Conditional probability is the probability of an event given that another event has occurred: P(A|B) = P(A and B) / P(B). It is the workhorse of updating beliefs with evidence. The probability of rain given that the sky is cloudy is higher than the unconditional probability of rain. The probability that a positive test is a true positive depends on the base rate of the condition and the test's accuracy — and often contradicts intuition.
The failure mode is base-rate neglect. People overweight the evidence (e.g. "the test is 95% accurate") and underweight the prior (e.g. "the condition is rare"). So a positive result in a rare disease can still mean the probability of having the disease is low — because most positives are false positives when the base rate is low. Conditional probability, applied correctly via Bayes, fixes that: posterior = (likelihood × prior) / normalising constant.
In decisions, you constantly face conditional probabilities: probability of success given the team, of default given the sector, of a deal closing given the stage. Making them explicit avoids conflating P(positive test | disease) with P(disease | positive test) — the first is sensitivity; the second is what you need to decide. The discipline is to write down the condition and the event, and to update with the right formula.
Section 2
How to See It
You see conditional probability when someone updates a belief or a forecast based on new information, or when they confuse P(A|B) with P(B|A). The diagnostic: "given that X, what's the chance of Y?" and the need to use both the likelihood of the evidence and the prior.
Business
You're seeing Conditional Probability when a founder says "our conversion rate for users who complete onboarding is 40%." That is P(convert | onboard). The unconditional conversion rate might be 5% because most users don't onboard. Decisions (e.g. where to invest) depend on which probability is relevant.
Technology
You're seeing Conditional Probability when a model outputs P(fraud | transaction features). The decision (block or allow) depends on that conditional probability and on the cost of false positives vs false negatives. Confusing P(fraud | features) with P(features | fraud) would mis-set thresholds.
Investing
You're seeing Conditional Probability when an investor asks: given that the company missed revenue twice, what is the probability it hits the next plan? That is P(hit | two misses). It is not the same as P(two misses | underlying quality), and it requires a prior on quality and a model of persistence.
Markets
You're seeing Conditional Probability when someone interprets a signal: given that the Fed raised rates, what is the probability of recession? The answer depends on P(recession), P(hike | recession), and P(hike | no recession). Conditional probability structures the update.
Section 3
How to Use It
Decision filter
"When you get new information, ask: what probability do I need? P(outcome | evidence) or P(evidence | outcome)? Use the base rate and the likelihood of the evidence to update. Write P(A|B) = P(B|A)P(A)/P(B) when it helps; at minimum, don't treat a positive test as a sure thing when the condition is rare."
As a founder
Use conditional probabilities when interpreting metrics and signals. Conversion given onboarding, retention given first week activity, win rate given pipeline stage — each is P(outcome | condition). Don't quote the conditional rate when the unconditional rate is what matters for the business (e.g. overall conversion). When you get a "positive signal" (e.g. a pilot yes), ask: given how many pilots we run, how often would we see this by chance? That is base rate and likelihood in disguise.
As an investor
Every due diligence finding is an update: P(good investment | findings). Combine the prior (how often do we see a company this good?) with the likelihood (how likely are these findings if it's good vs bad). Avoid base-rate neglect: a charismatic founder or a great demo is evidence, not proof. The posterior probability of success still depends on how many similar-looking companies fail.
As a decision-maker
Before acting on a test or a signal, clarify: P(condition | positive) or P(positive | condition)? They can differ by an order of magnitude. Use the base rate and the test's sensitivity and specificity (or their analogues) to compute the right one. Decisions that assume P(positive | condition) = P(condition | positive) are systematically wrong when base rates are extreme.
Common misapplication: Swapping condition and event — treating P(A|B) as if it equalled P(B|A). Second misapplication: Ignoring the base rate. When the prior is very small or very large, the posterior can remain small or large even after strong-looking evidence. Always include the prior in the update.
Munger has repeatedly emphasised "inversion" and "base rates." Inversion is asking what would have to be true for the opposite outcome; base rates are the prior in a Bayesian update. His point: people neglect base rates and overreact to recent evidence. Conditional probability done right — prior + likelihood → posterior — is the antidote.
