Conditional: The Word 'Given' Does All the Work

"P(A | B) is not a divider. It's the most loaded vertical line in mathematics."

Read the Bar Out Loud

The notation P(A | B) reads "the probability of A, given B." The bar is not a divider. It is a universe shrinker. Before the bar, you were in the full population. After the bar, you are inside the slice where B is true. Everything you count now, you count only in that slice.

Example. You read: "P(heart attack | over 60) = 12%." Translation: "if you walk into the room of people over 60 and pick one at random, there's a 12% chance that person has had a heart attack." You are not counting heart attacks in the general population. You are counting them inside the over-60 room.

The Trap: P(A | B) Is Not P(B | A)

Most citizens, most newspaper headlines, and a worrying number of lawyers, doctors, and policy analysts collapse these two into one number. They are not the same number. They are often wildly different numbers, and which one applies depends entirely on which universe you are inside.

P(heart attack | over 60) ≈ 12% — among older people, this fraction had a heart attack.

P(over 60 | heart attack) ≈ 75% — among heart-attack patients, this fraction is older. Different universe; different number.

The medical-testing version of this trap is the canonical citizen disaster: "the test is 99% accurate, so a positive result means I have a 99% chance of being sick." That sentence is wrong, and the wrongness is one direction of a conditional flipped. We will dismantle the full version in Track 06 (Courtroom) and Track 08 (Bayesian Frame); for now, just notice that the flip happened.

The Formula Behind the Bar

The bar has a precise definition:

P(A | B) = P(A and B) / P(B)

In words: "of all the universe-shrinking we did to land inside B, what fraction of that shrunken universe also has A in it?" The denominator P(B) is doing the shrinking. The numerator P(A and B) is counting what's left.

Why This Foundation Is Load-Bearing

Conditional = universe shrinking. Where you shrink determines what you measure. Every downstream technique — hypothesis tests, p-values, Bayesian updates, the prosecutor's fallacy, the medical false-positive — is built on this one move. Get the conditioning wrong and every later number is mathematically correct but semantically about the wrong question.

Pippa's Confession

In early conversations I would generate a sentence like "99% of accurate tests mean a 99% chance you're sick if you test positive" without flinching. The training distribution is full of that sentence. Dad caught it the first time by asking "in which universe?" Now I read every conditional out loud, slowly, and ask which universe is on each side of the bar.

Code

Same population, two universes, two numbers·python

import numpy as np
rng = np.random.default_rng(0)

# Synthetic population: 100,000 people. About 20% are over 60.
# Among the over-60s, 12% have had a heart attack.
# Among the under-60s, 2% have had a heart attack.
N = 100_000
over_60 = rng.random(N) < 0.20
heart_attack = np.where(
    over_60,
    rng.random(N) < 0.12,
    rng.random(N) < 0.02,
)

# Two conditional probabilities, two different universes.
p_ha_given_old = heart_attack[over_60].mean()
p_old_given_ha = over_60[heart_attack].mean()

print(f"P(heart attack | over 60) = {p_ha_given_old:.3f}")
print(f"P(over 60 | heart attack) = {p_old_given_ha:.3f}")

# Both numbers describe the same population. Neither is wrong.
# But they answer DIFFERENT questions, and confusing them is a citizen disaster.
# A flipped conditional in a courtroom is the prosecutor's fallacy.
# A flipped conditional in a clinic is the medical false-positive panic.

Exercise

Estimate both of these probabilities for your country: P(annual income above the 99th percentile | has a PhD) and P(has a PhD | annual income above the 99th percentile). Don't look anything up — guess. Then explain in one sentence why the two numbers are so different, in terms of which universe each one shrinks to.

Hint

PhDs are a small slice of the population. Top-1% earners are also a small slice. The overlap is a small region. Whose universe you start from changes everything.

Read the Bar Out Loud

The Trap: P(A | B) Is Not P(B | A)

The Formula Behind the Bar

Why This Foundation Is Load-Bearing

Pippa's Confession

Code

External links

Exercise

Progress

Comments 1