The vocabulary that chapters 4 through 13 are written in. An experiment is a planned operation whose result is not predetermined; an outcome is one result, the sample space is the set of all of them, and an event is any combination of outcomes. A probability is the long-term relative frequency of an outcome, always between zero and one, and when the outcomes are equally likely it can be computed by counting. The OR event is the union and the AND event is the intersection; the complement of A is everything not in A, and the two probabilities sum to one. Conditional probability is the one genuinely new idea: it reduces the sample space, and the probability of A given B is computed inside B alone.
Subject: Statistics · 66 slides · symbolic lesson
Open the interactive version of this deck
Title
Statistics · Chapter 3 — Probability Topics
Sample spaces, events, the long run, AND, OR, NOT, and what changes when you are given news
Objectives
Chapters 4 to 13 are written in the language of probability. Every distribution, every confidence interval and every hypothesis test is a sentence about the probability of some event. This section builds that language from the ground up, and it does it with counting, so every rule you meet can be checked by listing outcomes and counting them.
One habit runs through the whole lesson. Before you compute anything, write down the sample space, or at least say what it is, and name the event you want as a set of outcomes. Most wrong answers in probability come from a sample space that was never written down, not from bad arithmetic.
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, p. 168 — Terminology
Warm-up
Chapter 1 defined the relative frequency of a value: how often it occurs, divided by the total number of observations. Probability is built directly on top of that idea, so make sure it is solid.
Discussion prompt
In a class of 30 students, 12 rode a bus in the past month. What is the relative frequency of bus riders? Then reduce the fraction twelve thirtieths to lowest terms and write it as a decimal.
Keep both readings of that 0.4 in mind. As a description of the class it is a relative frequency: 12 of 30. As a prediction about one student picked at random from the class, it is a probability: the chance that the person you pick rode a bus.
That shift, from describing data you have to predicting a result you have not seen yet, is exactly what this chapter is about. The arithmetic stays the same: a count of the cases you care about, divided by a count of all the cases.
OpenStax Introductory Statistics 2e, §3.1 Terminology §3 introduction, pp. 167-168 — the collaborative exercise
Section
Part 1
Concept
Probability talks about activities whose result you cannot know in advance. The book calls such an activity an experiment, and its results outcomes.
experiment — A planned operation carried out under controlled conditions. If the result is not predetermined, it is a chance experiment. Flipping one fair coin twice is an experiment.
outcome — One possible result of an experiment. Flipping a coin once has two outcomes, heads and tails.
The word experiment here is wider than in a laboratory. Rolling a die, drawing a card, picking a student at random from a class list, spinning a roulette wheel and recording tomorrow's weather are all experiments. What makes them experiments is that you could describe the procedure precisely and that the result is uncertain.
The word controlled matters too. Flip the coin the same way each time, roll the die on the same table, choose the name with the same random method. When the conditions are fixed, the experiment can be repeated, and repetition is what will give probability its meaning in Part 2.
A good first question for any probability problem is: what exactly is the experiment, and what does one outcome look like? For two coin flips, one outcome is not heads; it is a pair, such as heads then tails. Getting the shape of an outcome right is half the battle, as the dime-and-nickel example will show.
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, p. 168 — experiment, chance experiment, outcome
Concept
Once you know what one outcome looks like, collect all of them. That collection has a name and a letter.
sample space — The set of all possible outcomes of an experiment, written S. For one coin flip, S is the set containing H and T.
\[ S = \{1, 2, 3, 4, 5, 6\} \quad \text{(one roll of a die)} \]
The book lists three ways to represent a sample space: write out the list of outcomes in braces, draw a tree diagram, or draw a Venn diagram. In this section you will use lists, grids and Venn diagrams; trees arrive in Section 3.5.
event — Any combination of outcomes, that is, any set of outcomes taken from the sample space. Events are named with capital letters such as A, B or E, and the probability of event A is written P(A).
Figure (svg): Six square cells numbered 1 to 6, the faces of one die, under the heading 'S, the sample space'. Cells 5 and 6 are shaded yellow and bracketed underneath with the label 'E: at least five, 2 of the 6 outcomes'.
\[ E = \{5, 6\} \quad \text{(at least five)} \]
Notice the difference in size. An outcome is one result; an event can hold one outcome, several, all of them or none. Rolling at least five is an event with two outcomes in it. Rolling a 6 is an event with exactly one. Any yes-or-no question you can ask about the result picks out an event.
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, p. 168 — sample space and event
Notation
When every outcome in the sample space is equally likely, the book gives a rule for computing the probability of any event: count and divide. Each piece of the formula has a job.
Annotate
On: \( P(A) = \frac{\text{number of outcomes in } A}{\text{number of outcomes in } S} \)
Why does dividing give the probability? If six faces share the certainty of 1 equally, each face carries one sixth. An event with two faces collects two of those shares, which is two sixths. Count over total is just a quick way of adding equal shares.
\[ P(E) = \frac{1}{6} + \frac{1}{6} = \frac{2}{6} = \frac{1}{3} \]
The condition is the part people forget. The formula is a consequence of equal shares, so when outcomes are not equally likely, as with a loaded die or tomorrow's weather, it gives a wrong answer. The trap later in this part shows how easily that happens.
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, p. 168 — equally likely outcomes
Worked example
Toss a fair dime and a fair nickel. Find the probability of event A, getting exactly one head.
The coins are different coins, so an outcome has to record what each coin did. Write the dime's face first and the nickel's second. A grid with one row per dime face and one column per nickel face makes sure every pairing appears exactly once.
Build the sample space
Why: two dime faces times two nickel faces gives four pairs.
\[ S = \{HH, HT, TH, TT\} \]
Figure (svg): A two-by-two grid. Rows are the dime, H or T; columns are the nickel, H or T. The cells read HH, HT, TH, TT. The cells HT and TH are shaded cyan; a note at the right reads 'exactly one head: 2 of 4 cells'.
Count the sample space
Why: four cells, and the coins are fair, so the four are equally likely.
\[ \text{outcomes in } S = 4 \]
Write the event as a set
Why: one head and one tail, in either order.
\[ A = \{HT, TH\} \]
Count the event
Why: two cells are shaded.
\[ \text{outcomes in } A = 2 \]
Divide
Why: equally likely outcomes, so count over total.
\[ P(A) = \frac{2}{4} = 0.5 \]
In context: when you toss a fair dime and a fair nickel, you get exactly one head half the time in the long run.
Check by counting the rest
Why: the cells outside A must hold the other half, and they do: both heads and both tails, two cells of four.
\[ \{HH, TT\}: \frac{2}{4}, \qquad 0.5 + 0.5 = 1 \]
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, p. 168 — the dime and nickel
Trap
Asked for the chance of exactly one head in two tosses, a common first line lists the possible numbers of heads:
\[ \{0, 1, 2\} \text{ heads} \]
\[ P(\text{one head}) = \frac{1}{3} \]
Wrong. The list is a correct list of what can happen, but its three items are not equally likely, so counting them is not allowed.
Figure (svg): A bar chart of the number of heads in two coin tosses. The bar at 0 heads has height 0.25, at 1 head 0.5, at 2 heads 0.25. A dashed red line across at one third is labelled 'if the counts were equally likely'.
One head can happen in two ways, a head on the dime or a head on the nickel, while no heads and two heads each happen in only one way. List the equally likely outcomes, then count:
\[ \{HH, HT, TH, TT\} \]
\[ P(\text{one head}) = \frac{2}{4} = 0.5 \]
The rule to remember: count over total is valid only on a sample space whose outcomes are equally likely. When in doubt, describe each coin separately, as the grid does, and the outcomes will be equally likely by symmetry.
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, p. 168 — the dime and nickel
Matching
Situation: a teacher picks one student at random from a class list of 30 and records whether that student rode a bus last month. Match each term to its part of the situation.
Match the pairs
Why: The experiment is the whole procedure, done once. One outcome is one specific result of doing it, here one particular student. The sample space collects every possible result, the full list of 30. An event is a set of outcomes picked out by a condition, here the 12 bus riders.
The pair most people mix up is outcome and event. An outcome is a single result; an event is a set, which may contain many outcomes. The bus-rider event contains twelve outcomes, and it happens whenever any one of them is the name drawn.
Once the four words are attached, the probability is immediate. The students are equally likely to be picked, so the chance of a bus rider is 12 out of 30, which is 0.4: the relative frequency from the warm-up, now read as a prediction.
If you matched the class list to the experiment, notice the difference between the procedure and the things it can produce. The experiment is something you do; the sample space is the menu of what can come out of doing it.
