DOES CAUSE PRECEDE EFFECT?
The arrow of causation demands temporal precedence — an effect cannot birth its own cause. Yet in quantum entanglement, correlated outcomes emerge without a propagating signal. Does causality require a timeline, or is it a pattern recognition imposed by observers trapped in sequential experience?
We have no direct impression of necessary connection between events — only constant conjunction and the expectation it breeds.
Causal diagrams formalize do-calculus: intervention, not observation, reveals genuine cause from confounded correlation.
CORRELATION IS NOT CAUSATION
The mantra every statistician chants — yet correlation remains the gateway drug to causal inference. Randomized controlled trials isolate variables, but in messy social systems, we build causal models from observational data saturated with confounders. The question is not whether correlation implies causation, but under what structural assumptions it can.
Aggregated data reverses direction when stratified — the paradox that makes naive causal inference from correlation treacherous.
Temporal predictive power as a proxy for causation: if X forecasts Y beyond Y's own history, X "Granger-causes" Y.
THE HIDDEN MECHANISM
To claim A causes B, must we identify the mechanism linking them? Mechanistic philosophy demands a chain of intermediate events — a domino sequence from cause to effect. But epidemiologists established that smoking causes cancer decades before molecular pathways were understood. Causation can be proven without mechanism.
Causal processes transmit a conserved quantity through spacetime — marking a process to trace the physical transfer of influence.
Mechanisms are entities and activities organized to produce change — causation understood through the nuts and bolts of nature.
THE COUNTERFACTUAL TEST
If the cause had not occurred, would the effect still have happened? The counterfactual theory of causation reduces causal claims to conditional statements about possible worlds. But overdetermination shatters this framework — when two independent causes each suffice, neither passes the "but-for" test, yet both intuitively caused the outcome.
Causation is a chain of counterfactual dependence — event C caused E if the nearest possible world without C is a world without E.
Interventionist causation: C causes E if there exists an intervention on C that changes E, holding other variables fixed.
CAUSALITY REMAINS OPEN
After centuries of philosophical debate and decades of formal causal modeling, causality resists a unified theory. It is simultaneously a metaphysical primitive, a statistical inference, a counterfactual conditional, and an interventionist construct. The debate is the destination. Every argument card you have tilted, every collision zone you have witnessed — these are the living edges of an unsolved problem.
Causation is pluralist — no single account captures all causal relations. Different domains demand different causal concepts.
Causal discovery algorithms extract structure from data, but require assumptions about faithfulness and sufficiency that nature may not honor.
The debate continues. Every cause contains the seed of its next question.