Capturing spatiotemporal dynamics of Alaskan groundfish catch using signed-rank estimation for varying coefficient models

Varying coefficient models (VCMs) are commonly used for their high degree of flexibility in modeling complex systems. Many applications in fisheries utilize VCMs to capture spatial variation in populations of marine fishes. All of these applications use the penalized least squares method for estimation. However, this approach is known to be sensitive to non-normal distributions and outliers, a common feature of ecological data. Robust estimation methods are more appropriate for handling noisy and non-normal data. We present the application of a signed-rank-based procedure for obtaining robust estimates in VCMs on a fisheries dataset from the North Pacific Ocean. We demonstrates that the signed-rank-based estimation method provides better fit and improved prediction in comparison to the classical likelihood VCM fits in both simulations and the real data application, particularly when the distributions are non-normal and may be misspecified. Rank-based estimation of VCMs is therefore valuable for modeling ecological data and obtaining useful inferences where non-normality and outliers are common.