Ecology is a discipline that has faced increasing challenges as the disconnect between its scientific and statistical methods has become more obvious. Growing demands for useful forecasts in an era of intensifying global change requires models that can capture the variability and underlying uncertainty of ecological systems and related data. Yet many ecologists are not trained in current methods to build the flexible robust models needed to address this challenge. Thus, there is often a reliance on a limited set of predefined models combined with null hypothesis testing or a temptation to adopt new approaches without fully understanding their limitations. The result is poor models that lead to incorrect predictions, alongside concerns of a looming replication crisis. Here we show how new advances in workflows can lead to better models and enhance training in ecology. Building on the increasingly computational toolkit of many ecologists, this approach leverages simulation to integrate model building and testing of empirical data more fully with ecological theory. We argue this approach can fit models that are more robust and better-suited to providing new ecological insights and improved predictions, and may provide a blueprint for other fields similarly challenged by complex systems, growing datasets and limited training in how to best approach them. This article is part of the theme issue 'Statistical workflow'.
Wolkovich et al. (Thu,) studied this question.