TB: On Causation

Core Thesis

Association and causation are fundamentally different epistemic objects: association is about prediction (what co-occurs), causation is about intervention (what would happen if we acted). Neyman's counterfactual framework makes this precise — causal inference is the problem of reasoning about the unobserved.

Key Takeaways

The core distinction

  • "Association is about prediction. Causation is about intervention."
  • Two questions that look similar but are logically independent: "Does X predict Y?" vs. "If I do X, will Y happen?"
  • A variable can be associated with an outcome without causing it (confounding), and a cause can be hard to detect via association (weak signal, measurement error).

The Neyman counterfactual frame

  • Neyman framed causal inference in terms of measuring the unobserved: the effect of a treatment that was not performed.
  • "Causal inference, in this view, is the problem of inferring something (the effect of the treatment that was not performed) about unobserved counterfactuals from observed data."
  • The fundamental problem of causal inference: for any given unit, you can only observe one potential outcome — the one that actually occurred.

Mental Models

See also

Source

Substack, 2026 Original article