Prediction
Tomorrow it will either rain or not rain. That is a sample space with exactly two outcomes. Before reading on, decide what follows.
Predict first
Since tomorrow has exactly two possible outcomes, rain or no rain, the probability of rain is
Correct: not settled by the count, because the two outcomes need not be equally likely
Why: Count over total needs equally likely outcomes. Nothing makes rain and dry equally likely; in a desert the chance of rain is tiny, in a rainy season it is large. The real chance comes from long-run records of similar days, which is the subject of Part 2.
Two outcomes do not make a fifty-fifty chance. A lottery ticket either wins or loses, and nobody believes the chance of winning is one half. The count of outcomes tells you how many things can happen; it says nothing about how likely each one is.
Ask what would make two outcomes equally likely. For a fair coin, it is physical symmetry: the two faces are interchangeable. For a true or false question answered by a pure guess, it is the guessing. When no symmetry or randomisation forces equal chances, you need data instead, and that is where relative frequency comes in.
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, p. 168 — equally likely
Worked example
A number is selected at random from 1 to 10. Describe the sample space, write each event as a set and compute its probability.
Write the sample space
Why: ten numbers, each equally likely because the pick is random.
\[ S = \{1, 2, 3, 4, 5, 6, 7, 8, 9, 10\} \]
Figure (svg): The numbers 1 to 10 in boxes. A cyan dot in the row 'A: more than 6' sits under 7, 8, 9, 10; an orange dot in the row 'B: odd' sits under 1, 3, 5, 7, 9. Boxes 7 and 9 are outlined in yellow because they carry both dots.
(a) F: the outcome is 5
Why: a single outcome is still an event.
\[ F = \{5\}, \quad P(F) = \frac{1}{10} \]
(b) A: more than 6
Why: more than excludes 6 itself.
\[ A = \{7, 8, 9, 10\}, \quad P(A) = \frac{4}{10} \]
(c) B: the number is odd
Why: every second number, starting at 1.
\[ B = \{1, 3, 5, 7, 9\}, \quad P(B) = \frac{5}{10} \]
(d) the complement of B
Why: everything in S that is not odd.
\[ B' = \{2, 4, 6, 8, 10\}, \quad P(B') = \frac{5}{10} \]
Parts (i) and (j) need two definitions from arithmetic. A composite number has a factor other than 1 and itself; 1 is neither prime nor composite, and the primes up to 10 are 2, 3, 5 and 7.
(i) P: a composite number
Why: the numbers that are neither 1 nor prime.
\[ P = \{4, 6, 8, 9, 10\}, \quad P(P) = \frac{5}{10} \]
(j) M: a multiple of 3
Why: three, six and nine.
\[ M = \{3, 6, 9\}, \quad P(M) = \frac{3}{10} \]
Check the prime and composite counts
Why: 1, the four primes and the five composites must account for all ten numbers exactly once.
\[ 1 + 4 + 5 = 10 \]
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, p. 171 — Try It 3.2
Section
Part 2
Concept
Counting equally likely outcomes is a way to compute some probabilities. It is not what a probability is. The book's definition is about repetition.
probability — The probability of an outcome is its long-term relative frequency: the share of times it occurs when the experiment is repeated a very large number of times.
Because a relative frequency is a count divided by a larger or equal count, a probability can never be negative and never exceed one. The book states the range, inclusive of both ends.
\[ 0 \le P(A) \le 1 \]
Figure (svg): A horizontal number line from 0 to 1 with ticks at quarter steps. Marked points: 0 'never happens, a 7 on one die'; one third 'at least five on one die'; 0.5 'as likely as not, heads on a fair coin'; 0.56 'euro coin heads, observed'; 1 'always happens, a face from 1 to 6'.
The landmarks on the scale have plain meanings. A probability of 0 means the event can never happen, like rolling a 7 on an ordinary die. A probability of 1 means it always happens. A probability of 0.5 means the event is equally likely to occur or not to occur, like heads on a fair coin.
The definition also explains why counting works when it is allowed. Roll a fair die thousands of times and each face turns up about one sixth of the time, because nothing favours any face. So an event made of two faces turns up about two sixths of the time, which is exactly what the counting formula said.
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, p. 168 — long-term relative frequency
Picture it
Here is the definition happening. A computer tossed a fair coin 20,000 times and, after every toss, recorded the share of heads so far. The horizontal axis counts tosses, with each step ten times the one before, so the early tosses get room to show their behaviour.
Figure (svg): A line graph of the running proportion of heads in 20,000 simulated tosses of a fair coin, with the number of tosses on a logarithmic axis from 1 to 20,000. The line swings widely at first (0.2 after 10 tosses), passes 0.48 at 100, 0.498 at 1,000 and 0.503 at 10,000, hugging a dashed horizontal line at 0.5.
After 10 tosses this coin had shown only 2 heads, a share of 0.2. Judged on those ten tosses alone, you might call it badly biased. Yet the coin is fair by construction: the computer gives heads exactly half the chance on every toss.
| tosses | heads | share of heads | distance from 0.5 |
|---|---|---|---|
| 10 | 2 | 0.200 | 0.300 |
| 100 | 48 | 0.480 | 0.020 |
| 1,000 | 498 | 0.498 | 0.002 |
| 10,000 | 5033 | 0.5033 | 0.0033 |
Read down the last column: the distance from one half shrinks from 0.3 to 0.02 to 0.002 to about 0.003. It does not shrink every single time, and it never needs to reach zero, but the trend is unmistakable. The long run pulls the relative frequency to the probability.
The book makes the same point with 20 to 2,000 to 20,000 flips. That settling-down is what the phrase long-term relative frequency means, and the next slide gives it its name.
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, p. 168 — flipping a fair coin repeatedly
Concept
The pattern in the coin picture is so dependable that it has a name, and the book states it carefully.
law of large numbers — As the number of repetitions of an experiment increases, the relative frequency obtained tends to become closer and closer to the theoretical probability.
Two words in the statement carry the weight. Tends means the closeness is a long-run trend, not a promise about every single run: a run of 2,000 flips can occasionally land further from one half than a run of 1,000. Theoretical means the probability computed from a model, such as one sixth for a face of a fair die.
The relative frequency you actually observe is called the empirical, or observed, probability. So there are two numbers in play whenever you repeat an experiment: one from the model, one from the data.
\[ \text{empirical} = \frac{\text{times } A \text{ occurred}}{\text{number of trials}} \longrightarrow P(A) \]
Notice what the law does not say. It does not say outcomes come in any pattern or order; the book stresses that they do not. Individual tosses stay unpredictable forever. Only the overall share becomes predictable, and only when the number of trials is large.
This is the law that makes statistics possible. A survey of a large random sample estimates a population proportion for exactly the reason that 10,000 coin tosses estimate the probability of heads.
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, p. 168 — the law of large numbers
Intuition
Whenever you can both model an experiment and repeat it, the same event gets two probabilities, and it pays to keep them apart in your head and on paper.
| theoretical probability | empirical probability | |
|---|---|---|
| where it comes from | a model: symmetry, design, counting | data: what actually happened in trials |
| heads on a fair coin | 0.5, from two equal faces | 0.48 after 100 simulated tosses |
| heads on the Belgian euro | unknown; 0.5 if it were fair | 0.56 after 250 trials |
| does it change with more trials? | no, it is fixed by the model | yes, it wobbles and settles |
The theoretical number is a claim about the experiment. The empirical number is a measurement of it. The law of large numbers is the bridge: as trials pile up, the measurement tends toward the claim, if the claim is right.
That last clause is where statistics begins. If a coin is really fair, its empirical share must drift toward 0.5. If after many trials it stubbornly sits somewhere else, the model was wrong, and the empirical number is telling you the real probability. Every hypothesis test in Chapter 9 is a careful version of that comparison.
So when a problem gives you a percentage, ask which kind it is. Fifty-six percent of 250 euro trials is empirical. One sixth for a die face is theoretical. They are compared all the time, but they are never the same kind of number.
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, p. 168 — the word empirical
Worked example
Roll one fair die and let E be rolling at least five. The book computes the probability, then says you would expect about a third of a large number of rolls to land in E, but not exactly a third. Make that concrete with 600 rolls.
The theoretical probability
Why: two equally likely faces out of six.
\[ P(E) = \frac{2}{6} = \frac{1}{3} \]
Expected count in 600 rolls
Why: a third of the rolls, in the long-run sense.
\[ 600 \times \frac{1}{3} = 200 \]
Now suppose one actual run of 600 rolls gives 230 rolls of five or six. Is the die suspicious? Turn the count into a relative frequency so that it sits on the same scale as the probability.
