Résumé
Most of the big questions that keep ecologists busy are causal in nature, be it to understand and forecast the effects of global warming on tree physiology or forest dynamics, the consequences of alien species invasions on native communities, or eco-evolutionary constraints in species traits and adaptations, to name a few. Whereas randomized controlled experiments are the gold standard to quantify causal effects, these are most often impractical or even impossible, for many ecological questions such as the ones above. The Structural Causal Model (SCM) framework however offers ways to still address such questions from observational data, such as the ones collected from natural ecosystems.Whereas other fields such as the Social Sciences, Epidemiology or Psychological Sciences have long tended to move towards explicitly causal approaches of their questions, Ecology has tended to lag behind on that aspect, with common problematic practices hindering correct inference still common place, notably through the paired use of prediction-oriented model selection and causal interpretation of the resulting variable coefficients.Explicitly acknowledging the causal nature of the questions we ask, and using appropriate quantitative tools to address these is a necessary step to avoid many avoidable issues, from the step of data collection to that of data analysis.In this talk, I aim to introduce some of the basic concepts needed to ask causal questions using the SCM framework, such as the use of causal diagrams to clearly represent our assumptions about the data-generating processes, the “elemental confounds” and how they can prevent or cause spurious associations among variables of interest, and the “backdoor criterion” to help choose what covariate to control for in regression-like models, in order to address a given causal effect.