Predictive models aren't for causal inference. (7th June 2022)

• Record Type: Journal Article
• Authors: Arif, Suchinta; MacNeil, M. Aaron
• Is Part Of: Ecology letters. Volume 25:Number 8(2022)
• Page Start: 1741
• Page End: 1745
• Publication Date: 2022-06-07
• Subjects: back‐door criterion -- causal inference -- directed acyclic graphs (DAGs) -- model selection -- prediction; Ecology -- Periodicals; 577
• DOI: 10.1111/ele.14033
• Languages: English
• ISSNs: 1461-023X

Abstract

Abstract: Ecologists often rely on observational data to understand causal relationships. Although observational causal inference methodologies exist, predictive techniques such as model selection based on information criterion (e.g. AIC) remains a common approach used to understand ecological relationships. However, predictive approaches are not appropriate for drawing causal conclusions. Here, we highlight the distinction between predictive and causal inference and show how predictive techniques can lead to biased causal estimates. Instead, we encourage ecologists to valid causal inference methods such as the backdoor criterion, a graphical rule that can be used to determine causal relationships across observational studies. Abstract : Although observational causal inferences methodologies exist, they have not been widely adopted in the ecological literature. Instead, many studies misuse predictive techniques to quantify causal ecological relationships. Here, we highlight the distinction between predictive and causal inference, and highlight the backdoor criterion as a graphical rule that can be widely used for observational causal inference in ecology.