The empirical probability
Why: observed count over number of trials.
\[ \frac{230}{600} \approx 0.383 \]
Compare with the theoretical value
Why: one third written as a decimal.
\[ 0.383 - 0.333 = 0.050 \]
The law of large numbers says this gap should tend to shrink as you keep rolling. If, after 60,000 rolls, the share were still near 0.38, you would have strong evidence that the die favours its high faces; after only 600, a gap of five hundredths is worth a second look but not a verdict.
Check by scaling back to counts
Why: each relative frequency times 600 must recover its count.
\[ 0.3833 \times 600 \approx 230, \qquad 0.3333 \times 600 \approx 200 \]
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, p. 168 — rolling a number that is at least five
Estimation
Roll a fair die 6,000 times and count the rolls that show five or six. Estimate before you read the answer.
Predict first
The count of fives and sixes in 6,000 rolls of a fair die will most likely be
Correct: somewhere near 2,000, typically within about 70 of it
Why: A third of 6,000 is 2,000, the long-run expectation. Hitting it exactly is unlikely, because a count wobbles around its expected value. For 6,000 rolls, the typical wobble is a few dozen rolls, and almost all runs land within about 70 of 2,000. That is a relative frequency between about 0.32 and 0.345.
Exactly 2,000 is the tempting answer, and it is the single most likely count, but it is still unlikely: there are so many possible counts near 2,000 that no single one of them gets much probability. The law of large numbers promises closeness of the share, not an exact hit on the count.
Notice how the two scales behave differently. Measured in rolls, the wobble grows as you roll more, roughly 37 rolls at 6,000. Measured as a share of all rolls, it shrinks: 37 is barely more than half a percent of 6,000. The law is a statement about the share.
\[ \frac{2000 \pm 70}{6000} \approx 0.333 \pm 0.012 \]
Later chapters will make typically within precise. For now, the habit to build is that an observed relative frequency is an estimate with some wobble, and the wobble gets smaller, as a share, when the number of trials gets larger.
Worked example
Two mathematics professors had their students test the Belgian one-euro coin. In 250 trials, heads came up 56 percent of the time and tails 44 percent. Turn those percentages into counts and weigh the evidence.
Count the heads
Why: 56 percent of the 250 trials.
\[ 0.56 \times 250 = 140 \]
Count the tails
Why: the rest of the trials.
\[ 250 - 140 = 110 \]
What a fair coin predicts
Why: half of 250 in the long run.
\[ 0.5 \times 250 = 125 \]
Figure (svg): Two bars: heads 140 and tails 110, out of 250 spins of a Belgian one-euro coin. A dashed yellow line across at 125 is labelled 'a fair coin's long-run share: 125 each'.
Measure the gap
Why: observed share minus the fair share.
\[ 0.56 - 0.50 = 0.06 \]
Is 0.06 a big gap for 250 trials? An exact calculation for a fair coin, which you will be able to do in Chapter 4, says it produces 140 or more heads in 250 trials only about 3 times in every 100 runs. So the euro's result is unusual for a fair coin, but not impossible.
\[ P(\text{140 or more heads in 250} \mid \text{fair coin}) \approx 0.033 \]
That is why the book's verdict is careful: the data seem to show the coin is not fair, and more repetitions would help. By the law of large numbers, a few thousand more trials would either pull the share back toward 0.5 or keep it near 0.56 and settle the matter.
Check the counts
Why: heads and tails must account for every trial, and their shares must add to one.
\[ 140 + 110 = 250, \qquad 0.56 + 0.44 = 1 \]
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, p. 168 — the Belgian one-euro coin
Trap
A fair coin has just landed heads seven times in a row. A common line of reasoning:
\[ \text{7 heads so far} \Rightarrow \text{tails is due} \]
\[ P(\text{tails next}) > 0.5 \]
Wrong. This is the gambler's fallacy. The coin has no memory of the seven heads and no way to make up for them.
It feels right because the law of large numbers promises the share of heads will approach one half, so it seems the coin must produce extra tails to get there.
The next toss of a fair coin is still fifty-fifty. The law works by dilution, not compensation: the seven early heads get swamped by thousands of later tosses that come out about evenly.
\[ \frac{7 + 500}{7 + 1000} \approx 0.503 \]
Seven heads followed by 1,000 perfectly ordinary tosses leaves the share at about 0.503, already close to one half, with no catching up at all. The streak does not vanish; it just stops mattering as a share.
The same fallacy fills casinos: a roulette wheel that has landed red eight times is no more likely to land black next.
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, p. 168 — outcomes do not happen according to any set pattern
Prediction
A small hospital has about 15 births a day and a large one about 45. At both, each baby is a boy with probability about one half. Each hospital counts the days on which more than 60 percent of the babies born were boys.
Predict first
Over a year, which hospital records more days on which over 60 percent of births were boys?
Correct: the small hospital
Why: By the law of large numbers, a relative frequency based on more trials tends to sit closer to the probability. Each day at the small hospital is a sample of 15 births, whose share of boys swings widely; at the large hospital it is a sample of 45, whose share stays nearer one half.
Figure (svg): Two bars: a small hospital with 15 births a day records a day of over 60 percent boys with probability 0.151; a large hospital with 45 births a day, 0.068.
The exact chances, from the counting methods of Chapter 4, are about 0.151 for the small hospital and 0.068 for the large one. Over 365 days that is roughly 55 lopsided days against 25. One extra boy moves a 15-birth day by almost 7 percentage points, but a 45-birth day by just over 2.
Most people guess about the same. The law of large numbers is not only a statement about the long run; it is also a warning about the short run. Small samples produce extreme relative frequencies far more often than large ones, which is why a small study's dramatic result deserves extra suspicion.
Concept
The counting formula needs equally likely outcomes, and the book warns that many real situations do not provide them. A coin or die may be unfair, or biased.
Look at the dice in a board game at home. The spots are usually small holes drilled out and painted. A face with six holes has lost more material than a face with one hole, so the die is very slightly lighter on the six side. That may tip the outcomes, ever so slightly, toward the heavier faces landing down.
Casinos cannot afford that. Casino dice have flat faces, and their holes are filled completely with paint of the same density as the die itself, so each face is equally likely. A casino earns its money from the small edge built into the rules of each game, and it does not want any extra edge for the players hidden in the dice.
| situation | equally likely? | how to get the probability |
|---|---|---|
| fair die, fair coin, random pick | yes, by symmetry or design | count over total |
| Belgian euro, homemade die | maybe not | long-run relative frequency from many trials |
| rain tomorrow, a team winning | no reason to think so | records of similar past cases |
For an unfair coin, the definition from earlier in this part still works when the counting formula fails. The probability of heads for the euro is whatever share of heads it settles toward over many trials. The book promises later techniques for unequal outcomes; in this section, whenever you count, say first why the outcomes are equally likely.
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, pp. 168-169 — biased coins and dice
Section
Part 3
Concept
Two events on the same sample space can be combined into new events. The first combination collects every outcome that belongs to at least one of them.
A OR B — An outcome is in the event A OR B if it is in A, or in B, or in both. The event is also called A union B.
\[ A \cup B = A\text{ OR }B \]
The book's example uses two small sets of numbers. Put every member of either set into one list, and write each member only once.
\[ A = \{1, 2, 3, 4, 5\}, \quad B = \{4, 5, 6, 7, 8\} \]
\[ A\text{ OR }B = \{1, 2, 3, 4, 5, 6, 7, 8\} \]
Notice that 4 and 5 are not listed twice. A set records which outcomes are in it, not how many times you found them, so an outcome that is in both events still counts as one outcome of the union. That one rule is behind almost every OR mistake.
The meaning of or here is the inclusive one, either or both, as in a sign saying applicants with a degree or experience may apply. Someone with both is certainly welcome. Everyday speech sometimes uses the exclusive one, soup or salad meaning one of the two; probability never does.
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, p. 169 — the OR event
Concept
The second combination is stricter. It keeps only the outcomes that satisfy both conditions at once.
A AND B — An outcome is in the event A AND B if it is in both A and B at the same time. The event is also called A intersection B.
\[ A \cap B = A\text{ AND }B \]
With the same two sets, look for the numbers that appear in both lists.
\[ A\text{ AND }B = \{1, 2, 3, 4, 5\} \cap \{4, 5, 6, 7, 8\} = \{4, 5\} \]
Compare the sizes. The union has eight members, the intersection two, and each of A and B has five. That ordering is no accident: the AND event sits inside each event, and each event sits inside the OR event.
\[ A\text{ AND }B \subseteq A \subseteq A\text{ OR }B \]
So an AND event can never be more likely than either event alone, and an OR event can never be less likely than either. Use that as a free check: if your AND probability is bigger than one of the single probabilities, something has gone wrong.
