Two relationships events can have, which the book opens by warning are not the same thing. Independence is about information: two events are independent when knowing that one occurred does not affect the chance of the other, which can be tested by any one of three equivalent conditions. Mutual exclusivity is about overlap: two events are mutually exclusive when they cannot occur at the same time, so their intersection is empty and its probability is zero. Sampling with replacement makes draws independent and sampling without replacement does not. The two ideas are very nearly opposites, because mutually exclusive events of non-zero probability are maximally dependent — learning that one occurred settles the other completely.
Subject: Statistics · 67 slides · symbolic lesson
Open the interactive version of this deck
Title
Statistics · Chapter 3 — Probability Topics
Two different relationships a pair of events can have, and why they are nearly opposites
Objectives
The section opens with a warning: independent and mutually exclusive do not mean the same thing. In everyday speech both sound like the two events have nothing to do with each other. In probability they answer two completely different questions, and this lesson is about keeping them apart.
Here is the one-line version to carry through every slide. Independence is a question about information: does learning that one event happened change the chance of the other? Mutual exclusivity is a question about overlap: can the two events happen at the same time? A pair of events gets an answer to each question separately.
Along the way you will see why the two answers are linked in one surprising way: if two events cannot happen together, then learning that one happened tells you everything about the other, so they are about as far from independent as events can be.
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, p. 172 — the opening warning
Warm-up
Section 3.1 ended with Example 3.3, a sample of 100 people sorted by sex and handedness. Everything in this lesson is built from conditional probabilities like the ones you computed there.
| right-handed | left-handed | total | |
|---|---|---|---|
| males | 43 | 9 | 52 |
| females | 44 | 4 | 48 |
| total | 87 | 13 | 100 |
Discussion prompt
Find the probability of right-handed, and the probability of right-handed given male. Are they equal? Say in one sentence what the comparison tells you.
Notice the move you just made. You computed the chance of an event in the whole sample, then again inside a smaller group, and compared the two. That comparison is the whole idea of independence. If the two numbers had matched, being male would carry no information about handedness.
Keep the formula in view as well, because every test in this lesson uses it: the probability of A given B is the probability of A AND B divided by the probability of B.
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.1, pp. 171-172 — Example 3.3
Concept
Take any two events A and B defined on the same experiment. There are two separate things you might want to know about how they relate.
| question | property | the number that settles it |
|---|---|---|
| Does knowing B happened change the chance of A? | independence (if no) | compare P(A given B) with P(A) |
| Can A and B happen at the same time? | mutual exclusivity (if no) | is P(A AND B) zero? |
The first question is about information. Think of a friend who peeks at the outcome and tells you only that B happened. If your estimate for A should not move, the events are independent. If it should move up or down, they are dependent.
The second question is about overlap. Look at the outcomes in A and the outcomes in B. If no outcome belongs to both, they cannot happen together and they are mutually exclusive.
\[ \text{independent: } P(A \mid B) = P(A) \qquad \text{exclusive: } P(A\text{ AND }B) = 0 \]
Look at the two conditions side by side. One compares a conditional with a plain probability; the other asks whether an intersection is empty. They use different numbers, so it should not surprise you that their answers are unrelated, and later you will see that they almost never agree.
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, pp. 172-174 — the two definitions
Section
Part 1
Concept
Two events A and B are independent if the knowledge that one occurred does not affect the chance the other occurs. Written as a conditional probability, that is one equation.
\[ P(A \mid B) = P(A) \]
Read it slowly. The left side is the chance of A once you know B happened; the right side is the chance of A with no news at all. Independence says the news was worthless for predicting A.
The book's example is two rolls of a fair die. The die has no memory. Whatever the first roll showed, the second roll still has six equally likely faces, so the chance of a six on the second roll is one sixth before and after you see the first.
\[ P(\text{six on roll 2} \mid \text{six on roll 1}) = \frac16 = P(\text{six on roll 2}) \]
dependent events — Two events that are not independent: learning that one occurred changes the probability of the other.
Independence is a precise numerical claim, not a feeling that two things are unrelated. Shoe size and reading ability feel unrelated, yet across all children they are strongly dependent, because both grow with age. The only way to know is to compare the numbers.
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, pp. 172-173 — the definition of independent events
Notation
The book lists three conditions and says that showing any one of them is enough. The first and the last are shown below; the middle one is the first with A and B swapped, and it has its own note.
Annotate
On: \( P(A \mid B) = P(A) \iff P(A\text{ AND }B) = P(A)P(B) \)
The first two are the definition, read in each direction. The third is the one people remember as the multiplication rule for independent events, and it has a big practical advantage: it never divides, so it works even when you have only the three probabilities of A, of B, and of both.
Which one should you use? The book's note on Example 3.9 answers this: the choice depends on the information you have. If a conditional is handed to you, compare it. If the joint probability is handed to you, multiply. Never compute more than you need.
One caution about rounded data. When the given numbers are rounded decimals, the two sides of a test can disagree in the last digit even for independent events. Treat 0.2667 against 0.2666 as a match and 0.45 against 0.30 as a real difference.
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, pp. 172-173 and 176 — the three conditions and the note on Example 3.9
Concept
It is worth seeing why the three conditions always stand or fall together, because then you will never wonder whether you checked the right one. Start from the definition of a conditional probability.
\[ P(A \mid B) = \frac{P(A\text{ AND }B)}{P(B)} \]
Suppose the first condition holds, so the left side equals the probability of A. Substitute that in.
\[ P(A) = \frac{P(A\text{ AND }B)}{P(B)} \]
Multiply both sides by the probability of B. The fraction clears and the third condition appears: the overlap is the product.
\[ P(A\text{ AND }B) = P(A)\,P(B) \]
Now use the conditional formula the other way round, for B given A, and substitute the product for the overlap.
\[ P(B \mid A) = \frac{P(A\text{ AND }B)}{P(A)} = \frac{P(A)\,P(B)}{P(A)} = P(B) \]
The probability of A cancels and the second condition appears. Every step can be run backwards too, so each condition implies the other two. One small print: this needs both probabilities to be positive, since you divide by them. An event with probability zero is a degenerate case the book never uses.
Intuition
The definition is written with A on the left of the bar, but nothing about independence prefers A. If news of B tells you nothing about A, then news of A tells you nothing about B. The derivation on the last slide proved this; here is why it makes sense.
Suppose a quarter of all students take art, and a quarter of the athletes take art too. Then art takers are spread evenly over athletes and non-athletes, in proportion to their sizes. So the athletes' share among art takers must be the same as their share of everyone.
\[ P(\text{art} \mid \text{ath}) = P(\text{art}) \iff P(\text{ath} \mid \text{art}) = P(\text{ath}) \]
Dependence runs both ways too. In Example 3.12, being a woman raised the chance of long hair from one half to three quarters; turning it round, having long hair raises the chance of being a woman from six tenths to nine tenths.
\[ P(W \mid L) = \frac{0.45}{0.50} = 0.9 > 0.60 = P(W) \]
This is why the word independent describes a pair, never one event on its own. It is also why dependence is not causation: long hair does not cause anyone to be a woman, yet the information flows in both directions just the same.
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, p. 178 — Example 3.12
Picture it
Here is a picture you can carry into every problem, built from Example 3.9. A whole college population is a square of area 1. Cut it into two columns by whether a student takes science: half do, so each column has width 0.5.
Figure (svg): A unit square split into two columns of equal width 0.5: 'science' and 'no science'. In each column the bottom 0.6 of the height is shaded and labelled math. A dashed yellow line at height 0.6 runs across both columns and is labelled 'P(G) = 0.6'.
Now shade, inside each column, the share of students taking math. Among science students the share is 0.6; among the others it is also 0.6. The shaded band runs at the same height straight across the square.
\[ \text{area of math AND science} = 0.5 \times 0.6 = 0.3 \]
That level band is independence, drawn. Being told a student is in the science column does not move your estimate of math, because math fills the same fraction of every column. And the area of the shaded block in the science column is width times height, which is exactly the product condition.
Keep this picture for what comes later. When the band steps up or down between columns, the events are dependent, and the size of the step tells you how strongly. You will see a sharply stepped band on the long-hair example in a few slides.
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, p. 176 — Example 3.9
Worked example
Let G be taking a math class and H taking a science class. You are told the probability of G is 0.6, of H is 0.5, and of both is 0.3. Are G and H independent?
\[ P(G) = 0.6, \quad P(H) = 0.5, \quad P(G\text{ AND }H) = 0.3 \]
You have all three numbers, so any of the three tests will work. The book asks you to run two of them, which is a good habit the first time: seeing two tests agree shows you they really are the same question.
