulations. The latter of which flexibly maneuvers its way around.