Predictive models are indeed useful for causal inference

Area of Science:

• Ecology
• Causal Inference
• Ecological Modeling

Background:

• Recent discussions in ecology highlight causal inference, particularly concerning structural causal models (SCM).
• SCM proponents have questioned the utility of predictive models for inferring causation.
• This study addresses the debate on using predictive modeling for causal inference in ecology.

Purpose of the Study:

• To argue for the validity of predictive modeling in assessing ecological causation.
• To distinguish between hypothesis-generating and hypothesis-testing predictive modeling approaches.
• To propose a hypothetico-deductive (H-D) framework for causal inference in ecology.

Main Methods:

• Defining causation with a focus on "probability-raisers-of-processes" suitable for ecological systems.
• Outlining scientific designs for generating observational data for causal investigations.
• Comparing components of SCM and H-D approaches, emphasizing H-D for vital rates.

Main Results:

• Predictive modeling, when guided by causal hypotheses within an H-D framework, is a valid method for causal inference.
• A distinction is made between undirected predictive modeling (hypothesis generation) and H-D predictive modeling (causal inference).
• Two ecological case studies demonstrate the successful application of predictive modeling for causal inference, addressing SCM criticisms.

Conclusions:

• Predictive models, particularly within a hypothetico-deductive framework, are essential tools for drawing causal inferences in ecology.
• The "probability-raisers-of-processes" definition of causation is well-suited for complex ecological systems.
• Predictive modeling continues to offer valuable insights into ecological causation when appropriately applied.