(a) Write the conditional test
Why: the chance of math among science students, by the definition of a conditional.
\[ P(G \mid H) = \frac{P(G\text{ AND }H)}{P(H)} \]
Substitute
Why: overlap 0.3 over science 0.5.
\[ P(G \mid H) = \frac{0.3}{0.5} \]
Divide and compare with the plain chance of math
Why: the news changed nothing.
\[ P(G \mid H) = 0.6 = P(G) \]
(b) Run the product test
Why: multiply the two plain chances and compare with the given overlap.
\[ P(G)\,P(H) = (0.6)(0.5) = 0.3 = P(G\text{ AND }H) \]
Both tests say independent. In context: knowing that a student takes science does not change the 60 percent chance that they take math.
Check with the third condition, the one the book leaves for practice
Why: the chance of science among math students must equal the plain chance of science.
\[ P(H \mid G) = \frac{0.3}{0.6} = 0.5 = P(H) \]
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, p. 176 — Example 3.9
Trap
After Example 3.9, a common line of reasoning:
\[ P(G \mid H) = 0.6 \]
\[ P(H \mid G) = 0.5 \]
\[ 0.6 \ne 0.5 \Rightarrow \text{dependent} \]
Wrong. Every number is right; the comparison is not one of the tests.
It feels natural because the two conditionals are the two numbers you just computed, and symmetry suggests they should agree.
Each conditional is compared with its own event's plain chance, never with the other conditional:
\[ P(G \mid H) = 0.6 = P(G) \]
\[ P(H \mid G) = 0.5 = P(H) \]
Under independence the two conditionals equal two different numbers, the chances of G and of H, so they match each other only when those two happen to be equal. Comparing them tests nothing.
The habit that prevents this: before comparing, say aloud which plain chance the conditional should be compared with. It is always the event to the left of the bar.
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, p. 176 — Example 3.9
Prediction
Let A be learning Spanish and B learning German. You are told the chance of A is 0.4, of B is 0.2, and of both is 0.08. Before computing anything, make a call.
Predict first
Using only the three given numbers, are learning Spanish and learning German independent?
Correct: independent
Why: The product test needs exactly these three numbers. Four tenths times two tenths is eight hundredths, which matches the given overlap, so the events are independent. Equivalently, Spanish among German learners is 0.08 over 0.2, which is 0.4, the same as for everyone.
The book's hint lists the three conditions again, and the point of the hint is that you get to choose. You have the joint probability and both single probabilities, so the product test is a single multiplication.
If you guessed that the numbers cannot settle it, notice why they can. Independence is a statement about three numbers, the two single chances and the overlap, and you were handed all three. Nothing about the people or the languages is needed beyond them.
If you guessed dependent, ask what made you think so. Perhaps that someone keen on languages takes both. That is a reasonable story, but stories are hypotheses; the numbers here say that the share of Spanish learners is exactly the same among German learners as among everyone.
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, p. 176 — Try It 3.8
Worked example
Now write it out properly. Name the test, compute what independence predicts, and compare with what you were given.
\[ P(A) = 0.4, \quad P(B) = 0.2, \quad P(A\text{ AND }B) = 0.08 \]
The logic of any test is the same. Assume the events are independent and work out what the overlap would then have to be. If the real overlap matches, independence holds. If it misses, the assumption was false.
Notice that the test never needed the conditional chances. That is the practical strength of the product form: whenever a problem gives you the overlap directly, one multiplication and one comparison settle the question.
Write the product condition
Why: chosen because all three of its ingredients are given.
\[ P(A\text{ AND }B) \overset{?}{=} P(A)\,P(B) \]
Compute the right side
Why: four tenths times two tenths.
\[ P(A)\,P(B) = (0.4)(0.2) = 0.08 \]
Compare with the given overlap
Why: the two numbers are equal exactly.
\[ 0.08 = P(A\text{ AND }B) \]
So learning Spanish and learning German are independent. In context: 40 percent of these students learn Spanish, and among the German learners the share is still 40 percent.
Check with a conditional test
Why: Spanish among German learners must come back to the plain 0.4.
\[ P(A \mid B) = \frac{0.08}{0.2} = 0.4 = P(A) \]
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, p. 176 — Try It 3.8
Worked example
In a class, 60 percent of the students are women, 50 percent have long hair, 45 percent are women with long hair, and of the women 75 percent have long hair. Let W be woman and L long hair. Are W and L independent?
\[ P(W) = 0.60, \quad P(L) = 0.50, \quad P(W\text{ AND }L) = 0.45, \quad P(L \mid W) = 0.75 \]
The book's note says you can use the first or the last condition here but not the middle one, because you are not given the chance of W given L. You could compute it, but you do not need to: one failed test settles the question.
Run the product test
Why: compute what independence would predict for the overlap.
\[ P(W)\,P(L) = (0.60)(0.50) = 0.30 \]
Compare with the given overlap
Why: 0.45 is half as large again as 0.30.
\[ P(W\text{ AND }L) = 0.45 \ne 0.30 \]
Run the conditional test as well
Why: long hair among women against long hair overall.
\[ P(L \mid W) = 0.75 \ne 0.50 = P(L) \]
Figure (svg): A unit square split into a 'women' column of width 0.6 and a 'men' column of width 0.4. The women column is shaded to height 0.75; the men column only to height 0.125. A dashed yellow line at height 0.5 is labelled 'P(L) = 0.5'.
The picture shows how big the gap is. Among the men, long hair is the remaining 0.05 of the class spread over a column of width 0.4, which is 0.125. Knowing a student is a woman changes the chance of long hair dramatically.
Check that the given numbers are consistent
Why: the 75 percent must be the overlap divided by the share of women, and it is.
\[ P(L \mid W) = \frac{0.45}{0.60} = 0.75 \]
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, p. 178 — Example 3.12
Trap
Asked for the chance that a random student is a woman with long hair, someone writes:
\[ P(W\text{ AND }L) = P(W)\,P(L) \]
\[ = (0.60)(0.50) = 0.30 \]
Wrong. The arithmetic is perfect, but the first line assumed independence that was never shown.
The product is the formula everyone remembers, so it gets used whenever two chances appear side by side.
The product is only the overlap when the events are independent. Here the overlap is given, and it is not the product:
\[ P(W\text{ AND }L) = 0.45 \]
The error costs a third of the answer: 30 long-haired women predicted in a class of 100, where 45 were counted. The book's rule exists to prevent this: if you do not know, assume dependent and do not multiply.
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, pp. 173 and 178 — the default assumption; Example 3.12
Worked example
Let C be taking an English class and D a speech class. You are told the chance of C is 0.75, of D is 0.3, of C given D is 0.75, and of both is 0.225. Justify each answer numerically.
\[ P(C) = 0.75, \quad P(D) = 0.3, \quad P(C \mid D) = 0.75, \quad P(C\text{ AND }D) = 0.225 \]
Part (a) asks about independence and you are handed a conditional, so compare it with the plain chance. Part (b) asks about overlap, which is a different question with a different number.
Notice that the book gives you more than you need. The chance of C given D alone settles part (a), the overlap alone settles part (b), and part (c) uses the overlap again. Picking the right number for each question is half the skill.
(a) Compare the conditional with the plain chance
Why: English among speech students against English overall.
\[ P(C \mid D) = 0.75 = P(C) \]
Conclude
Why: one condition is enough, so the events are independent.
\[ C \text{ and } D \text{ are independent} \]
(b) Look at the overlap
Why: mutually exclusive would need this to be zero.
\[ P(C\text{ AND }D) = 0.225 \ne 0 \]
So C and D are independent but not mutually exclusive. In context: 22.5 percent of students take both classes, and taking speech does not change the 75 percent chance of taking English.
(c) The other conditional
Why: overlap divided by the chance of C.
\[ P(D \mid C) = \frac{0.225}{0.75} = 0.3 \]
Check part (c) against independence
Why: since the events are independent, speech among English students must equal speech overall, and the product must equal the overlap.
\[ P(D \mid C) = 0.3 = P(D), \quad (0.75)(0.3) = 0.225 \]
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, p. 177 — Example 3.10
Worked example
At a library, let B be checking out a book and D a DVD. The chance of B is 0.40, of D is 0.30, and of both is 0.20. Find both conditionals and decide both questions.
\[ P(B) = 0.40, \quad P(D) = 0.30, \quad P(B\text{ AND }D) = 0.20 \]
This time no conditional is given, so you build each one from the definition. Then each conditional is compared with its own plain chance, never with the other conditional.
(a) Book given DVD
Why: overlap over the chance of the condition, the DVD.
\[ P(B \mid D) = \frac{0.20}{0.30} \approx 0.667 \]
(b) DVD given book
Why: the same overlap, now over the chance of a book.
\[ P(D \mid B) = \frac{0.20}{0.40} = 0.5 \]
(c) Compare each with its own plain chance
Why: both conditionals are far above their baselines.
\[ 0.667 \ne 0.40, \qquad 0.5 \ne 0.30 \]
Figure (svg): A unit square split into a 'DVD' column of width 0.3 and a 'no DVD' column of width 0.7. The DVD column is shaded to height 0.667, the other to height 0.286. A dashed yellow line at height 0.4 is labelled 'P(B) = 0.40'.