An AND event can also be empty. A die cannot show a face that is both 1 and 6, so that intersection has no outcomes and probability zero. Section 3.2 gives events like that a name.
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, p. 169 — the AND event
Notation
The book writes events in words and in symbols. You need to read both fluently, because problems switch between them without warning.
Annotate
On: \( A \cup B, \quad A \cap B, \quad A', \quad P(A \mid B) \)
Translate the words of a problem into these symbols as your first move. Both and also usually mean AND. Either, at least one of and or usually mean OR. Not, fails to and other than mean a complement. Given, among, of those and if it is known that mean a condition.
One caution about the word and in ordinary English. The students who play soccer and the students who play tennis can mean two separate groups put together, which is an OR, not the students who play both. When the wording is ambiguous, ask whether a single outcome has to satisfy both conditions at once.
The book itself mostly uses the words AND and OR in capitals, and this deck follows it. The symbols are worth knowing because other books and software use them.
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, p. 169 — union and intersection notation
Picture it
Example 3.1 uses the whole numbers from 1 to 19, with A the even numbers and B the numbers greater than 13. Before computing anything, see where every number goes.
Figure (svg): The numbers 1 to 19 in a row of boxes. Under each even number a cyan dot sits in the row labelled 'A: even'; under each number from 14 to 19 an orange dot sits in the row labelled 'B: above 13'. Boxes 14, 16 and 18 are outlined in yellow because they carry both dots.
The strip marks each number once for each event it belongs to. The Venn diagram below takes the same marks and sorts the numbers into regions, so that each number appears exactly once.
Figure (svg): A rectangle S holding two overlapping circles, A (even) and B (above 13). Only in A: 2, 4, 6, 8, 10, 12. In the overlap: 14, 16, 18. Only in B: 15, 17, 19. Outside both circles: 1, 3, 5, 7, 9, 11, 13.
Four regions, four kinds of outcome: in A only, in both, in B only, in neither. The OR event is the first three regions together; the AND event is the middle one. The fourth region, outside both circles, will be useful in Part 4.
The diagram also shows why adding the sizes of A and B overcounts the union. The circle for A holds nine numbers and the circle for B six, but the three numbers in the lens are inside both circles, so nine plus six counts them twice.
\[ 9 + 6 - 3 = 12 \]
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, p. 169 — Example 3.1
Worked example
The sample space is the whole numbers starting at one and less than 20. A is the even numbers, B the numbers greater than 13. A number is chosen at random, so all are equally likely.
(a) List the sample space
Why: starting at one, stopping before 20.
\[ S = \{1, 2, 3, \ldots, 19\}, \quad 19 \text{ outcomes} \]
(b) List A
Why: the even numbers up to 18.
\[ A = \{2, 4, 6, 8, 10, 12, 14, 16, 18\} \]
(b) List B
Why: greater than 13 starts at 14.
\[ B = \{14, 15, 16, 17, 18, 19\} \]
(c) Probabilities of A and B
Why: count each event, divide by 19.
\[ P(A) = \frac{9}{19}, \quad P(B) = \frac{6}{19} \]
(d) A AND B
Why: the even numbers greater than 13.
\[ A\text{ AND }B = \{14, 16, 18\} \]
(d) A OR B
Why: every number in either list, written once.
\[ A\text{ OR }B = \{2, 4, 6, 8, 10, 12, 14, 15, 16, 17, 18, 19\} \]
(e) Probability of AND
Why: three outcomes of nineteen.
\[ P(A\text{ AND }B) = \frac{3}{19} \approx 0.158 \]
(e) Probability of OR
Why: twelve outcomes of nineteen.
\[ P(A\text{ OR }B) = \frac{12}{19} \approx 0.632 \]
In context: if you pick a whole number from 1 to 19 at random, about 63 percent of the time it is even or bigger than 13, and about 16 percent of the time it is both.
Check the OR count against the overlap
Why: the union must hold the nine even numbers plus the six big ones, minus the three counted twice; and the AND must not exceed either event.
\[ 9 + 6 - 3 = 12, \qquad \frac{3}{19} \le \frac{6}{19} \le \frac{12}{19} \]
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, pp. 169-170 — Example 3.1(a)-(e)
Trap
For Example 3.1, two opposite slips are common. The first adds the two counts:
\[ P(A\text{ OR }B) = \frac{9 + 6}{19} = \frac{15}{19} \]
The second reads OR as exactly one of the two, dropping 14, 16 and 18:
\[ P(\text{only one}) = \frac{9}{19} \]
Both wrong. The first counts the overlap twice; the second does not count it at all.
OR in probability is inclusive: either or both. Each outcome in the union is counted exactly once, including the outcomes in the overlap.
\[ P(A\text{ OR }B) = \frac{9 + 6 - 3}{19} = \frac{12}{19} \]
Check yourself against the Venn diagram. Fifteen would mean the lens numbers were in the union twice; nine would mean they were not in it at all. Twelve is the number of distinct numbers inside either circle.
If a problem really wants exactly one, it will say so, and you then take the union minus the intersection.
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, p. 169 — Notice that 4 and 5 are NOT listed twice
Sorting
Use Example 3.1 again: S is 1 to 19, A the even numbers, B the numbers greater than 13. Sort each number into the one region of the Venn diagram it belongs to.
Sort into buckets
Sort each number into its region of the Venn diagram for Example 3.1.
The number 13 is the one people misplace. Greater than 13 means 14 and up, so 13 itself is not in B, and being odd it is not in A either. Boundary words such as greater than, at least and at most deserve a second reading every time.
Once every outcome has a region, the events you will meet are just unions of regions. A OR B is three regions, A AND B one, the complement of A two, and neither one.
This is why the Venn diagram is worth drawing even when the sets are small. It forces each outcome into exactly one place, so nothing gets counted twice or forgotten.
Worked example
A fair six-sided die is rolled; the outcome is the number of dots showing. A is an even number, B is less than four. Identify each event as a subset of S and compute its probability.
The sample space
Why: six equally likely faces.
\[ S = \{1, 2, 3, 4, 5, 6\} \]
(a) T: the outcome is two
Why: one face.
\[ T = \{2\}, \quad P(T) = \frac{1}{6} \]
(b) and (c) A even, B less than four
Why: three faces each.
\[ A = \{2, 4, 6\}, \quad B = \{1, 2, 3\}, \quad \frac{3}{6} \text{ each} \]
(d) the complement of A
Why: the faces that are not even.
\[ A' = \{1, 3, 5\}, \quad P(A') = \frac{3}{6} \]
Figure (svg): A rectangle S holding two overlapping circles, A (even) and B (less than four). Only in A: 4 and 6. In the overlap: 2. Only in B: 1 and 3. Outside both: 5.
(g) A AND B
Why: the only even face below four.
\[ A\text{ AND }B = \{2\}, \quad P = \frac{1}{6} \]
(h) A OR B
Why: every face except 5.
\[ A\text{ OR }B = \{1, 2, 3, 4, 6\}, \quad P = \frac{5}{6} \]
(i) A OR B prime
Why: B prime is 4, 5, 6; join it with the even faces.
\[ A\text{ OR }B' = \{2, 4, 5, 6\}, \quad P = \frac{4}{6} \]
(j) N: a prime number
Why: 2, 3 and 5; 1 is not prime.
\[ N = \{2, 3, 5\}, \quad P(N) = \frac{3}{6} \]
(k) I: the outcome is seven
Why: no face has seven dots.
\[ I = \{\,\}, \quad P(I) = 0 \]
Parts (e) and (f), the conditional ones, come back in Part 5. Part (k) is the book's reminder that an impossible event is still an event, with probability zero.
Check (h) with the overlap rule
Why: three faces in A, three in B, one in both.
\[ \frac{3 + 3 - 1}{6} = \frac{5}{6} \]
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, pp. 170-171 — Example 3.2
Fill the middle
Back to the number from 1 to 10, with A more than 6 and B odd. Fill in the counts, then the probabilities. Use the strip figure from Part 1 if it helps: count the numbers carrying at least one mark.
Fill in the blanks
P(A\text2B) = \frac7}}___ = \frac___}___
Why: A has four numbers and B five, but 7 and 9 are in both, so two are subtracted to avoid counting them twice. The union is 1, 3, 5, 7, 8, 9, 10: seven numbers, a probability of seven tenths.
