← 🎲 Counting & Probability
Subtopic 4 of 5

4.4Conditional probability & Bayes

The idea

Conditional probability and Bayes: update a probability once you know something has happened.

Think of it like this: Inspecting parts from two suppliers: given that this part passed, who most likely made it?

Why it exists

Draws without replacement and "given that" questions need this.

How it works

  1. Split into cases by the first event.
  2. Probability of each case × probability of the next event.
  3. Add the cases.
  4. Bayes: wanted case ÷ all cases that produce the evidence.

Worked example: X supplies 60% (96% reliable), Y supplies 40% (72% reliable)

P(Y | reliable) = 0.288 / (0.576 + 0.288) = 0.333.

Now play with it

  • →Change the percentages and watch the posterior.
  • →Use the ball tile for the replaced orange ball.

Take 100 items:

X items (60)60
...of which reliable (57.6)57.6
Y items (40)40
...of which reliable (28.8)28.8
P(reliable)0.864
P(made by Y | reliable)0.3333
P(made by X | reliable)0.6667

Bayes: the answer is the part of the reliable items that came from Y. The reliable items from Y divided by all reliable items.

First draw green, then orange3/10 = 0.3000
First draw orange (replaced), then orange4/25 = 0.1600
P(orange on the next draw)0.4600

Condition on the first draw: green is not replaced, orange is replaced by another orange. 3/5 · 2/4 + 2/5 · 2/5 = 23/50.

Check yourself

3 green, 2 orange; green not replaced, orange replaced. P(orange next) =

Remember: Split by the first outcome, then add the cases.
Exam notes (the dense version)

P(A | B) = P(A and B) ÷ P(B). Bayes: P(cause | evidence) = P(cause) × P(evidence | cause) ÷ P(evidence), where P(evidence) adds all ways it can happen (total probability). A draw that depends on the first draw (an orange ball is replaced, a green one is not) is solved by conditioning on the first outcome.

Previous year questions: Counting & Probability (10)

PYQs are grouped by topic, so this is the full set for Counting & Probability; try the ones that match this subtopic. Open all 10 on the topic page.