So the events are dependent: a DVD borrower is much more likely than average to take a book too. And part (d): the overlap is 0.20, not zero, so they are not mutually exclusive either.
Check with the product test
Why: it must reach the same verdict, and it does: independence would predict 0.12, but the real overlap is 0.20.
\[ (0.40)(0.30) = 0.12 \ne 0.20 \]
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, p. 177 — Try It 3.10
Worked example
A box holds three red cards numbered 1 to 3 and five blue cards numbered 1 to 5. You draw one card without looking. Let R be red, B blue, and E an even number.
Figure (svg): Eight cards in a row: R1, R2, R3 in red, then B1 to B5 in blue. B4 and B5 have thick yellow outlines and are labelled G, number above 3. A green bracket under B1 to B4 is labelled H. The cards R2, B2 and B4 carry the word even.
All eight cards are equally likely, so each probability is a count over 8, and a conditional is a count inside a smaller group. That makes every test here a matter of counting cards in the picture.
Single chances
Why: three red, five blue, three even (R2, B2, B4).
\[ P(R) = \tfrac38, \quad P(B) = \tfrac58, \quad P(E) = \tfrac38 \]
Even given blue
Why: of the five blue cards, two are even.
\[ P(E \mid B) = \tfrac25 = 0.4 \]
Blue given even
Why: of the three even cards, two are blue. The book's text says to are blue, a slip for two.
\[ P(B \mid E) = \tfrac23 \]
Let G be a number above 3, and H blue numbered 1 to 4
Why: G is B4 and B5; H is B1 to B4.
\[ P(G) = \tfrac28 = \tfrac14 \]
G given H
Why: only B4 of the four cards in H is above 3.
\[ P(G \mid H) = \tfrac14 = P(G) \]
So G and H are independent, while R and B are mutually exclusive: no card is both red and blue. In context: being told the card is a blue one from 1 to 4 leaves a one-in-four chance that its number is above 3, the same as for the whole box.
Check G and H with the product test
Why: one card, B4, is in both.
\[ P(G\text{ AND }H) = \tfrac18 = \tfrac14 \cdot \tfrac48 \]
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, p. 177 — Example 3.11
Comparison
Each row is a pair of events from this part of the lesson. Fill in the blanks, then read across. Every row compares a conditional with the plain chance of the same event, which is the definition test.
Comparison matrix
| events | chance given the other | plain chance | independent? |
|---|---|---|---|
| Ex 3.9: math given science | 0.6 | 0.6 | yes |
| Ex 3.12: long hair given woman | 0.75 | 0.5 | no |
| TI 3.10: book given DVD | 0.667 | 0.40 | no |
| Ex 3.11: even given blue | 0.4 | 0.375 | no |
The last row is the subtle one. Two of the five blue cards are even, which is 0.4, while three of the eight cards are even, which is 0.375. The gap is small, but these are exact fractions, so it is not rounding: being blue nudges the chance of an even number up a little.
Contrast that with G and H in the same box, which came out exactly independent. Independence is not a property of a box of cards; it is a property of a particular pair of events, and two pairs from the same box can get different verdicts.
The long-hair row is the other extreme: three quarters against one half is a big jump, and the picture of that class showed a steeply stepped band. The size of the gap is a rough measure of how dependent the events are.
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, pp. 176-178 — Examples 3.9, 3.11, 3.12 and Try It 3.10
Fill the middle
A bag holds six red marbles numbered 1 to 6 and four green marbles numbered 1 to 4. Let G be a green marble and O an odd-numbered one. The sample space has ten equally likely outcomes.
Figure (svg): Ten marbles in a row: six red marked R1 to R6 and four green marked G1 to G4. The odd-numbered marbles R1, R3, R5, G1 and G3 have dashed yellow rings around them.
Fill in the blanks
P(G\text2O) = \frac0.2}___ = ___
Why: The marbles that are green and odd are G1 and G3, two of the ten. Two tenths is one fifth, or 0.2.
Count the marbles that are both green and ringed in the picture: G1 and G3. Every probability here is a count over ten, because each marble is equally likely to be drawn.
The Try It stops there, but you now have all three numbers for the product test. Green has chance four tenths and odd has chance five tenths, since R1, R3, R5, G1 and G3 are odd.
\[ P(G)\,P(O) = (0.4)(0.5) = 0.2 = P(G\text{ AND }O) \]
So green and odd are independent. In context: half of the green marbles are odd, and half of all the marbles are odd, so learning the colour tells you nothing about whether the number is odd. Change one marble, say renumber G3 as G6, and the independence would break.
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, p. 176 — Try It 3.9
Worked example
In an arena, 70 percent of fans root for the home team, 25 percent wear blue, 20 percent wear blue and root for the away team, and of the away fans 67 percent wear blue. Let A be rooting for the away team and B wearing blue.
First translate every sentence into notation. The phrase of the fans rooting for the away team marks a conditional, and the away team's share comes from the complement of the home team's.
Away fans are the complement of home fans
Why: every fan roots for exactly one side.
\[ P(A) = 1 - 0.70 = 0.30 \]
Write the other givens
Why: wearing blue, both, and blue among the away fans.
\[ P(B) = 0.25, \quad P(A\text{ AND }B) = 0.20, \quad P(B \mid A) = 0.67 \]
Run the conditional test
Why: blue among away fans against blue overall.
\[ P(B \mid A) = 0.67 \ne 0.25 = P(B) \]
Run the exclusivity test
Why: can a fan root away and wear blue?
\[ P(A\text{ AND }B) = 0.20 \ne 0 \]
Figure (svg): A bar chart of the chance of wearing blue: away fans 0.667, home fans 0.071, and a dashed line across at 0.25 for all fans.
So the events are dependent and not mutually exclusive. In context: wearing blue is a strong sign of an away fan, since two thirds of away fans wear blue but only about 7 percent of home fans do.
Check that the 67 percent agrees with the other numbers
Why: the overlap divided by the away share is two thirds, which the book rounded to 67 percent.
\[ P(B \mid A) = \frac{0.20}{0.30} = 0.6667 \]
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, pp. 177-178 — Try It 3.11
Section
Part 2
Concept
Many probability problems draw several items from one population: cards from a deck, marbles from a bag, people from a list. How the drawing is done decides whether the draws are independent.
sampling with replacement — Each member is put back after it is picked, so the same member can be chosen more than once. The draws are independent.
sampling without replacement — Each member is set aside once picked, so it can be chosen only once. The draws are dependent.
The book's own examples are cards from a 52-card deck, with four suits of thirteen cards each, numbered 1 to 10 and then jack, queen and king. The picture shows the deck as it stands just before the second pick, under each scheme.
Figure (svg): Two 13 by 4 grids of small card cells, one row per suit (S, H, D, C) and one column per rank 1 to K. Left, titled 'with replacement': all 52 cells present, the queen of spades highlighted in yellow as drawn and put back; caption 52 cards to choose from. Right, titled 'without replacement': the king of hearts crossed out in red; caption 51 cards to choose from.
With replacement, the second draw faces exactly the same 52 cards as the first, so nothing about the first draw can change any chance on the second. That is independence built in by the procedure.
\[ \text{with: } \frac{1}{52}, \frac{1}{52}, \frac{1}{52} \qquad \text{without: } \frac{1}{52}, \frac{1}{51}, \frac{1}{50} \]
Without replacement, the population shrinks and changes. Each specific remaining card becomes slightly more likely, and the card already drawn becomes impossible. The second draw's chances depend on what the first draw removed, and that is dependence.
Notice the difference in where the independence comes from. In Example 3.9 independence was discovered in the data by running a test. Here it is guaranteed by the procedure itself: put the card back, reshuffle, and the second draw cannot know what the first one was.
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, p. 173 — with and without replacement
Worked example
Draw three cards from a fair, well-shuffled 52-card deck. In part (a) each card goes back and the deck is reshuffled; in part (b) each card is set aside. Find the chance of the exact sequence drawn each time.
The question is really about denominators. With replacement, each draw is from 52 cards. Without replacement, the deck loses one card per draw. Each draw is one specific card, so its numerator is 1.
(a) With replacement: Q of spades, ten of clubs, Q of spades
Why: every draw is from the full deck, so multiply three independent chances.
\[ \frac{1}{52} \cdot \frac{1}{52} \cdot \frac{1}{52} = \frac{1}{140608} \]
Note the repeat
Why: the queen of spades came up twice, which is possible only because it went back.
\[ \text{picks} = \{QS, 10C, QS\} \]
The second and third draws of part (a) faced exactly the same deck as the first, because the book reshuffles after replacing. Whatever came out before has no bearing on what comes out next.