Part (h) asks for A prime OR B. A prime is the numbers up to 6, which is six numbers. Joining the odd numbers adds only 7 and 9, because 1, 3 and 5 are already in A prime.
\[ A' \cup B = \{1, 2, 3, 4, 5, 6, 7, 9\}, \quad P = \frac{8}{10} \]
Both answers can be checked by looking at what is left out. The union of A and B misses only 2, 4 and 6, three numbers, and seven plus three is ten. The union of A prime and B misses only 8 and 10, and eight plus two is ten.
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, p. 171 — Try It 3.2(g), (h)
Section
Part 4
Concept
Every event splits the sample space into two pieces: the outcomes in it and the outcomes not in it. The second piece is an event in its own right.
complement — The complement of event A, written A prime and read A prime, consists of all outcomes in the sample space that are not in A.
The book's example: S is the six faces of a die and A is 1, 2, 3, 4. The complement is what is left.
Figure (svg): A long rectangle S divided into six equal cells numbered 1 to 6. The first four cells are shaded cyan and labelled A; the last two are shaded orange and labelled A prime, not in A. Underneath: 4/6 + 2/6 = 1.
\[ A = \{1, 2, 3, 4\}, \quad A' = \{5, 6\} \]
Add the two probabilities
Why: both are counts over the same six faces.
\[ P(A) + P(A') = \frac{4}{6} + \frac{2}{6} = \frac{6}{6} = 1 \]
This is not a coincidence of these numbers. Every outcome is in A or in A prime, and none is in both, so the two counts always add to the size of S, and the two probabilities always add to one. The rule holds for unequal outcomes too, because the long-run shares of A and not A must add to the whole.
\[ P(A) + P(A') = 1 \quad \Longrightarrow \quad P(A') = 1 - P(A) \]
The second form is the useful one. Whenever an event is awkward to count directly, its complement may be easy, and one subtraction finishes the job.
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, p. 169 — the complement
Fill the middle
In Example 3.1, A is the nine even numbers among the nineteen. Fill in the size of A prime and its probability without listing it, then check the sum.
Fill in the blanks
P(A') = \frac101 = \frac___}___, \qquad P(A) + P(A') = ___
Why: Every number below 20 is even or odd and never both, so the odd numbers are the 19 minus the 9 even ones, which is 10. Nine nineteenths plus ten nineteenths is nineteen nineteenths, which is 1.
The book lists A prime as 1, 3, 5, 7, 9, 11, 13, 15, 17, 19: ten numbers, matching the subtraction. Notice that it includes 15, 17 and 19, which are in B. A complement is taken against the whole sample space, not against the other event.
Subtracting is safer than listing as soon as the sample space is big. For a raffle of 1,000 tickets with 500 even ones, nobody wants to list 500 odd numbers, and nobody can miss one by subtracting.
One more thing to notice: the complement depends on the sample space. If S were only 2 to 19, A would still have nine members, but A prime would have nine as well, not ten.
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, p. 170 — Example 3.1(f), (g)
Worked example
Toss a fair coin ten times. Find the probability of getting at least one head. Counting this directly is miserable; the complement makes it a one-liner.
Count the sample space
Why: each toss doubles the number of sequences, ten times over.
\[ 2^{10} = 1024 \text{ equally likely sequences} \]
At least one head includes one head, two heads, and so on up to ten: hundreds of sequences to count. What is not in the event? Only the sequences with no heads at all.
Name the complement
Why: not at least one head means zero heads.
\[ A' = \{\text{TTTTTTTTTT}\} \]
Probability of the complement
Why: one sequence out of 1024.
\[ P(A') = \frac{1}{1024} \]
Use the complement rule
Why: subtract from one.
\[ P(A) = 1 - \frac{1}{1024} \]
Simplify
Why: as a fraction and as a decimal.
\[ P(A) = \frac{1023}{1024} \approx 0.9990 \]
In context: in ten tosses of a fair coin you get at least one head about 999 times in every 1,000 tries. The complement of at least one is none, never at most one; that is the most common slip with this kind of question.
Check the boundary case of one toss
Why: the same method must give the obvious answer one half.
\[ 1 - \frac{1}{2^1} = \frac{1}{2} \]
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, p. 169 — the complement rule
Error analysis
A student wrote the complements of four events about tossing a coin ten times, and then a probability for one of them. Read each line before the notes.
Annotate
On: \( \text{(at least one head)}' = \text{at most one head} \)
The other three lines on the page were: the complement of all heads is all tails; the complement of exactly five heads is not exactly five heads; and the complement of more than six heads is fewer than six heads. Run the two-part test on each one.
All heads against all tails fails: a sequence with three heads is in neither set, so they do not cover S. The correct complement is at least one tail. More than six against fewer than six fails too: exactly six heads is in neither set. The correct complement is six or fewer.
Only the exactly-five line is right, and it is right because it uses the word not. When you write a complement, it is often safest to say not first and translate afterwards, checking the boundary value as you go.
Section
Part 5
Concept
Roll a fair die. Let A be the face is 2 or 3; its probability is two sixths. Now a friend glances at the die and tells you only that the face is even. How likely is A now?
Figure (svg): Six die faces. Faces 1, 3 and 5 are drawn faint with dashed outlines and labelled 'ruled out by the news'. Faces 2, 4 and 6 have thick blue outlines labelled 'B: even, the new sample space'. Face 2 is filled yellow as the only outcome of A (2 or 3) left inside B; face 3 carries a small note 'in A, but ruled out'.
The news rules out the odd faces 1, 3 and 5. They could have happened before; now they cannot. What is left is a smaller sample space, the event B, the face is even.
\[ B = \{2, 4, 6\} \]
The three even faces were equally likely before the news, and the news did not favour any of them, so they remain equally likely. Inside this new sample space, A is the outcomes of A that survive: only the face 2, since 3 was ruled out.
\[ P(A \mid B) = \frac{\text{outcomes of } A \text{ in } B}{\text{outcomes in } B} = \frac{1}{3} \]
conditional probability — The probability of A given B, written P(A given B), is the probability that A occurs given that B has already occurred. The condition reduces the sample space: compute the probability of A within B alone.
The news moved the probability from two sixths, about 0.333, to one third. The same number here, which is no accident, as you will see shortly; other news changes a probability a great deal. The method is what matters: shrink the sample space to B, then count.
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, p. 169 — A conditional reduces the sample space
Concept
The book also gives a formula for the conditional probability that works with probabilities rather than counts. It is not a new idea, just the count rewritten. Start from the die answer.
\[ P(A \mid B) = \frac{1}{3} \]
Divide the top and bottom of the fraction by 6, the size of the original sample space. That never changes a fraction's value.
\[ \frac{1}{3} = \frac{1 \div 6}{3 \div 6} = \frac{1/6}{3/6} \]
Now read the pieces. One sixth is the probability of the single outcome in both A and B, measured on the whole die. Three sixths is the probability of B on the whole die.
\[ P(A \mid B) = \frac{P(A\text{ AND }B)}{P(B)}, \qquad P(B) > 0 \]
That is the book's formula. The top is the overlap, the part of A that survives the news. The bottom is the probability of the news itself. Dividing by the probability of B rescales so that B becomes the whole new world, with total probability one.
The formula needs the probability of B to be greater than zero: you cannot be given news that is impossible, and dividing by zero is undefined. And unlike the counting version, the formula works when outcomes are not equally likely, which is why the book takes it as the definition.
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, p. 169 — the formula for P(A given B)
Notation
This formula will appear in nearly every remaining section of the chapter. Read each piece slowly once, and it will never trip you again.
Annotate
On: \( P(A \mid B) = \frac{P(A\text{ AND }B)}{P(B)} \)
The quickest way to keep the formula straight is to say it in words: of the outcomes where B happens, what share also have A? The of the outcomes where B happens is the denominator. The also have A is the overlap in the numerator.
Notice which event goes on the bottom. It is always the event to the right of the bar. The most common error in all of probability is to put the wrong event there, and the trap later in this part is that error.
Compare with ordinary probability. The probability of A is the probability of A given S: of all outcomes, the share in A. Every probability you met in Parts 1 to 4 was secretly a conditional one with the whole sample space as the news.
