Conditional Probability and Independence Explained

Defining conditional probability

Conditional probability answers a very specific question: given that we already know event B happened, how likely is event A? It is written P(A | B), read "the probability of A given B", and defined as

P(A | B) = P(A ∩ B) / P(B), provided P(B) > 0.

The intuition is a shrinking sample space. Once you know B has occurred, every outcome outside B becomes impossible, so B effectively becomes the new universe. The probability of A given B is then just the fraction of that new, smaller universe that A still occupies — the overlap A ∩ B, rescaled so that B itself has probability 1. This rescaling is why dividing by P(B) is necessary rather than optional: without it, the numbers would no longer sum to 1 across the new reduced sample space.

The multiplication rule and probability trees

Rearranging the definition gives the multiplication rule: P(A ∩ B) = P(A | B) · P(B) = P(B | A) · P(A). This is the engine behind probability trees for sequential experiments. Draw a first-stage branch for each outcome of the first event, label it with its unconditional probability, then draw second-stage branches labelled with the probability of the second event given the first. Multiplying along any complete path from root to leaf gives the joint probability of that specific sequence of outcomes, and because the tree accounts for every possible path, the leaf probabilities always sum to 1.

This extends naturally to more than two stages via the chain rule: P(A₁ ∩ A₂ ∩ ... ∩ A_n) = P(A₁) · P(A₂ | A₁) · P(A₃ | A₁ ∩ A₂) · ... · P(A_n | A₁ ∩ ... ∩ A_n-1). Drawing cards without replacement is the classic example: the probability of drawing three aces in a row from a standard deck is (4/52) · (3/51) · (2/50), because each draw changes the composition of the remaining deck and therefore the conditional probability of the next event.

Independence: a strong assumption in disguise

Events A and B are independent if knowing one occurred tells you nothing about the other: formally, P(A | B) = P(A), which is algebraically equivalent to the cleaner, symmetric condition P(A ∩ B) = P(A) · P(B). This second form is the one almost always used in practice because it works even when P(B) = 0, where the conditional definition would divide by zero.

Independence is a much stronger claim than people intuitively treat it as. Two successive fair coin flips are independent because the coin has no memory. But most real-world "coincidences" people call independent are not: whether it rains today and whether it rains tomorrow are correlated through weather systems; two students' exam scores are correlated through shared teaching quality; two stocks' daily returns are correlated through shared market and sector exposure. Treating correlated events as independent is one of the most consequential modelling errors in finance and risk management — it was a central technical failure behind the mispricing of mortgage-backed securities before the 2008 financial crisis, where default probabilities across different mortgages were modelled as far more independent than they actually were.

Pairwise independence is not the same as mutual independence

A subtlety that trips up even careful students: three or more events can be pairwise independent (every pair satisfies P(A_i ∩ A_j) = P(A_i)P(A_j)) without being mutually independent (which additionally requires P(A ∩ B ∩ C) = P(A)P(B)P(C), and the analogous product condition for every subset). A standard textbook example: flip two fair coins, let A be "first coin is heads", B be "second coin is heads", and C be "the two coins show the same result". Each pair among A, B, C is independent, but once you know any two of them, the third is fully determined — so all three together are not mutually independent. This matters in statistics because many formulas (variance of a sum, for instance) only simplify correctly under the full mutual independence assumption, not merely pairwise independence.

Where conditional reasoning goes wrong in practice

Two well-documented cognitive traps stem directly from confusing P(A | B) with P(B | A) — two quantities that are related by Bayes' theorem but are not generally equal. The prosecutor's fallacy occurs when a very small P(evidence | innocent) is presented as if it were the (much more relevant, and often very different) P(innocent | evidence). A DNA match with a 1-in-a-million random match probability does not mean there is only a 1-in-a-million chance the defendant is innocent; that conclusion also depends on how many people could plausibly have been tested and on the prior evidence in the case.

Base rate neglect is the related tendency to ignore how common or rare an event is in the general population when interpreting a conditional probability, a pattern explored in depth in the context of medical testing under Bayes' theorem. Both errors share a root cause: conditional probability is directional, and reversing the direction without doing the actual calculation almost never gives the right answer.