(b) Without replacement: K of hearts, three of diamonds, J of spades
Why: 51 cards remain for the second draw and 50 for the third.
\[ \frac{1}{52} \cdot \frac{1}{51} \cdot \frac{1}{50} = \frac{1}{132600} \]
Each specific sequence without replacement is slightly more likely, because the denominators shrink. In context: no card can appear twice in part (b), so the probability lost on repeats is spread over the sequences that remain possible.
Check by counting sequences
Why: 52 times 51 times 50 ordered sequences of three different cards, and each is equally likely, so one of them has probability one over that count.
\[ 52 \cdot 51 \cdot 50 = 132600 \]
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, p. 173 — Example 3.4
Sorting
You see only the cards that were drawn. Sort each sequence by what it tells you about how it was drawn. The key is a repeated card: only one of the two schemes can ever produce one.
Sort into buckets
Sort each sequence of picks by which sampling scheme could have produced it.
Example 3.5 asks which of two sequences was drawn which way: the one with no repeat is the without-replacement draw, and the one with KH twice is the with-replacement draw. Try It 3.4 asks the harder question of whether you can decide at all.
A repeat settles it; no repeat settles nothing. With replacement, three draws avoid any repeat about 94 percent of the time, so a clean sequence is the expected outcome under both schemes.
That is the answer to Try It 3.4: the sequence with the queen of spades twice must be with replacement; the one with three different cards cannot be decided. Try It 3.5 is the same question reversed. Without replacement only the sequences with no repeat are possible; with replacement all three are.
\[ \frac{52 \cdot 51 \cdot 50}{52^3} \approx 0.943 \]
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, pp. 173-174 — Example 3.5, Try Its 3.4 and 3.5
Worked example
Draw two cards. What is the probability that both are hearts, first with replacement and then without? This is the comparison the whole of Part 2 has been building toward.
Let H1 be a heart on the first draw and H2 a heart on the second. The general rule for both is the chance of the first times the chance of the second given the first. Only the second factor differs between the schemes.
Before computing, predict the direction. Removing a heart leaves fewer hearts in a slightly smaller deck, so a second heart should be a little less likely without replacement.
With replacement: the second draw sees the full deck
Why: the first heart went back, so the conditional equals the plain chance.
\[ P(H_2 \mid H_1) = \frac{13}{52} = P(H_2) \]
Multiply
Why: independent draws, so a plain product.
\[ P(H_1\text{ AND }H_2) = \frac{13}{52} \cdot \frac{13}{52} = \frac{1}{16} = 0.0625 \]
Without replacement: one heart is gone
Why: 12 hearts remain among 51 cards.
\[ P(H_2 \mid H_1) = \frac{12}{51} \ne \frac{13}{52} \]
Multiply
Why: use the conditional, not the plain chance.
\[ P(H_1\text{ AND }H_2) = \frac{13}{52} \cdot \frac{12}{51} = \frac{156}{2652} \approx 0.0588 \]
Figure (svg): A two-stage tree for two cards dealt without replacement. First branches: heart 13/52 and not 39/52. After a heart: heart 12/51 and not 39/51. After a non-heart: heart 13/51 and not 38/51. Path ends: both hearts 0.0588, heart then not 0.1912, not then heart 0.1912, neither 0.5588.
In context: dealing two cards from one deck gives both hearts about 5.9 percent of the time; drawing twice with replacement gives 6.25 percent. The difference is small for a 52-card deck, but it is real.
Check that the without-replacement tree is consistent
Why: the four path ends must add to 1, and the chance of a heart on the second draw must still be one quarter overall.
\[ 0.0588 + 0.1912 + 0.1912 + 0.5588 = 1, \quad 0.0588 + 0.1912 = 0.25 \]
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, p. 173 — sampling with and without replacement
Trap
Two cards are dealt from one deck. The chance both are aces:
\[ \frac{4}{52} \cdot \frac{4}{52} \approx 0.00592 \]
Wrong. It treats the second deal as if the first ace had gone back into the deck.
The slip is natural because each card, taken alone, is an ace with chance four in 52. That is true, but it is not what the second factor asks for.
The second card is an ace with chance four in 52 only before you know what the first card was.
The second factor is the chance of an ace given the first was an ace. One ace and one card are gone:
\[ \frac{4}{52} \cdot \frac{3}{51} \approx 0.00452 \]
The error overstates the answer by about 31 percent. Update both numerator and denominator after every draw without replacement: the population that the next draw faces is the one left behind.
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, p. 173 — sampling without replacement
Concept
Without replacement the draws are dependent, but how much does it matter? It depends on how big the population is compared with the sample. Suppose a quarter of a population of size N are successes.
\[ P(\text{2nd success} \mid \text{1st success}) = \frac{N/4 - 1}{N - 1} \]
For a deck of 52 cards that is 12 over 51, about 0.235, noticeably below a quarter. For a population of 10,000 it is 2,499 over 9,999, about 0.24992, which no survey could tell apart from 0.25.
Figure (svg): A graph against population size N from 4 to 100. An upper curve, the chance of a success on the second draw after a failure, falls toward 0.25; a lower curve, after a success, rises toward 0.25; a dashed line marks 0.25. Dots at N = 52 read 0.255 and 0.235.
The graph shows both conditional chances, after a success and after a failure. They straddle the dashed line at one quarter, and the gap between them shrinks like one over N.
This is why surveys of a few thousand people from a city of millions are treated as independent draws even though nobody is asked twice. The dependence is there, but it is far too small to matter. Later chapters use a rule of thumb: if the sample is under about 10 percent of the population, treat draws as independent.
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, p. 173 — sampling without replacement
Tweak it
The curve below is the chance that the second draw is a success, given that the first draw was one, plotted against the population size N. The slider sets the fraction of the population that are successes.
Parameter explorer
Start with a success fraction of 0.25 and read the curve at N = 52 and at N = 100. Then raise the fraction to 0.5 and find the population size at which the second-draw chance first gets within 0.01 of 0.5.
\[ P(\text{2nd} \mid \text{1st}) = \frac{{p}N - 1}{N - 1} \]
At a quarter, the curve reads about 0.235 at a deck of 52 and about 0.242 at 100: already close to the with-replacement value of 0.25. At small N the curve dips far below the fraction, because removing one success is a big share of a small population.
Raise the fraction to one half. The second-draw chance is within 0.01 of a half from about N equals 51 onward. Removing one success from a large population changes almost nothing, which is the whole reason without-replacement surveys behave like independent draws.
Notice that the curve never reaches the fraction for any finite N. Without replacement is always a little dependent. The question a statistician asks is not whether the dependence exists but whether it is big enough to change any conclusion.
Estimation
Try It 3.4 showed that a repeated card proves the sampling was with replacement. But how often does a repeat actually happen? Estimate before you compute.
Predict first
You draw four cards with replacement from a 52-card deck. Roughly how often does at least one card appear twice?
Correct: about 11 percent of the time
Why: The chance of no repeat is the product of 52 over 52, 51 over 52, 50 over 52 and 49 over 52, about 0.889, so a repeat happens about 11 percent of the time. Most with-replacement draws of four cards look exactly like draws without replacement.
\[ 1 - \frac{52 \cdot 51 \cdot 50 \cdot 49}{52^4} \approx 1 - 0.889 = 0.111 \]
The complement does the work: no repeat means the second card avoids one card, the third avoids two, the fourth avoids three. Each factor is a with-replacement chance, so the draws are independent and the factors multiply.
This is why Try It 3.4(b) cannot be decided. A sequence with no repeat is what you see almost nine times in ten under replacement, and every time without it. A missing repeat is weak evidence; a present repeat is proof.
The number may feel high or low depending on your intuition. With 52 cards and only four draws a repeat seems unlikely, and it is; but it is not rare. Add more draws and it climbs quickly: by ten draws with replacement, a repeat happens about 60 percent of the time.
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, p. 173 — Try It 3.4
Section
Part 3
Concept
A and B are mutually exclusive if they cannot occur at the same time. In terms of outcomes, A and B share no outcome, so their intersection is empty and its probability is zero.
\[ A, B \text{ mutually exclusive} \iff P(A\text{ AND }B) = 0 \]
The book's example uses the numbers 1 to 10 as the sample space, with A the numbers 1 to 5, B the numbers 4 to 8, and C the numbers 7 and 9.
Figure (svg): The whole numbers 1 to 10 as dots on a line. A blue bar above 1 to 5 is labelled A; an orange bar above 4 to 8 is labelled B; the bars overlap above 4 and 5, which are coloured yellow and labelled A AND B. Green rings surround 7 and 9, labelled C.
