Extended molt phenology models improve inferences about molt duration and timing

TL;DR

This paper introduces an extended statistical modeling framework for molt phenology that accounts for real-world data complexities, improving inference accuracy for molt duration and timing in birds and mammals.

Contribution

The authors develop a novel, flexible modeling framework that handles re-encounters, misclassification, and sampling bias in molt data, enhancing analysis of real-world datasets.

Findings

Extended models reduce bias in molt timing estimates.

Framework improves inference accuracy across various sampling conditions.

Applicable to other phenological processes with ordered categories.

Abstract

Molt is an essential life-history event in birds and many mammals, as maintenance of feathers and fur is critical for survival. Despite this molt remains an understudied life-history event. Non-standard statistical techniques are required to estimate the phenology of molt from observations of plumage or pelage state, and existing molt phenology models have strict sampling requirements which can be difficult to meet under real-world conditions. We present an extended modelling framework that can accommodate features of real-world molt datasets, such re-encounters of individuals, misclassified molt states, and/or molt state-dependent sampling bias. We demonstrate that such features can lead to biased inferences when using existing molt phenology models, and show that our model extensions can improve inferences about molt phenology under a wide range of sampling conditions. We hope that our novel modelling framework removes barriers for modelling molt phenology data from realworld datasets and thereby further facilitates the uptake of appropriate statistical methods for such data. Although we focus on molt, the modelling framework is applicable to other phenological processes which can be recorded using either ordered categories or approximately linear progress scores.