Intuition
A condition removes outcomes. Whether that makes A more likely or less likely depends on which outcomes it removes: mostly ones outside A, or mostly ones inside A.
| experiment | event A | news B | before | after the news |
|---|---|---|---|---|
| one card from 52 | a king | a face card: J, Q, K | 4/52 ≈ 0.077 | 4/12 ≈ 0.333 |
| one die | an even face | less than four | 3/6 = 0.5 | 1/3 ≈ 0.333 |
| one die | a 2 or a 3 | an even face | 2/6 ≈ 0.333 | 1/3 ≈ 0.333 |
In the card row, the news throws away 40 cards and not one of them is a king, so the kings become a much larger share of what is left. The probability rises more than fourfold.
\[ \frac{4}{52} \approx 0.077 \quad \longrightarrow \quad \frac{4}{12} \approx 0.333 \]
In the first die row, the news throws away the faces 4 and 6, two of A's three faces, but only half of the whole die. A loses a bigger share than the sample space does, so its probability falls.
In the last row, the news throws away half of the die and exactly half of A, the face 3. A keeps the same share and its probability does not move. News that changes nothing is special, and Section 3.2 calls the two events independent.
So never guess the direction. Reduce the sample space, count, and compare.
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, p. 169 — the die example
Counterexample
A classmate argues: conditioning throws outcomes away, so the probability of A given B can never be bigger than the probability of A. Fewer outcomes, smaller chance.
Discussion prompt
Give a counterexample with one card drawn from a standard 52-card deck, and show both probabilities with numbers.
The flaw is looking only at the numerator. A conditional is a ratio, the outcomes of A that survive over the outcomes of B. Throwing outcomes away can shrink the top, the bottom, or both, and the ratio moves in whichever direction the bottom wins.
The extreme case makes the point vivid. Given that the card is a king, the probability that it is a king is one: the news removed every card that was not a king. A condition can push a probability all the way to 1, and the opposite news, a non-king, pushes it all the way to 0.
Keep this example beside Example 3.2(e), where news lowered a probability from one half to one third. Between them they show the direction is never automatic.
Worked example
One fair die. A is the face is 2 or 3, B is the face is even. Find the probability of A given B, first by counting in the reduced sample space and then by the formula.
Reduce the sample space
Why: the news is B, so only the even faces remain.
\[ B = \{2, 4, 6\} \]
Find the outcomes of A inside B
Why: of the faces 2 and 3, only 2 is even.
\[ A\text{ AND }B = \{2\} \]
Count and divide inside B
Why: one outcome of three, and they are equally likely.
\[ P(A \mid B) = \frac{1}{3} \]
Now the formula, which works with probabilities measured on the whole sample space of six faces. You should get the same number, because the formula is the count with top and bottom divided by six.
Probability of the overlap
Why: one face of six.
\[ P(A\text{ AND }B) = \frac{1}{6} \]
Probability of the condition
Why: three faces of six.
\[ P(B) = \frac{3}{6} \]
Divide
Why: one sixth over three sixths: the sixes cancel.
\[ P(A \mid B) = \frac{1/6}{3/6} = \frac{1}{3} \]
In context: if you know the die shows an even face, it is a 2 or a 3 one time in three.
Check that the conditional probabilities inside B add to one
Why: B is now the whole world, so A and not A within B must cover it.
\[ P(A \mid B) + P(A' \mid B) = \frac{1}{3} + \frac{2}{3} = 1 \]
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, p. 169 — the die example for P(A given B)
Worked example
Return to the numbers from 1 to 19, with A even and B greater than 13. Part (h) asks for the probability of A given B, the probability of B given A, and whether they are equal.
\[ P(A\text{ AND }B) = \frac{3}{19}, \quad P(A) = \frac{9}{19}, \quad P(B) = \frac{6}{19} \]
A given B: divide by the probability of B
Why: the news is greater than 13.
\[ P(A \mid B) = \frac{3/19}{6/19} \]
Cancel the nineteenths
Why: what is left is a count inside B.
\[ P(A \mid B) = \frac{3}{6} = 0.5 \]
B given A: divide by the probability of A
Why: now the news is even.
\[ P(B \mid A) = \frac{3/19}{9/19} \]
Cancel the nineteenths
Why: a count inside A.
\[ P(B \mid A) = \frac{3}{9} \approx 0.333 \]
Figure (svg): Two rows of boxes. Top row, labelled 'given B (above 13)': the six numbers 14 to 19, with 14, 16 and 18 highlighted, and the note '3 of 6 = 0.5'. Bottom row, labelled 'given A (even)': the nine even numbers 2 to 18, with 14, 16 and 18 highlighted, and the note '3 of 9, about 0.333'.
The two answers share their numerator, the three numbers 14, 16 and 18. They differ in their denominators, because they are measured in different groups: three of the six big numbers are even, but only three of the nine even numbers are big.
In context: told that the number is greater than 13, you should give even-ness a 50 percent chance; told that it is even, you should give greater than 13 only about a 33 percent chance. The answer to the book's question is no, they are not equal.
Check both by counting in the figure
Why: count highlighted boxes in each row.
\[ \text{3 of the 6 in } B, \qquad \text{3 of the 9 in } A \]
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, p. 170 — Example 3.1(h)
Trap
Asked for the probability that a left-handed person in Example 3.3 is female, a common line divides by the wrong group:
\[ P(F \mid L) = \frac{4}{48} \]
\[ \approx 0.083 \]
Wrong. That is the share of females who are left-handed, the answer to a different question.
The two questions use the same four words, so they sound alike. The bar is what separates them.
The news is L, so the reduced sample space is the 13 left-handed people. Of those, 4 are female:
\[ P(F \mid L) = \frac{4}{13} \]
\[ \approx 0.308 \]
Always ask which group the question puts you in. Given that a person is left-handed puts you among left-handers, and their total, 13, is the denominator. Getting this backwards is so common in courts and clinics that it has a name, the prosecutor's fallacy.
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, p. 172 — Example 3.3(l), (m)
Worked example
One fair die, A even, B less than four. Part (e) asks for A given B and part (f) for B given A. After the last example you might expect different answers.
The overlap
Why: the only even face below four.
\[ A\text{ AND }B = \{2\}, \quad P(A\text{ AND }B) = \frac{1}{6} \]
(e) A given B: reduce to B
Why: B is 1, 2, 3; one of them is even.
\[ P(A \mid B) = \frac{1}{3} \]
(f) B given A: reduce to A
Why: A is 2, 4, 6; one of them is below four.
\[ P(B \mid A) = \frac{1}{3} \]
This time they are equal. Look at the formula to see why: both conditionals have the same numerator, the overlap, and their denominators are the probabilities of B and of A. Here those are both three sixths, so the two fractions are identical.
\[ P(A \mid B) = \frac{1/6}{3/6}, \qquad P(B \mid A) = \frac{1/6}{3/6} \]
So the two conditionals agree exactly when the two events have the same probability, as long as they overlap at all. In Example 3.1 they did not, nine nineteenths against six, and the conditionals came apart. Never assume equality: check the denominators.
Notice also what the news did here. Before, A had probability one half. Told that the face is below four, A drops to one third. News can lower a probability as easily as raise it.
Check against the book's lists
Why: the book writes A given B and B given A each as the set containing 2, with probability one third.
\[ \frac{1}{3} = \frac{2}{6} < \frac{3}{6} = P(A) \]
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, p. 170 — Example 3.2(e), (f)
Tweak it
The conditional formula divides the overlap by the probability of the condition. Fix the overlap with the slider and let the probability of B vary along the horizontal axis.
Parameter explorer
Set the overlap to 0.16, close to Example 3.1's three nineteenths. Read the conditional when the condition has probability 0.32, about six nineteenths, and when it has probability 0.47, about nine nineteenths. Where does the curve cross height 1, and why can the condition never sit to the left of that point?
\[ P(A \mid B) = \frac{ {a} }{ P(B) } \]
At 0.32 the curve reads 0.5 and at 0.47 it reads about 0.34: the two conditionals of Example 3.1, three sixths and three ninths, on one curve. The same overlap divided by a smaller group gives a bigger share.
The curve crosses height 1 exactly where the probability of B equals the overlap. That is the case where B lies entirely inside A, so everyone in B is also in A. To the left of that point the curve would claim a probability above one, and that region is impossible, because the overlap is part of B and can never be bigger than B.
Slide the overlap up and the whole curve rises. A conditional probability is large when the overlap is a large share of the condition, and that, in a sentence, is the entire formula.
Worked example
The sample space is the ordered pairs whose first number is 1 to 3 and second number is 1 to 4. A is the sum is even; B is the first number is prime.
(a) Count the sample space
Why: three choices of first number times four of second.
\[ 3 \times 4 = 12 \text{ pairs} \]
Figure (svg): A three-by-four grid of the ordered pairs (first number 1 to 3, second number 1 to 4), each cell showing the pair and its sum. Cells with an even sum are shaded cyan: (1,1), (1,3), (2,2), (2,4), (3,1), (3,3). Rows 2 and 3 are enclosed by a dashed orange outline labelled 'B: first number prime'.