Look for shared outcomes. A and B share 4 and 5, so they are not mutually exclusive. A and C share nothing at all, so they are. The test is a search for a common member, and a single one is enough to fail it.
mutually exclusive — Two events with no outcome in common. Their AND probability is zero.
Notice what the definition does not mention: conditional probability, information, or whether one event affects the other. It is purely about whether the two sets of outcomes overlap, which is why it is a different property from independence.
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, p. 174 — the definition and the numbers 1 to 10
Notation
Mutual exclusivity has one condition, not three. Each piece of it says something precise, and two of the pieces are easy to misread.
Annotate
On: \( A\text{ AND }B = \varnothing \iff P(A\text{ AND }B) = 0 \)
Notice what is missing from the condition: the probabilities of A and of B by themselves. Two events can each be very likely or very unlikely and still be exclusive; all that matters is whether they share an outcome.
Also missing is any statement about what is left over. Exclusive events need not cover the sample space. And there is no conditional probability anywhere in the condition, which is the clearest sign that this property has nothing to do with independence's question about information.
In practice you test it in one of two ways. With a listed sample space, write out the intersection and see whether it is empty. With given probabilities, look at the AND probability and see whether it is zero.
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, p. 174 — the definition of mutually exclusive
Worked example
With S the numbers 1 to 10, equally likely, A the numbers 1 to 5, B 4 to 8 and C the numbers 7 and 9, test each pair for mutual exclusivity. The book does two pairs; the third is from the homework.
\[ A = \{1,2,3,4,5\}, \quad B = \{4,5,6,7,8\}, \quad C = \{7, 9\} \]
For each pair, list the intersection first. Then count it over 10. An empty list means zero and the pair is mutually exclusive; anything else means it is not.
A AND B
Why: 4 and 5 are in both lists.
\[ A\text{ AND }B = \{4, 5\}, \quad P(A\text{ AND }B) = \frac{2}{10} \]
Verdict for A and B
Why: a positive AND probability.
\[ \frac{2}{10} \ne 0 \Rightarrow \text{not mutually exclusive} \]
A AND C
Why: neither 7 nor 9 is in A.
\[ A\text{ AND }C = \varnothing, \quad P(A\text{ AND }C) = 0 \]
B AND C
Why: 7 is in both B and C.
\[ B\text{ AND }C = \{7\}, \quad P(B\text{ AND }C) = \frac{1}{10} \]
So A and C are mutually exclusive, and the other two pairs are not. In context: if the number drawn is 7, then B and C have both happened, so they cannot be exclusive.
Check A and C by listing their union
Why: if they share nothing, their sizes simply add: five plus two is seven distinct numbers.
\[ A\text{ OR }C = \{1,2,3,4,5,7,9\}, \quad 5 + 2 = 7 \]
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, p. 174 — the 1-to-10 example
Worked example
Flip two fair coins. The sample space has four equally likely outcomes, HH, HT, TH and TT, where HT means heads on the first coin and tails on the second.
Figure (svg): A 2 by 2 grid. Rows are the first coin, H then T; columns are the second coin, H then T. Cells HH, HT, TH are shaded blue as event A, at most one tail. Cell TT is ringed red as B, all tails, and cell HH is ringed green as C, all heads.
The book defines six events in a row. Write each as a set of outcomes first; then each probability is the count over 4, and each relationship can be read off the grid.
A: at most one tail, meaning zero or one
Why: HH has zero tails; HT and TH have one.
\[ A = \{HH, HT, TH\}, \quad P(A) = \tfrac34 \]
B: all tails
Why: only TT; B is exactly the outcomes not in A.
\[ B = \{TT\} = A', \quad P(B) = \tfrac14 \]
C: all heads, and the relationship with B
Why: C is HH, and B and C share no outcome.
\[ C = \{HH\}, \quad P(B\text{ AND }C) = 0 \]
D: more than one tail; E: a head on the first flip
Why: the book says roll here, a slip for flip.
\[ D = \{TT\}, \tfrac14; \quad E = \{HT, HH\}, \tfrac24 \]
F: at least one tail
Why: one or two tails.
\[ F = \{HT, TH, TT\}, \quad P(F) = \tfrac34 \]
So B and C are mutually exclusive: you cannot get all tails and all heads on the same two flips. And A and B are mutually exclusive in a special way, since together they cover everything.
Also notice that B and D are the same set, all tails and more than one tail, and that A and F overlap in HT and TH. Different words can name the same event, and only the list of outcomes tells you.
Check the complement pair
Why: A and B share nothing and between them fill the sample space, so their chances add to 1.
\[ P(A) + P(B) = \tfrac34 + \tfrac14 = 1 \]
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, p. 174 — Example 3.6
Worked example
Draw two cards from a standard deck with replacement. Find the probability of getting at least one black card. This combines the two ideas of the lesson so far.
At least one is awkward to count directly: it includes black then red, red then black, and black then black. Its complement, no black at all, is a single case. So find the complement and subtract from 1.
Name the complement
Why: at least one black fails only when both cards are red.
\[ \text{(at least one black)}' = \text{red AND red} \]
Chance of red on each draw
Why: 26 of 52 cards are red.
\[ P(\text{red}) = \frac{26}{52} = \frac12 \]
Both red
Why: with replacement, the draws are independent, so the product rule applies.
\[ P(\text{red AND red}) = \frac12 \cdot \frac12 = \frac14 \]
Subtract from 1
Why: the complement rule.
\[ P(\text{at least one black}) = 1 - \frac14 = \frac34 \]
In context: three times in four, two cards drawn with replacement include at least one black card. The independence from replacement is what let you multiply in the third line.
Check by listing the colour patterns
Why: four equally likely patterns, BB, BR, RB and RR, and three of them contain a black card.
\[ \frac{|\{BB, BR, RB\}|}{4} = \frac34 \]
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, p. 174 — Try It 3.6
Worked example
Flip two fair coins again. Let F be at most one tail, G both faces the same, H a head on the first flip, and J all tails. Find the probabilities, then test two pairs for mutual exclusivity.
Figure (svg): The same 2 by 2 grid of two coin flips. The top row, HH and HT, is shaded orange as H, a head first. The diagonal cells HH and TT are ringed yellow as G, faces the same. Cell TT carries a purple label J, all tails.
(a) F: zero or one tail
Why: HH, HT and TH.
\[ P(F) = \tfrac34 \]
(b) G: both faces the same
Why: HH or TT.
\[ P(G) = \tfrac24 \]
(c) H: a head first, then either face
Why: HH or HT.
\[ P(H) = \tfrac24 \]
(d) F and G
Why: they share HH.
\[ F\text{ AND }G = \{HH\}, \quad P(F\text{ AND }G) = \tfrac14 \ne 0 \]
(e) J and H
Why: J is TT, and H holds only outcomes with a head first.
\[ J\text{ AND }H = \varnothing, \quad P(J\text{ AND }H) = 0 \]
So F and G are not mutually exclusive, and J and H are. In context: two heads count as at most one tail and as matching faces, so F and G can happen on the same flips; all tails can never begin with a head.
One habit worth building: when an event is described in words, translate it into a list of outcomes before testing anything. The phrase at most one tail hides HH, and that hidden HH is exactly the shared outcome that decides part (d).
Check part (d) with the addition rule from the counts
Why: if F and G overlap in one outcome, their union should have three plus two minus one outcomes, all four.
\[ F\text{ OR }G = \{HH, HT, TH, TT\}, \quad 3 + 2 - 1 = 4 \]
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, p. 175 — Example 3.7
Matching
A box holds one white ball and one red. You draw, replace, and draw again. The outcomes are WW, WR, RW and RR, equally likely. Match each event or pair to its result.
Match the pairs
Why: White twice is only WW, one outcome of four. Different colours is WR or RW, two of four. White twice and different colours can never both happen, so they are exclusive. Different colours and white first share WR, so they are not.
Each answer comes from listing outcomes. The event F is the single outcome WW; G is the two mixed outcomes WR and RW; H is WW and WR. Every relationship question is then a search for a shared outcome.
Because the ball is replaced, each draw is independent of the other, so each of the four outcomes has chance one half times one half, a quarter. That is what makes counting over four legitimate here.
The pair G and H is the one that catches people, because the descriptions sound unrelated: one is about colours being different, the other about the first draw. But the outcome WR is white first and two different colours at once, so the events overlap.
Notice too that white twice and different colours are exclusive without being complements: the outcome RR is in neither. A search for one shared outcome beats any amount of reasoning about what the events sound like.
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, p. 175 — Try It 3.7
Worked example
Roll one fair die. Let A be an odd face, B an even face, C an odd face larger than two, D an even face smaller than five, and E any face less than five.
