Probabilistic Causation

mental-model

Probabilistic causation changes the likelihood of an outcome rather than determining each case. Long delays and ambiguous individual cases can conceal a strong causal pattern that becomes visible only in aggregate.

A smoker develops cancer 20 years after the exposure—and that case alone still cannot prove smoking caused it. The causal signal exists, but not at the level of a single, neatly attributable outcome.

E1

Causes that load the dice

Some causes work by shifting a probability distribution: exposure raises the odds of harm without guaranteeing who will suffer it. Each individual outcome therefore remains compatible with several explanations. Add a latency of 10 or 20 years, and the link becomes even easier to deny because cause and effect are separated in time.

The pattern emerges by comparing aggregates—how outcomes differ across exposed and less-exposed populations—not by demanding that every case carry an identifiable causal signature. This is why probabilistic causation must be distinguished from mere correlation while still resisting the impossible standard that a cause must determine every outcome.

E1

Where it shows up

Smoking and delayed cancer

No single cancer diagnosis settles the causal question. Across many cases, however, the altered odds become visible even though disease appears years after exposure and not every smoker develops it.

E1

A pattern is not an individual verdict

An aggregate causal effect does not establish that the cause produced one particular case, nor does it mean every exposed person will experience the outcome. The model supports claims about changed odds; it cannot supply certainty where the mechanism itself is probabilistic.

E1

Replace anecdotes with rate comparisons

When evaluating a delayed harm, specify the exposed and comparison groups, the outcome rate in each, and the plausible latency window. Do not let one unaffected person—or one ambiguous case—stand in for the population-level test the claim requires.

E1

Episodes that teach this