(b) List A
Why: the cells with an even sum.
\[ A = \{(1,1), (1,3), (2,2), (2,4), (3,1), (3,3)\} \]
(b) Describe B
Why: 2 and 3 are prime; 1 is not.
\[ B = \text{rows 2 and 3}, \quad 8 \text{ pairs} \]
(c) Probabilities
Why: count over twelve.
\[ P(A) = \frac{6}{12}, \quad P(B) = \frac{8}{12} \]
(d), (e) AND and OR
Why: shaded cells inside the outline; then all of B plus row 1's two shaded cells.
\[ P(A\text{ AND }B) = \frac{4}{12}, \quad P(A\text{ OR }B) = \frac{10}{12} \]
(f) B prime
Why: the pairs whose first number is 1.
\[ B' = \{(1,1), (1,2), (1,3), (1,4)\}, \quad P(B') = \frac{4}{12} \]
Part (g) asks for the probability of A plus the probability of A prime, although part (f) was about B prime: the book has mixed its letters. Both sums are 1, so either reading gives the same answer.
(h) The two conditionals
Why: the overlap over each condition.
\[ P(A \mid B) = \frac{4}{8} = 0.5, \quad P(B \mid A) = \frac{4}{6} \approx 0.667 \]
Check (h) by counting in the grid
Why: four shaded cells among the eight outlined, four outlined cells among the six shaded; not equal.
\[ \frac{4}{8} \ne \frac{4}{6} \]
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, p. 170 — Try It 3.1
Concept
Example 3.3 hands you data in a table instead of a list. Before computing anything from it, see every piece of the table as a piece of probability language.
| part of the table | what it is | probability it gives |
|---|---|---|
| each of the 100 people | one equally likely outcome | a random pick gives each person 1/100 |
| an inner cell, such as 43 | an AND event: male AND right-handed | cell over 100 |
| a row or column total, such as 52 | a single event: male | total over 100 |
| a row or column, entered alone | a reduced sample space, the condition | cell over that total |
The people are equally likely because the person is drawn at random from the sample. That is the condition that lets you count, exactly as the fairness of the die did. Every probability in the table is therefore a count over a count.
The last row of the table above is the one to underline. A conditional probability in a two-way table is never a new kind of calculation: you put your finger on the row or column named after the bar, and you read the table as if nothing else existed. The grand total of 100 plays no part.
Two quick checks keep table work honest. The inner cells of a row must add to the row total, and the totals of the rows must add to the grand total. If they do not, you misread a cell before you even started.
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, p. 171 — Table 3.1
Worked example
A random sample of 100 people is classified by sex assigned at birth and by handedness. M is male, F female, R right-handed, L left-handed. A person is drawn at random from the 100.
| right-handed | left-handed | total | |
|---|---|---|---|
| males | 43 | 9 | 52 |
| females | 44 | 4 | 48 |
| total | 87 | 13 | 100 |
The book prints only the four inner counts. Add the row and column totals first, because every single-event probability is a total over 100.
Figure (svg): A unit square split into two columns, males of width 0.52 and females of width 0.48. In the male column a cyan block of 43 right-handers sits above an orange block of 9 left-handers; in the female column a cyan block of 44 sits above an orange block of 4.
(a) to (d) Single events
Why: row and column totals over 100.
\[ P(M) = 0.52, \; P(F) = 0.48, \; P(R) = 0.87, \; P(L) = 0.13 \]
(e), (f) AND events
Why: one inner cell each.
\[ P(M\text{ AND }R) = \frac{43}{100}, \quad P(F\text{ AND }L) = \frac{4}{100} \]
(g) M OR F
Why: every person is one or the other.
\[ P(M\text{ OR }F) = \frac{52 + 48}{100} = 1 \]
(h) M OR R
Why: 52 males plus 87 right-handers counts the 43 right-handed males twice.
\[ P(M\text{ OR }R) = \frac{52 + 87 - 43}{100} = 0.96 \]
(i) F OR L
Why: 48 females plus 13 left-handers, minus the 4 counted twice.
\[ P(F\text{ OR }L) = \frac{48 + 13 - 4}{100} = 0.57 \]
(j) M prime
Why: not male means female here.
\[ P(M') = 1 - 0.52 = 0.48 \]
Check (h) with the complement
Why: the only people not in M OR R are the left-handed females.
\[ 1 - \frac{4}{100} = 0.96 \]
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, pp. 171-172 — Example 3.3(a)-(j)
Worked example
The same 100 people. In a table, conditioning on an event means stepping into its row or column and ignoring the rest of the table.
(k) R given M: step into the male row
Why: 52 males, 43 of them right-handed.
\[ P(R \mid M) = \frac{43}{52} \]
Divide
Why: four decimal places, as the book does.
\[ P(R \mid M) \approx 0.8269 \]
(l) F given L: step into the left-handed column
Why: 13 left-handers, 4 of them female.
\[ P(F \mid L) = \frac{4}{13} \approx 0.3077 \]
(m) L given F: step into the female row
Why: 48 females, 4 of them left-handed.
\[ P(L \mid F) = \frac{4}{48} \approx 0.0833 \]
Figure (svg): Two rows of small squares, one square per person. Top: 48 squares for the females, 4 of them orange. Bottom: 13 squares for the left-handers, the same 4 orange. Notes at the right read '4 of 48, about 0.083' and '4 of 13, about 0.308'.
Parts (l) and (m) use the same four people, so they have the same numerator. They differ because 4 is a big share of the 13 left-handers and a small share of the 48 females. In context: about 31 percent of left-handers in this sample are female, but only about 8 percent of females are left-handed.
Part (k) is worth comparing with the overall rate. Among everyone, 87 percent are right-handed; among males, about 83 percent; and among females, 44 of 48, about 92 percent. In this sample, knowing someone is male lowers the chance that they are right-handed a little.
Check (k) with the formula
Why: joint probability over the condition's probability gives the same value as the count.
\[ \frac{P(M\text{ AND }R)}{P(M)} = \frac{0.43}{0.52} \approx 0.8269 \]
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, p. 172 — Example 3.3(k)-(m)
Comparison
A random sample of 100 people is classified by gender and by whether they prefer tea or coffee: 22 men and 16 women like tea, 26 men and 36 women like coffee. Fill in the blanks, working from the counts.
Comparison matrix
| probability | count | denominator | value |
|---|---|---|---|
| P(M) | 22 + 26 | 100 | 0.48 |
| P(T) | 38 | 100 | 0.38 |
| P(M OR T) | 48 + 38 − 22 | 100 | 0.64 |
| P(W OR C) | 52 + 62 − 36 | 100 | 0.78 |
| P(T | M) | 22 | 48 | 0.4583 |
| P(W | C) | 36 | 62 | 0.5806 |
Read down the denominator column. Every unconditional probability, AND or OR, is over the full 100. Every conditional is over the size of its condition: 48 men for tea given man, 62 coffee drinkers for woman given coffee. The denominator column is the whole difference between the two kinds of question.
The two OR rows subtract the overlap cell once. Men or tea drinkers is 48 men plus 38 tea drinkers, but the 22 men who like tea are in both groups. Seen the other way, the people outside M OR T are just the 36 women who prefer coffee, and 100 minus 36 is 64.
The remaining Try It answers follow the same patterns: the probability of a woman is 0.52, of coffee 0.62, of M AND T 0.22, of W AND C 0.36, of M OR W 1, of M prime 0.52, and of coffee given woman 36 over 52, about 0.6923.
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, p. 172 — Try It 3.3
Worked example
One more pair of conditionals, from the number picked at random from 1 to 10. A is more than 6, B is odd. Use the strip from Part 1: 7 and 9 carry both marks.
The overlap
Why: odd numbers above 6.
\[ A\text{ AND }B = \{7, 9\} \]
(e) A given B: reduce to the odd numbers
Why: five odd numbers, two of them above 6.
\[ P(A \mid B) = \frac{2}{5} = 0.4 \]
(f) B given A: reduce to the numbers above 6
Why: four numbers, two of them odd.
\[ P(B \mid A) = \frac{2}{4} = 0.5 \]
Same numerator, two different denominators, two different answers, just as in Example 3.1. The probability of A is four tenths and of B five tenths; the events have different sizes, so their conditionals on each other differ.