Thorp's work on blackjack and later quantitative investing relied on correct probability updates. Knowing P(win | count) and updating the count with each card is conditional probability in action. His book A Man for All Markets stresses using probability correctly and avoiding the confusions (e.g. prosecutor's fallacy) that come from swapping condition and event.
Section 6
Visual Explanation
Conditional probability: P(A|B) = P(A and B) / P(B). Bayes: posterior ∝ likelihood × prior. Don't confuse P(positive|disease) with P(disease|positive).
Section 7
Connected Models
Conditional probability is at the centre of Bayesian reasoning and decision under uncertainty. The grid below shows what reinforces it, what creates tension, and what it leads to.
Reinforces
Bayes Theorem
Bayes' theorem is the rule for updating probability given evidence: posterior ∝ likelihood × prior. It is conditional probability written in a form that makes the update explicit. The two are inseparable in practice.
Reinforces
Base Rate
The base rate is the prior probability before evidence. Conditional probability says the posterior depends on both the likelihood of the evidence and the base rate. Neglecting the base rate is the most common error; reinforcing it fixes that.
Tension
False Positives & False Negatives
Sensitivity is P(positive | condition); specificity is P(negative | no condition). The probability you need for decisions is P(condition | positive), which depends on both and on the base rate. The tension: good sensitivity does not imply that a positive result usually means the condition — base rate can dominate.
Tension
Prior and Posterior
The prior is the belief before evidence; the posterior is after. Conditional probability is the update rule. The tension: people often anchor on the likelihood (e.g. "test is 95% accurate") and forget the prior. Making prior and posterior explicit reduces the error.
Section 8
One Key Quote
"Nothing in life is as important as you think it is when you are thinking about it. Unless you apply base rates."
— Daniel Kahneman
The quote is about focusing illusion, but it applies to conditional probability: without base rates, you overweight the case in front of you. The correct update — posterior = (likelihood × prior) / P(evidence) — forces the base rate into the calculation. What you think when you're "thinking about it" is often the likelihood; what you need for the decision is the posterior.
Section 9
Analyst's Take
Faster Than Normal — Editorial View
Always specify the condition. When you hear "the probability of X," ask: probability given what? P(success | seed round) is different from P(success | Series B). P(win | this team) is different from P(win) overall. Writing the condition avoids confusion and forces the right reference class.
Do the Bayes update for high-stakes signals. When a test, a metric, or a signal is positive, compute P(condition | positive) using the base rate and the test's accuracy. You will often find the posterior is lower than intuition says. That prevents overreacting to false positives.
Don't swap condition and event. P(A|B) and P(B|A) are not the same. Sensitivity is P(positive | disease); the decision-relevant quantity is P(disease | positive). Training yourself to name both and to compute the one you need eliminates a large class of reasoning errors.
Section 10
Test Yourself
Is this mental model at work here?
Scenario 1
A test is 99% accurate. 1 in 10,000 people have the condition. Someone tests positive. A doctor says 'you're almost certainly fine.'
Scenario 2
A founder says '80% of users who complete onboarding convert.' They conclude 'onboarding is the key to conversion.'
Scenario 3
An investor says 'given that they missed twice, the probability they hit next quarter is low.'
Scenario 4
A recruiter says 'this candidate passed our bar, and our bar is 90% accurate, so they're 90% likely to be great.'
Section 11
Summary & Further Reading
Summary: Conditional probability is P(A|B) — the probability of A given B. It is the basis for updating beliefs with evidence. Use it to avoid base-rate neglect and to avoid swapping P(A|B) and P(B|A). Apply Bayes when you have a prior and a likelihood and need the posterior.
Rigorous treatment of conditional probability, Bayes' rule, and their use in inference.
Leads-to
Probability Theory
Conditional probability is a core concept in probability theory. It leads to conditional expectation, martingales, and full Bayesian inference. The basic definition is the starting point for all of that.
Leads-to
Information Asymmetry
When one side has private information, the other updates beliefs conditional on observed actions or signals. Conditional probability structures how beliefs evolve in games and markets with asymmetric information.
