Comparing Bayesian and frequentist methods of fisheries models: Hierarchical catch curves

Bayesian inference is an emerging statistical paradigm and is becoming an increasingly used alternative to frequentist inference. Unfortunately, little is known about the efficacy of Bayesian inference and how it relates to the historical methodology of evaluating fisheries related models. Mortality information is routinely used in fisheries management to describe fish population abundance over time and has been historically estimated using catch curves and frequentist inference (i.e., maximum likelihood estimation). The objective of this study was to compare frequentist and Bayesian inference approaches to estimate instantaneous mortality (Z) from a hierarchical catch curve model. The data used in the comparison were from a long term monitoring program of yellow perch Perca flavescens from southern Lake Michigan in addition to a simulated dataset where parameter estimates were compared to known values. Point estimates of Z were similar among both methods. Similarly, Bayesian inference 95% credible intervals were concordant with frequentist 95% confidence intervals. However, the root mean squared error of frequentist inference increased at a higher rate than Bayesian inference with increasing variability in the simulated dataset. Our study builds on the literature that seeks to compare results between these two paradigms to assist managers to make the best decision possible when deciding what statistical paradigm to employ.