Shared-catch simulation model

Paired counts that are both correlated and overdispersed are hard to model well. So, we built a model for exactly that: a negative-binomial trivariate reduction, reparameterised so its parameters represent quantities of interest that are intuitive and meaningful. See how it works by adjusting the dials in the interactive panel below.

Each longline set has a logbook (LB) count and an electronic monitoring (EM) count. Each count comprises fish seen by both methods (A = ) plus fish seen only by that method (LB-only = , EM-only = ). M is the mean count, and congruence f is the fraction of it that is shared. The box on the left shows the model parameters; the ten stacks are sets randomly generated from the model. Non-zero bias β distributes the unshared fish unevenly between the two methods, without changing M. With κ = 0 each part of the count is generated by a Poisson distribution; with κ > 0, by a negative binomial (higher variance).

Software

The panel above is a browser simulation of the model. Fitting it to data is done with bicountbrms, an R package of custom brms families for jointly modelling paired counts. It provides bipois() and binegbin() — Poisson and negative-binomial trivariate reductions — and a partial-observation variant of each, for when one of the two counts is missing on some sets. Its native rate and dispersion parameters reparameterise into the four coordinates used as dials above: overall count rate, congruence, bias, and excess dispersion.

A companion package, skellambrms, models the difference between the two counts directly rather than the pair jointly, with Skellam, discrete-Laplace and discrete-normal families. Each is available in a fixed-location form and a free-location form that estimates the systematic disagreement between the two sources.

Both packages are archived on Zenodo under concept DOIs, which resolve to the current version rather than to a fixed release.

Built by Sea Through Science as part of a project to evaluate electronic monitoring of Belize-flagged longline vessels, and how this method compares with logbook data — a project run in collaboration with JPEC Consulting for The Nature Conservancy.

Tuna silhouette: Thunnus albacares by T. Michael Keesey, PhyloPic, public domain.