Compare each conditional with the plain probability. Knowing the number is odd leaves the chance of more than 6 at 0.4, the same as before. Knowing it is more than 6 leaves the chance of odd at 0.5, also the same. This news changes nothing, and Section 3.2 will call such events independent.
\[ P(A \mid B) = 0.4 = P(A), \qquad P(B \mid A) = 0.5 = P(B) \]
Check both with the formula
Why: two tenths over the probability of each condition.
\[ \frac{0.2}{0.5} = 0.4, \qquad \frac{0.2}{0.4} = 0.5 \]
OpenStax Introductory Statistics 2e, §3.1 Terminology §3.1, p. 171 — Try It 3.2(e), (f)
Ranking
All four of these come from Example 3.3's table of 100 people. Put them in order without a calculator: think about numerator and denominator.
Put in order
Order these from smallest to largest.
Why: Four of 100 is 0.04, the smallest because the joint event is counted against everyone. Four of 48 is about 0.083. Thirteen of 100 is 0.13. Four of 13 is about 0.308, the largest, because the same four people are measured against the smallest group.
The joint probability is always the smallest of the three that share the numerator 4. It has the biggest denominator, all 100 people, while each conditional divides by a group inside the 100. A conditional is never smaller than the matching AND probability.
\[ P(A\text{ AND }B) = P(A \mid B) \, P(B) \le P(A \mid B) \]
That inequality is the formula rearranged: multiplying the conditional by a probability, a number no bigger than one, can only shrink it. Section 3.3 turns the rearranged formula into the multiplication rule.
If you placed the probability of L above the probability of F given L, recheck the denominators: 13 people out of 100 is one share; 4 out of those 13 is another, and the second is much larger.
Step zero
The chapter begins with a class survey. Suppose a class of 30 answers it: 18 students have change in a pocket or purse, 12 rode a bus in the past month, and 8 did both. The book asks for the probability of change given bus.
Discussion prompt
Before computing anything, say which students go in the denominator of the probability of change given bus, and which go in the numerator. Then compute it and compare it with the probability of change for the whole class.
The book's own instruction is a perfect description of reducing the sample space: count all the students who rode a bus; from that group, count those who have change; divide. The 30 students in the room play no part once the condition is stated.
The class numbers here are invented to show the method; your own class will give different ones. The step-zero habit is the same either way: name the group before you count, and the group is always the event to the right of the bar.
OpenStax Introductory Statistics 2e, §3.1 Terminology §3 introduction, pp. 167-168 — the collaborative exercise
Real world
A condition affects 1 percent of a population. A screening test catches 99 percent of people who have it and wrongly flags 1 percent of people who do not. You test positive. How worried should you be?
Discussion prompt
Work with 10,000 people. Find how many have the condition and test positive, how many do not have it and test positive, and then the probability of having the condition given a positive test.
Figure (svg): Two bars side by side for the 198 people who test positive: 99 who have the condition and 99 who do not. Beneath, a note: 100 of the 10,000 have the condition; 9,900 do not.
The surprise comes from the denominator. The condition positive puts you among all 198 positives, and half of them come from the huge healthy group, where even a 1 percent error rate produces 99 false alarms.
The test's 99 percent is the probability of a positive given sickness. What you want is the probability of sickness given a positive. They are the two conditionals of this lesson, with different denominators, and confusing them leads doctors and juries astray. This is why a positive screening result is followed by a second, different test.
Section
Part 6
Pattern
Every question in this section has the same shape. Name the sample space, translate the words into an event, and count that event in the right sample space: the whole of S, or the reduced sample space given by a condition.
| the question says | the event | count | divide by |
|---|---|---|---|
| A | the outcomes in A | outcomes in A | all of S |
| A and B, both | A AND B, the overlap | outcomes in both | all of S |
| A or B, either, at least one | A OR B, the union | A + B − both | all of S |
| not A | A′, the complement | S − A, or use 1 − P(A) | all of S |
| A given B, A among B | A within the reduced space B | outcomes in both | outcomes in B only |
The last column is where most marks are lost. Only one row of the table changes the denominator, and it is the row with a condition in it.
Check
One question on the counting rule. Decide first whether the outcomes you are counting are equally likely.
Check your understanding
Two fair coins are tossed. What is the probability of at least one tail?
Answer: A
Why: The equally likely outcomes are HH, HT, TH and TT. At least one tail holds HT, TH and TT, three of the four, so the probability is three quarters. The complement, no tails, is only HH.
Write the four outcomes of the grid before anything else. With the list in front of you, at least one tail is three cells out of four; without it, the head totals look like a natural sample space and they are not equally likely.
The complement gives a second route. The only outcome with no tail is two heads, one cell of four, so at least one tail is one minus a quarter, which agrees.
Check
Now the combinations. Use Example 3.2's die: A is even, B is less than four.
Check your understanding
On one roll of a fair die, with A even and B less than four, what is the probability of A OR B prime?
Answer: A
Why: B prime is the faces 4, 5 and 6. Joining them with the even faces 2, 4 and 6 gives 2, 4, 5 and 6, four faces with each written once, so the probability is four sixths.
Build the complement first, against the whole die: the faces that are not less than four are 4, 5 and 6. Then take the union with the even faces, writing each face once.
The quick check is the complement of the answer. The faces not in A OR B prime are 1 and 3, two faces, and four plus two is six. If you had six sixths, the faces 4 and 6 were counted twice.
Check
Last, a conditional from a table. Find the bar, then find the denominator.
Check your understanding
In Try It 3.3, 22 men and 16 women prefer tea, and 26 men and 36 women prefer coffee. What is the probability that a person prefers tea, given that the person is a woman?
Answer: A
Why: Given woman puts you in the women's row: 16 plus 36 is 52 women, and 16 of them prefer tea. The probability is sixteen fifty-seconds, about 0.308.
The bar says given woman, so the reduced sample space is the women's row. Add it first, 16 plus 36 is 52, and only then look for the tea drinkers inside it.
The three wrong options are the three classic slips: dividing by the other event's total, dividing by everyone, and taking the wrong cell inside the right row. Each one gives a number that answers some question, just not this one.
Explain it to yourself
You have computed conditional probabilities two ways: counting inside B, and dividing the probability of the overlap by the probability of B. Explain why they always agree when outcomes are equally likely.
Discussion prompt
In two or three sentences, explain why counting inside the reduced sample space and using the formula give the same answer. Use the die example, where A is 2 or 3 and B is even.
A strong answer names the six: both probabilities in the formula are counts over the same sample space, so the size of that sample space cancels. That is why the formula is the count in disguise.
An even stronger answer adds when the formula is needed: when outcomes are not equally likely there is nothing to count, but the probabilities still exist, and the formula still works. That is why the book uses it as the definition.
If your explanation was only the formula, try saying it in words: of the outcomes in B, the share also in A.
Exit ticket
A number is chosen at random from 1 to 12. Let A be a multiple of 3 and B be greater than 8.
Discussion prompt
List A, B, A AND B and A OR B. Find P(A OR B), P(A prime) and P(A given B), and say in one sentence what the last one means.
Every part of the lesson is here. The sample space has twelve equally likely numbers; the union subtracts the overlap of 9 and 12 once; the complement is one subtraction; and the conditional counts inside the four numbers above 8.
If one part felt shaky, go back to the matching Part of the deck: Part 3 for unions, Part 4 for complements, Part 5 for the bar. The conditional is the one the next three sections depend on most.
Recap
Probability has its own vocabulary, and every word in it is a set or a count. An experiment produces outcomes, the sample space collects them all, and an event is any set of them. A probability is a long-term relative frequency, between 0 and 1, and with equally likely outcomes it can be computed by counting.
| idea | in words | in symbols |
|---|---|---|
| counting rule | outcomes in A over outcomes in S, if equally likely | P(A) = n(A) / n(S) |
| law of large numbers | the relative frequency tends to the probability | empirical → theoretical |
| OR, the union | in A, in B, or in both; count the overlap once | A ∪ B |
| AND, the intersection | in both at the same time | A ∩ B |
| complement | everything not in A | P(A′) = 1 − P(A) |
| conditional | A counted inside B | P(A | B) = P(A AND B) / P(B) |
\[ P(A \mid B) = \frac{P(A\text{ AND }B)}{P(B)} \ne P(B \mid A) \text{ in general} \]
The idea to carry forward is the condition. It reduces the sample space, it changes the denominator, and the two conditionals of a pair are usually different numbers. Section 3.2 asks when news changes nothing at all, and Section 3.3 rearranges the conditional formula into the multiplication rule.
OpenStax Introductory Business Statistics 2e, §3.1 Terminology §3.1 — the same terminology with business examples
Want this taught 1-on-1? Alexander tutors Statistics — $55/session, free consultation.