What is Causation?
What is causation? – Scientists, philosophers, and statisticians have all written extensively on the topic of causation. As a mechanistic scientist, I align with a mechanistic view of causation, as does the National Academies of Sciences Consensus Report on Causal Methods (2022). From this perspective I would propose that causation itself can be thought of as the operation of some collection of spatio-temporally contiguous structures and processes capable of transferring state changes from one or more mechanistic element to another.
How Might We Study Causation?
(1) Statisticians have focused on the simplest approach, which is often described as “Difference Making”. For them, the randomized manipulative experiment is the most obvious way to study difference making. This is a very indirect, approximate, and limited approach to learning about underlying mechanisms. That said, it can be very useful as a practical approach, though it only leads to causal knowledge when supplemented by mechanistic knowledge.
(2) Scientists frequently engage in “Mechanistic causal investigation”. In doing so they view the manifestations observable in the world as being produced by some real-world structures and processes. Referring to those real-world structures and processes as “machinery” can be a useful even if not perfect analogy. “Causal knowledge analysis” represents a formal approach to mechanistic causal investigation. Scientists (and others) frequently engage in this practice informally. There are great advantages for us to adopt a more formal approach moving forward.
(3) Repeatable manifestations (e.g., as studied using empirical dynamic modeling) provide another source of evidence that can be informative, especially when there is accompanying mechanistic knowledge.
(4) Theoretical analyses can also provide causal insights when they can be tied to structures and processes in the real world.
Expectations
It is difficulet for me to imagine that a singular approach to building causal knowledge and understanding can be defended. As Grace et al. (2025a) state, “Scientists and statisticians have spent enough years arguing for the supremacy of various approaches for causal investigation and we should recognize that for the larger goal of building causal knowledge, many approaches and forms of evidence can contribute.”
Situations, Choices, & Decisions
It is not unusual that people will recommend an approach or protocol without recognizing that there can be a variety of study purposes, contexts, alternatives, constraints, and risk profiles. These will all come into play in causal investigations where there are complex decisions to be made. Not everyone is doing a drug trial or investigating a new organism. Recommendations for this important aspect of causal investigations is still under development. Smart Choices by Hammond et al. bring to our attention the criteria to consider.