Figure (svg): Six die faces drawn with pips, 1 to 6; odd faces outlined blue, even faces outlined orange. Beneath them three rows of bars: C under 3 and 5, D under 2 and 4, E under 1, 2, 3 and 4. A dashed yellow line runs down through the 3, where the C and E bars both sit.
A and its complement
Why: the odd and even faces together make up the sample space.
\[ A = \{1,3,5\}, \quad A' = B = \{2,4,6\}, \quad P(A) = P(B) = \tfrac36 \]
C and D
Why: odd faces above two, even faces below five.
\[ C = \{3,5\}, \quad D = \{2,4\}, \quad P(C\text{ AND }D) = 0 \]
C and E
Why: E is the faces below five, and 3 is in both.
\[ E = \{1,2,3,4\}, \quad C\text{ AND }E = \{3\}, \quad P(C\text{ AND }E) = \tfrac16 \]
So C and D are mutually exclusive, because no face is odd and even at once, and C and E are not. In context: a roll of 3 is both an odd face above two and a face below five.
The last part asks for the chance of C given A. A conditional shrinks the sample space to the condition: once you know the face is odd, only 1, 3 and 5 remain, and two of those are in C.
C given A
Why: the reduced sample space is the three odd faces.
\[ P(C \mid A) = \tfrac23 \]
Check with the conditional formula
Why: the overlap of C and A is C itself, with chance two sixths, divided by the chance of A, three sixths.
\[ P(C \mid A) = \frac{2/6}{3/6} = \tfrac23 \]
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, p. 175 — Example 3.8
Trap
In Example 3.8, C is 3 and 5, D is 2 and 4. A common claim:
\[ C, D \text{ exclusive} \Rightarrow D = C' \]
\[ \Rightarrow P(C) + P(D) = 1 \]
Wrong. Exclusive says only that they do not overlap. It says nothing about covering the sample space.
The slip comes from the textbook's favourite exclusive pair, odd and even, which happens to cover the die.
Add them and see:
\[ P(C) + P(D) = \tfrac26 + \tfrac26 = \tfrac46 \]
The faces 1 and 6 are in neither event. A complement pair is a special exclusive pair that also covers everything, like odd and even. Every complement pair is exclusive; most exclusive pairs are not complements.
To show a complement pair you need two facts: no overlap, and the chances add to 1.
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, p. 175 — Example 3.8
Worked example
Mark takes one of two routes to work: the Interstate, I, with chance 0.44, or Fifth Street, F, with chance 0.56. He takes only one route, so the chance of both is zero. Find the chance of I OR F.
The addition rule from Section 3.3 subtracts the overlap so it is not counted twice. When events are mutually exclusive, the overlap is zero and there is nothing to subtract.
Write the addition rule
Why: always true, for any two events.
\[ P(I\text{ OR }F) = P(I) + P(F) - P(I\text{ AND }F) \]
Substitute, including the zero overlap
Why: he cannot take both routes.
\[ P(I\text{ OR }F) = 0.44 + 0.56 - 0 = 1 \]
In context: Mark certainly takes one of the two routes, which makes sense, since they are his only choices. Now the practice exercise: U and V are mutually exclusive, with chances 0.26 and 0.37.
Exercise 42(a): the overlap
Why: the definition of mutually exclusive.
\[ P(U\text{ AND }V) = 0 \]
Exercise 42(b): U given V
Why: the overlap over the chance of V.
\[ P(U \mid V) = \frac{0}{0.37} = 0 \]
Exercise 42(c): U OR V
Why: nothing to subtract.
\[ P(U\text{ OR }V) = 0.26 + 0.37 = 0.63 \]
Check the room left over
Why: 0.37 of the time neither happens, which is fine, since U and V are exclusive but not complements.
\[ 1 - 0.63 = 0.37 \ge 0 \]
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, p. 178 and practice, p. 204 — Try It 3.12, Exercise 42
Anomaly
Someone tells you that event A has probability 0.6, event B has probability 0.5, and the two are mutually exclusive. Before you believe them, add the two numbers.
Predict first
Can events with probabilities 0.6 and 0.5 be mutually exclusive?
Correct: no, they must share at least 0.1 of probability
Why: If they shared nothing, the outcomes in A or B would have total probability 0.6 plus 0.5, which is 1.1, more than the whole sample space. So at least 0.1 of probability must sit in both events: they overlap and are not exclusive.
Think of the sample space as a strip of length 1. Lay A down as a piece of length 0.6. There is only 0.4 of strip left, so a piece B of length 0.5 cannot fit without covering at least 0.1 of A.
\[ P(A\text{ AND }B) \ge P(A) + P(B) - 1 = 0.6 + 0.5 - 1 = 0.1 \]
This gives you a quick check on any claim of exclusivity. Exclusive events can have chances that add to at most 1, as Mark's two routes did exactly. If the chances add to more than 1, the claim is impossible, whatever the story behind the events.
The inequality comes from the addition rule of the next section, read with the fact that no probability exceeds 1. You will use it there as a free test on data that look suspicious.
Section
Part 4
Intuition
It is tempting to think mutually exclusive events must be independent, since they have nothing in common. The truth is the reverse. Suppose A and B are mutually exclusive and each has a positive chance.
\[ P(A\text{ AND }B) = 0, \quad P(A) > 0, \quad P(B) > 0 \]
Learn that A happened. Because the two events cannot happen together, you now know for certain that B did not. Your estimate for B drops from something positive all the way to zero.
\[ P(B \mid A) = \frac{P(A\text{ AND }B)}{P(A)} = \frac{0}{P(A)} = 0 \ne P(B) \]
That is the definition of dependence, and as strong as dependence gets: the news did not nudge the chance of B, it settled it. The product test agrees, since the product of two positive chances is positive while the overlap is zero.
\[ P(A)\,P(B) > 0 = P(A\text{ AND }B) \]
So the two properties are close to opposites. Mutually exclusive events with positive probabilities are never independent. The only loophole is an event with probability zero, which is trivially both, and which never appears in a real problem.
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, pp. 172-174 — the two definitions combined
Picture it
Here are three ways an event B of probability 0.3 can sit relative to an event A of probability 0.4, drawn in the unit square. A is always the blue strip on the left.
Figure (svg): Three unit squares, each with a blue strip A of width 0.4 on the left. Left square, 'mutually exclusive': an orange block B of area 0.3 sits entirely outside the strip. Middle square, 'independent': B is a horizontal band of height 0.3 running across the whole square. Right square, 'dependent, overlapping': B sits mostly inside the strip.
On the left, B lies entirely outside A: mutually exclusive. Learn that A happened and you know you are in the blue strip, where there is no B at all. The chance of B given A is zero.
In the middle, B is a horizontal band across the whole square. Whichever strip you are told you are in, B fills 30 percent of it. That is independence, and notice that the band has to cross A to manage it. Independent events with positive chances always overlap.
\[ \text{independent: } P(A\text{ AND }B) = (0.4)(0.3) = 0.12 > 0 \]
On the right, B is crowded into A, so news of A makes B much more likely: 0.24 of A's 0.4 is B, a conditional chance of 0.6. Dependence can push the chance up as well as down. Only the middle arrangement is independent; the exclusive one is dependence in its most extreme downward form.
Worked example
Flip a fair coin, then roll a fair die. Let A be heads followed by an even number, and B heads followed by a three. Are A and B mutually exclusive? Are they independent?
Count the outcomes
Why: two coin faces times six die faces, each pairing equally likely.
\[ 2 \times 6 = 12 \]
Figure (svg): A grid of 12 outcomes, two rows (coin T, coin H) by six columns (die 1 to 6). In the H row, cells H2, H4 and H6 are shaded blue as A, and cell H3 is shaded orange as B. The T row is unshaded.
List A and find its chance
Why: heads, then 2, 4 or 6.
\[ A = \{H2, H4, H6\}, \quad P(A) = \frac{3}{12} \]
List B and find its chance
Why: heads, then 3.
\[ B = \{H3\}, \quad P(B) = \frac{1}{12} \]
Test for mutual exclusivity
Why: no outcome is in both lists.
\[ P(A\text{ AND }B) = 0 \]
Test for independence
Why: the product of the two chances is not zero. The book's solution prints this line garbled, as 0.P(A)P(B).
\[ P(A)\,P(B) = \frac{3}{12} \cdot \frac{1}{12} = \frac{3}{144} = \frac{1}{48} \ne 0 \]
So A and B are mutually exclusive and dependent, exactly as the argument on the last two slides predicted. In context: if you know the coin and die showed heads and an even number, a heads-and-three outcome is ruled out entirely.
Compare this with the coin and die themselves. The coin flip and the die roll are physically separate, so heads on the coin and an even number on the die are independent. A and B are different: both are built on heads, and they compete for the same coin result with different die faces.
