4.3 Binomial Distribution

The first named family of distributions. A binomial experiment has three characteristics: a fixed number of trials n, exactly two outcomes on each trial with success probability p and failure probability q summing to one, and independent trials repeated under identical conditions so that p never changes. The random variable X counting the successes then has a binomial distribution, written X follows B of n and p, and its probabilities come from a formula whose three factors count the orderings and give the probability of one of them. Its mean is np and its standard deviation is the square root of npq, so two numbers determine the whole distribution — which is what makes a named family worth having.

Subject: Statistics · 70 slides · symbolic lesson

Open the interactive version of this deck

What this lesson covers

The lesson, slide by slide

1. The Binomial Distribution

Title

Statistics · Chapter 4 — Discrete Random Variables

Count the successes in a fixed number of independent tries, and two numbers tell you everything

2. What this lesson gives you

Objectives

Section 4.2 built every distribution by hand: list the values, find each probability, fill the table. That works, but it means starting from nothing every time. This section meets the first distribution with a name, the binomial, where one recipe produces the whole table from two numbers.

Keep one question in mind throughout: what exactly is being counted? The binomial counts successes, nothing else, and nearly every mistake in this section comes from losing track of what a success is, how many trials there are, or which counts the question includes.

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, pp. 237-242 — the section this deck covers

3. Before anything new: independent events and complements

Warm-up

Two tools from Chapter 3 carry this whole section. Use them on a small problem first, so they are warm when the binomial needs them.

Discussion prompt

Joe guesses on true-false questions and gets each one right with probability 0.6, independently of the others. What is the probability that he gets three questions in a row right? That he gets all three wrong? That he gets at least one right?

Notice what you did not compute: the probability of exactly one right, or exactly two. There are several different sequences that give two rights, and they have to be counted. That counting is the one new idea in this section, and Part 3 builds it.

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, p. 237 — Joe guesses correctly with p = 0.6

4. The three characteristics

Section

Part 1

5. A distribution worth giving a name

Concept

Every distribution in Section 4.2 was made from scratch: the soccer team, the night wakings, the earthquake bet. Each needed its own table. But many real questions share one shape: repeat the same yes-or-no try a fixed number of times and count the yeses.

The details differ, the shape does not. When a problem fits a known shape, you can skip building the table and use the shape's formula instead. The book puts it plainly: a probability distribution is a pattern, and you fit the problem to the pattern to do the calculation.

That makes the first job recognition. Before you compute anything, you must be able to say whether a situation really has the binomial shape. The next slides give the three conditions that decide it.

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.2, p. 236 — a probability distribution function is a pattern

6. The three characteristics of a binomial experiment

Concept

A binomial experiment is any process with all three of these properties. Test every one; a situation that fails any of them is not binomial, however close it looks.

  1. A fixed number of trials. The trials are repetitions of the same experiment, and their number, written n, is decided before you start.
  2. Two outcomes per trial. Each trial ends in success or failure. The probability of success on one trial is p, of failure q, and p + q = 1.
  3. Independent trials under identical conditions. One trial's outcome tells you nothing about the next, so p and q stay the same on every trial.

\[ p + q = 1 \quad\Longrightarrow\quad q = 1 - p \]

The book's example is Joe guessing on true-false questions. A success is a correct guess, with p equal to 0.6, so q is 0.4. Because Joe guesses independently each time, those two numbers are the same on the first question and the fiftieth.

binomial random variable — X, the number of successes in the n independent trials of a binomial experiment. It takes the whole-number values 0, 1, 2, and so on up to n.

Success is only a label. It means the outcome you are counting, even if that outcome is a failure in ordinary life, like a student withdrawing from a class.

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, p. 237 — the three characteristics

7. A binomial experiment as a row of boxes

Picture it

Picture the experiment as a row of n boxes, one per trial. Each box is filled with S or F, and the chances of each letter are the same in every box.

Figure (svg): Five boxes in a row labelled trial 1 to trial 5, showing S, F, F, S, F. Under each box: P of S equals p and P of F equals q. An arrow underneath says n equals 5 trials fixed before you start; the heading says each trial is independent with the same p. At the bottom: X counts the S boxes, here X equals 2.

Everything a binomial experiment needs is in this strip: a fixed number of boxes, two possible letters in each, the same chances in every box. The random variable only counts the S letters; it ignores where they fall.

