EIC (Expert Information Criterion) not AIC: the cautious biologist's guide to model selection

TL;DR

This paper reviews causal inference methods and proposes an analytical workflow tailored for ecology, evolution, and behavior research, emphasizing the importance of explicit model selection and causal analysis in observational data.

Contribution

It introduces a clear framework for applying causal inference in EEB research, bridging disciplinary gaps and enhancing data analysis practices.

Findings

Defines four analytical tasks: description, prediction, association, causal inference.

Provides a step-by-step workflow for causal inference with EEB examples.

Highlights the importance of explicit model selection in observational studies.

Abstract

1.A goal of many research programs in biology is to extract meaningful insights from large, complex data sets. Researchers in Ecology, Evolution and Behavior (EEB) often grapple with long-term, observational data sets from which they construct models to address fundamental questions about biology. Similarly, epidemiologists analyze large, complex observational data sets to understand the distribution and determinants of human health and disease. A key difference in the analytical workflows for these two distinct areas of biology is delineation of data analysis tasks and explicit use of causal inference methods, widely adopted by epidemiologists. 2.Here, we review the most recent causal inference literature and describe an analytical workflow that has direct applications for EEB researchers. 3.The first half of this commentary defines four distinct analytical tasks (description, prediction, association, and causal inference), and the corresponding approaches to data analysis and model selection. The latter half is dedicated to walking the reader through the steps of casual inference, focusing on examples from EEB. 4.Given increasing interest in causal inference and common misperceptions regarding the task of causal inference, we aim to facilitate an exchange of ideas between disciplinary silos and provide a framework for analyses of all data, though particularly relevant for observational data.