Check with the conditional test
Why: B given A must be zero, which differs from the chance of B.
\[ P(B \mid A) = \frac{0}{3/12} = 0 \ne \frac{1}{12} \]
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, pp. 178-179 — Example 3.13
Worked example
The section's practice exercises are short, and each one turns on using a definition as an equation. Write the definition first and the answer drops out.
Exercise 40: E and F are mutually exclusive, with chances 0.4 and 0.5; find E given F
Why: exclusive means the overlap is zero.
\[ P(E \mid F) = \frac{P(E\text{ AND }F)}{P(F)} = \frac{0}{0.5} = 0 \]
Exercise 40 is the dependence argument with numbers: the chance of E drops from 0.4 to zero once you learn F. Exercise 41 uses the other definition.
In Exercise 41 you are not told the chance of J directly, only the conditional. Independence is the bridge: it says the conditional and the plain chance are the same number, so knowing one is knowing the other.
Exercise 41: J and K are independent and J given K is 0.3; find the chance of J
Why: independence says the conditional equals the plain chance.
\[ P(J) = P(J \mid K) = 0.3 \]
Exercise 43: Q and R are independent, Q has chance 0.4, both 0.1; find R
Why: the product condition is an equation with one unknown.
\[ P(Q\text{ AND }R) = P(Q)\,P(R) \Rightarrow 0.1 = 0.4\,P(R) \]
Divide both sides by 0.4
Why: solve for the unknown.
\[ P(R) = \frac{0.1}{0.4} = 0.25 \]
Notice how Exercise 43 used a test as a tool. Once independence is given, the product condition is a fact you can solve with, not just something to check.
Check Exercise 43 with a conditional
Why: R given Q must equal the plain chance of R if they are independent.
\[ P(R \mid Q) = \frac{0.1}{0.4} = 0.25 = P(R) \]
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2 practice, p. 204 — Exercises 40, 41 and 43
Counterexample
Someone claims that independent events can never happen together, because independent means the events have nothing to do with each other. Try It 3.13 is the counterexample.
Figure (svg): A 2 by 2 grid of two draws with replacement. Rows are the first draw, W then R; columns are the second draw, W then R. The top row WW, WR is shaded blue as F, white first. The left column WW, RW is ringed orange as S, white second. The cell WW is labelled T, white twice.
Discussion prompt
A box holds one white and one red ball; you draw, replace, and draw again. Let F be white first and S white second. Show with numbers that F and S are independent and that they can happen together.
The root of the error is the everyday phrase nothing to do with each other. In probability it should mean no information, and two events can carry no information about each other while still happening at the same moment, like sunshine today in two distant cities.
Keep this pair as your standard example of independent and overlapping, just as the coin-then-die events are your standard example of exclusive and dependent.
A one-line rule settles every claim of this kind: independent events with positive chances always overlap, because their overlap is the product of two positive numbers, and a product of positive numbers is never zero.
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, p. 179 — Try It 3.13
Worked example
Same box, same two draws with replacement. Let T be white twice. The Try It asks for the chance of T, the chance of T given F, and whether T and F are independent.
(a) White twice
Why: one outcome of four.
\[ P(T) = P(\{WW\}) = \tfrac14 \]
(b) T given F
Why: once the first draw is white, only WW and WR remain, and one of them is WW.
\[ P(T \mid F) = \tfrac12 \]
(c) Compare with the plain chance of T
Why: the news raised the chance from a quarter to a half.
\[ P(T \mid F) = \tfrac12 \ne \tfrac14 = P(T) \]
So T and F are dependent, even though every draw is independent of every other. That is no contradiction: T is about both draws and already contains F's information, since white twice requires white first.
(d) F and S: is the overlap empty?
Why: WW is in both.
\[ P(F\text{ AND }S) = \tfrac14 \ne 0 \]
(e) F and S: run the product test
Why: two halves multiply to the quarter overlap.
\[ P(F)\,P(S) = \tfrac14 = P(F\text{ AND }S) \]
In context: the first draw tells you nothing about the second, so F and S are independent; but it tells you a lot about whether both are white.
Check (c) with the product test
Why: independence of T and F would need the overlap to be one eighth, but T lies inside F, so the overlap is T itself, one quarter.
\[ P(T\text{ AND }F) = \tfrac14 \ne \tfrac14 \cdot \tfrac12 = \tfrac18 \]
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, p. 179 — Try It 3.13
Matching
For each pair of events from the lesson, two questions: do they overlap, and does one change the chance of the other? Match each pair to its combination.
Match the pairs
Why: Green and odd share G1 and G3, a fifth of the bag, which equals four tenths times one half, so they are independent. Heads-even and heads-three share nothing and leave nine of twelve outcomes uncovered. Woman and long hair overlap in 0.45 against a product of 0.30. Odd and even faces are complements.
The marble pair came from the fill-in slide in Part 1: green and odd share two marbles, a fifth of the bag, which is exactly the product of four tenths and one half. The two exclusive pairs differ only in whether they cover everything: heads-even and heads-three leave nine outcomes uncovered, odd and even faces leave none.
The long-hair pair is the ordinary case: the events overlap, and the overlap is far from the product, 0.45 against 0.30. Most pairs of real events are like this, which is exactly why the book's default is to assume dependence.
Notice there is no match for exclusive and independent. With positive probabilities, that combination cannot exist, and the four rows here are the only shapes a pair of events can take.
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, pp. 175-178 — Try It 3.9, Examples 3.8, 3.12, 3.13
Error analysis
Here is an argument a classmate hands in for Example 3.8's events C, odd above two, and D, even below five. Every line looks like algebra. Find the line that breaks.
Annotate
On: \( P(C\text{ AND }D) = 0 = P(C)P(D) \Rightarrow \text{independent} \)
\[ P(C)\,P(D) = \tfrac26 \cdot \tfrac26 = \tfrac19 \ne 0 \]
The classmate probably reasoned that since the events share nothing, every number connected with them must be zero. But the product of the two chances is not about sharing; it is the overlap that independence would produce, and here independence would predict a one-in-nine overlap that is not there.
Run the conditional test as a check: given an odd face above two, the chance of an even face below five is zero, while its plain chance is a third. The news changed the chance completely.
When you mark an argument like this, do not stop at the wrong conclusion. Find the first false line, say what it should have been, and say what the corrected line shows. Here the corrected line, zero against one ninth, proves the opposite of the classmate's claim.
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, p. 175 — Example 3.8
Section
Part 5
Concept
The book gives two rules for what to assume when a problem does not tell you the relationship between two events.
Both defaults point the same way: assume the less convenient case. Independence and exclusivity each let you drop a term from a general rule, multiplying plain chances or skipping a subtraction. A default of dependent and overlapping means you never drop a term without a reason.
These are rules about how to work, not claims about the world. Assuming dependence does not say the events are related; it says you will use the general formula, which is right either way. If the events turn out independent, the general formula gives the same answer as the shortcut.
\[ P(A\text{ AND }B) = P(A)\,P(B \mid A) \quad \text{is always true} \]
The asymmetry is the point. Using the general rule on independent events costs nothing. Using the shortcut on dependent events gives a wrong answer, as the long-hair trap did when it predicted 30 women with long hair instead of 45.
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, pp. 173-174 — the two default assumptions
Step zero
Before any arithmetic, decide which test the information allows. In Example 3.12 you are given the chance of W, of L, of W AND L, and of L given W.
Discussion prompt
Of the three independence conditions, which can you check directly with those four numbers, and which would need an extra step? Say why.
This is exactly the book's note: the choice depends on the information you have. The step-zero habit is to list what you are given, then pick the test whose ingredients are all on the list.
Notice the extra condition still works once computed. Nine tenths of the long-haired students are women, against six tenths of the class: the news of long hair moves the chance of woman, which is the same dependence seen from the other side.
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, p. 178 — Example 3.12, note
Real world
A hospital has a main power supply and a backup generator. Each fails on a given day with probability 0.01. A report says both fail together with probability 0.01 times 0.01, one in ten thousand.
Discussion prompt
What assumption is hidden in the report's calculation? Give one reason it might be false for a generator in the same building, and say which way the error goes.
This is the default assumption doing real work. The general rule is the chance the main supply fails times the chance the generator fails given that the main supply failed. Without evidence that the second factor equals 0.01, you may not replace it with 0.01.
\[ P(\text{both}) = P(\text{main})\,P(\text{gen} \mid \text{main}) \ge (0.01)(0.01) \]
Engineers who want the product to be honest make the systems as close to independent as they can: a separate building, a separate fuel supply, a different maker. Each change reduces the shared causes that make the conditional bigger than the plain chance.
Ranking
Here is the routine the worked examples have been following for any pair of events. Put its steps in the order you would carry them out on a new problem.
Put in order
Order the steps for deciding how two events are related.
