Course

Introduction to the Theory and Practice of Causal Inference in Ecology

Organizer(s):

Most major questions in Ecology are causal. Randomised controlled experiments—the gold standard for causality—are often impractical or impossible for complex issues, such as understanding global warming’s influence on forests or the consequences of alien species invasions. The Structural Causal Modelling (SCM) framework provides tools to address cause-effect relationships using observational data from natural ecosystems. Importantly, this framework is also central to experimental studies, where statistical biases can still undermine inference despite good experimental design.

While Social and Psychological Sciences have long embraced explicit causal approaches, Ecology has lagged behind, with common practices introducing biases. Using appropriate quantitative tools to address causality is essential to avoid pitfalls, from data collection to analysis.

This course introduces key concepts for thinking causally using the SCM framework. We will present causal diagrams to represent assumptions about data-generating processes, distinguish causal from spurious associations, and apply graphical rules to identify ‘control variables’ for causal interpretation in statistical models. A hands-on session with a toy dataset will guide participants in using dagitty and R to create a causal diagram and quantify causal effects.

Target audience: anyone working with questions of causal nature, as much early career as senior researcher.
Pre-requisites: Participants should be familiar with multiple linear regressions, and have some experience using R (e.g. writing simple code, running a linear regression using base R lm function).
While the SCM framework can be used with any type of statistical - Bayesian or frequentist - or non-statistical modeling approaches, we will only use simple linear regressions in this short course, in order to focus on the causal thinking part and to keep statistical considerations to a minimum.

By the end of the course, participants will:



Recognize main forms of causal and spurious associations.


Learn to create a causal diagram.


Use it to determine control variables for a statistical model, ensuring causal interpretation.



In a nutshell, participants will learn to create a causal model, and use it to carefully guide the construction of a statistical model aimed at causal inference. Be it from experimental, or observational data.