Counterfactual Reasoning¶

Why It Matters¶

We can't directly observe causation - we only observe what actually happened. Causal reasoning requires imagining what would have happened otherwise. Without this, we can't distinguish: genuine effects from coincidence, necessary causes from sufficient causes, or effective interventions from placebos. Counterfactual reasoning is also how we learn from mistakes and near-misses: what would have happened if we'd chosen differently?

The Key Move¶

For any causal claim "X caused Y," make explicit the counterfactual: "If X hadn't occurred, Y wouldn't have occurred (or would have occurred differently)." Then test: How confident are we in this counterfactual? What's our evidence? For interventions, ask: "What's the control group?" - the counterfactual of what happens without the intervention. Make counterfactual comparisons explicit rather than implicit.

Classic Application¶

Historical analysis. "Did Churchill's leadership cause Britain to survive WWII?" requires imagining: what would have happened with a different leader? We can't rerun history, but we can look at: decisions unique to Churchill, moments where different choices were available, and outcomes that changed after his specific actions. The counterfactual becomes testable through comparing actual timeline to plausible alternatives.

Surprising Application¶

Personal performance review. "This project succeeded because of my contribution" requires the counterfactual: "Without my contribution, the project would have failed or succeeded less." Often reveals that you're taking credit for things that would have happened anyway (team was competent, problem was easy) or blaming yourself for failures outside your control (impossible deadline, no resources).

Failure Modes¶

Imagining unrealistic counterfactuals (comparing actual messy reality to idealized alternative). Neglecting multiple counterfactuals (many alternative scenarios exist, not just one). Hindsight bias - what seems obviously better in retrospect wasn't obvious prospectively. Treating closest-possible-world counterfactuals as if they're what would "really" have happened (we don't know). Counterfactual overdetermination - Y would have happened even without X through alternative cause Z.

Go Deeper¶

Lewis, "Counterfactuals"; Pearl, "Causality" (structural causal models)