Why: Everything depends on translating the problem first. The AND probability serves both questions, so find it early. Exclusivity is a single comparison with zero; if it holds and both events have positive chances, they are dependent with no further work. Otherwise pick the cheapest independence test, compare, and say both answers in context.
The step people skip is the first. The words at most one tail or of the women students hide a set or a conditional, and nothing can be tested until they are written out.
The step people forget is the last. Both questions always have answers, and the problem usually wants both: independent and not exclusive, or exclusive and therefore dependent, each with the number that decides it.
Why does the exclusivity test come before the independence test? Because a yes answers both questions at once. If the overlap is zero and both chances are positive, the events are dependent, and the next two steps can be skipped.
Concept
The routine fits in one picture. Ask the overlap question first, because a yes ends the whole story.
Figure (svg): A flow diagram. A box 'two events A and B' leads to the question 'is P(A AND B) = 0?'. The yes arrow leads to 'mutually exclusive, and so dependent (if both have positive probability)'. The no arrow leads to 'not mutually exclusive', then to a second question 'does P(A given B) = P(A)?', whose yes arrow leads to 'independent' and whose no arrow leads to 'dependent'.
If the AND probability is zero, the events are mutually exclusive, and if both have positive chances they are automatically dependent. You do not need a separate independence test for an exclusive pair.
If the AND probability is not zero, the events are not exclusive, and the independence question is still open. Then you run one of the three tests: compare a conditional with a plain chance, or the overlap with the product. That gives either branch at the bottom.
The diagram also shows why exactly three combinations can appear: exclusive and dependent, overlapping and independent, overlapping and dependent. The fourth box, exclusive and independent, is missing because no path leads to it.
Use the diagram as a script when you write up an answer. Two sentences, in this order: first the overlap and the verdict on exclusivity, then the independence test and its verdict, each with the number that decides it.
\[ \text{exclusive} \Rightarrow P(B \mid A) = 0 \Rightarrow \text{dependent} \]
OpenStax Introductory Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2, pp. 172-174 — the two definitions
Elimination
You are told that P(A) = 0.5, P(B) = 0.4 and P(A AND B) = 0.2. Four conclusions are offered. Knock out every one that a number contradicts.
Eliminate the wrong options
Which conclusion is supported by the numbers?
Survives elimination: k
Why: Product test: one half times four tenths is two tenths, exactly the overlap, so the events are independent. Exclusivity test: the overlap is two tenths, not zero, so they are not mutually exclusive.
Each wrong conclusion makes a different mistake. One mixes the two properties up; one treats overlap as if it were dependence; one forgets that the product test needs no conditional at all.
The quickest way to these verdicts is the order in the diagram. Is the overlap zero? No, it is 0.2. Then does the overlap equal the product? Yes. Two numbers, two verdicts.
Translate the verdict into a conditional to see what it means: the chance of A given B is 0.2 over 0.4, which is one half, the same as the plain chance of A. Learning B happened leaves A exactly where it was, even though both can and do happen together a fifth of the time.
Section
Part 6
Pattern
Everything in this section fits in one table. Each row is one of the two relationships, with the test that decides it and what you assume when you cannot test it.
| property | the question | the test (any one) | default if unknown |
|---|---|---|---|
| independent | does news of one change the other's chance? | P(A given B) = P(A), P(B given A) = P(B), or P(A AND B) = P(A)P(B) | assume dependent |
| mutually exclusive | can they happen together? | P(A AND B) = 0, no shared outcome | assume not exclusive |
The two properties answer different questions, so a full answer always has two verdicts, each backed by a number. And when nothing settles them, the defaults keep you on the general rules, which are never wrong.
And one link between the rows that is worth memorising: an exclusive pair with positive chances is always dependent. So of the four possible combinations of verdicts, only three ever occur, and the missing one, exclusive and independent, is the most common wrong answer on a test.
Check
One question on independence. Choose the test your numbers allow before computing anything.
Check your understanding
P(A) = 0.5, P(B) = 0.6 and P(A AND B) = 0.36. Which statement is correct?
Answer: A
Why: The product test multiplies the two single chances: one half times six tenths is three tenths. The given overlap is 0.36, which is different, so the events are dependent. In fact the chance of A given B is 0.36 over 0.6, which is 0.6, higher than 0.5.
The product test needs exactly the three numbers you were given, so it is the natural choice. Multiply the two single chances, compare with the overlap, and the verdict follows.
If you chose the option with 0.6 squared, check which two numbers the condition multiplies. It is the chance of A times the chance of B, one of each event, never one event's chance twice.
Read the verdict as a sentence. A given B is 0.6, higher than the plain 0.5, so learning that B happened makes A more likely. Dependence has a direction, and here it is upward.
Check
Now a draw without replacement. Update the population after the first draw.
Check your understanding
A bag holds 4 red and 6 blue marbles. Two are drawn without replacement. What is the probability that both are red?
Answer: A
Why: After a red marble is removed, three red remain among nine marbles, so the second factor is three ninths. Four tenths times three ninths is twelve ninetieths, which is two fifteenths, about 0.133.
Without replacement, the second factor is a conditional: the chance of red given that one red marble is already gone. Both the count of red and the total drop by one.
Compare with replacement, where the answer would be four twenty-fifths, 0.16. Taking a red marble out makes a second red less likely, which is why the without-replacement answer is smaller.
A bounds check catches the remaining options. A chance of both must be smaller than the chance of either, four tenths, and adding two chances can only make the number bigger, so the sum cannot be right.
Check
Two separate verdicts again. Run the overlap test, then the independence test.
Check your understanding
Roll one fair die. A = the face is 1 or 2, and B = the face is 5 or 6. Which statement is true?
Answer: A
Why: No face is in both events, so the overlap is zero and they are mutually exclusive. Given A, the chance of B drops from one third to zero, so they are dependent, as every exclusive pair with positive chances must be.
The overlap test first: A is 1 and 2, B is 5 and 6, nothing shared. That makes them exclusive, and with each having a chance of one third, dependence follows at once.
If you picked exclusive and independent, the product test is the cure: one third times one third is one ninth, and the overlap is zero. They do not match.
Say it in words to make it stick: if the die shows 1 or 2, there is no way it shows 5 or 6. That certainty is information, and information is exactly what independence forbids.
Explain it to yourself
Independence and mutual exclusivity are the two properties this lesson keeps apart. Explain them so that someone who confuses them would stop.
Discussion prompt
In two or three sentences, say what each property means, what number tests each one, and why mutually exclusive events with positive chances cannot be independent.
A strong answer names a different number for each property: a conditional or a product for independence, the overlap for exclusivity. If your two explanations used the same test, they are describing one property twice.
Test your explanation on a fresh pair: one card is drawn, A is a heart, B is a queen. Could someone using your words decide both questions? The queen of hearts is in both, so not exclusive; one in four queens is a heart, the same as one in four cards, so independent.
\[ P(\text{heart} \mid \text{queen}) = \tfrac14 = \tfrac{13}{52} = P(\text{heart}) \]
Exit ticket
At a school, 30 percent of students play an instrument, 20 percent play a sport, and 6 percent do both. A student is chosen at random.
Discussion prompt
Are playing an instrument and playing a sport independent? Are they mutually exclusive? Give the number that decides each, and say what each verdict means for the students.
Write the three givens in notation first. The overlap is given directly, so the product test is a single multiplication and the exclusivity test is a single comparison with zero.
If either verdict felt uncertain, go back to the part of the lesson it belongs to: Part 1 for the three tests, Part 3 for the overlap, and Part 4 for why the two cannot both hold here.
Recap
Two properties, two questions, and two defaults. Independence is about information; mutual exclusivity is about overlap. A pair of events gets a verdict on each, and each verdict rests on one number.
| independent | mutually exclusive | |
|---|---|---|
| meaning | news of one leaves the other's chance unchanged | they cannot happen together |
| test | P(A given B) = P(A), or P(A AND B) = P(A)P(B) | P(A AND B) = 0 |
| typical source | draws with replacement, separate coins or dice | one route, one face, one card |
| default if unknown | assume dependent | assume not exclusive |
\[ \text{exclusive and } P(A), P(B) > 0 \Rightarrow P(B \mid A) = 0 \ne P(B) \]
Every verdict in this lesson came from the same short routine: translate, find the overlap, test exclusivity, then test independence with whichever condition your numbers allow, and say both answers in context.
Sampling with replacement keeps the population fixed and makes draws independent; without replacement every draw changes what is left, though in a big population hardly at all. Next, Section 3.3 turns these properties into the multiplication and addition rules, where each one lets you delete a term.
OpenStax Introductory Business Statistics 2e, §3.2 Independent and Mutually Exclusive Events §3.2 — the same definitions with business examples
Want this taught 1-on-1? Alexander tutors Statistics — $55/session, free consultation.