The random variable X ignores the order of the letters and just counts the S boxes. The sequence S F F S F gives X equal to 2, and so do nine other sequences with two S letters in different places. That fact is where the formula in Part 3 comes from.

Each condition can be read off the picture. Fixed n: the number of boxes is set in advance, not decided as you go. Two outcomes: only two letters are allowed. Independent: what went into one box has no effect on the odds in the next.

Now imagine breaking each condition on the picture. If you kept adding boxes until the first S appeared, the row would have no fixed length. If a box could hold a number from 1 to 6, there would be more than two letters. If filling one box with S made the next S less likely, the boxes would no longer be independent. Those are exactly the three ways a situation fails to be binomial, and the sorting slide ahead asks you to spot each one.

8. One trial is a Bernoulli trial

Concept

Strip the experiment down to a single box, n equal to 1. What is left satisfies conditions two and three and has just two possible values for the count: 0 or 1.

Bernoulli trial — An experiment with exactly two outcomes, success with probability p and failure with probability q, and n equal to 1. Named after Jacob Bernoulli, who studied them in the late 1600s.

Figure (svg): A probability histogram with two bars: at x equals 0 a bar of height 0.4 labelled wrong, and at x equals 1 a bar of height 0.6 labelled right. A dashed line at 0.6 marks the mean.

One trial on its own has only two values, 0 and 1. Its mean is p itself, because the value 1 carries probability p and the value 0 contributes nothing.

Compute its mean with the Section 4.2 table: value times probability, added.

\[ \mu_{\text{one trial}} = (0)(q) + (1)(p) = p \]

So the mean of one trial is p itself: over many single questions Joe averages 0.6 correct per question. A binomial experiment is simply a string of n Bernoulli trials whose successes you add up, which is why the binomial mean in Part 4 will turn out to be n copies of p.

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, p. 237 — Bernoulli trials

9. Example 4.9 and Try It 4.9: choosing the success

Worked example

At ABC College the withdrawal rate from an elementary physics course is 30 percent in any given term. Set the course up as a binomial experiment.

Choose what to count

Why: the book counts withdrawals, so a success is a student who withdraws, even though no instructor would call that a success.

\[ \text{success} = \text{a student withdraws} \]

Why count withdrawals rather than students who stay? Because that is what the college is worried about. Either choice gives a valid binomial; they are mirror images of each other. What you may not do is switch halfway through the problem.

Read off p

Why: the probability of a success on one trial, for one student.

\[ p = 0.30 \]

Find q from p

Why: every student either withdraws or stays, so the two probabilities add to one.

\[ q = 1 - 0.30 = 0.70 \]

The 70 percent is the students who stay for the whole term. The book states it as a separate fact, but you never need to be told q: it always follows from p.

Name the random variable in words

Why: the count of successes in one class.

\[ X = \text{number of students who withdraw from the class} \]

The book does not give the class size, so n is not yet known and the distribution cannot be written in full. That is fine: the success, p and q are settled first, and n is simply the number of students on the register once you know it. For a class of 25, X would run from 0 to 25.

Try It 4.9 turns this round. Forty-eight percent of a state's schools offer fruit at lunch every day. The health board is counting schools that do, so a success is a school that offers fruit daily, with p equal to 0.48 and q equal to 0.52.

Check that p and q describe the same two outcomes

Why: they must be complements: withdraw or stay, fruit or no fruit, and add to one in each case.

\[ 0.30 + 0.70 = 1, \quad 0.48 + 0.52 = 1 \]

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, p. 237 — Example 4.9 and Try It 4.9

10. Before any numbers: what is the success here?

Step zero

Four short situations. For each one, the first move is not arithmetic; it is deciding what you are counting. Say the success in words and give p before you read the answer.

Discussion prompt

For each situation, name the success and give p and q. (a) A trainer counts how many of 20 attempts a dolphin catches a fish; it succeeds 35 percent of the time. (b) A tutor counts how many of 30 students do not volunteer; 32 percent volunteer. (c) A die is rolled ten times and you count the ones. (d) A basketball player's 80 shots are checked for misses; he scores 61.3 percent of shots.

Situations (b) and (d) are there to catch you. The percentage printed in the problem is not automatically p. It is p only when it is the probability of the thing being counted. When the count is of the other outcome, p is the complement.

Get into the habit of writing the success in words before writing any number. A problem set up with the wrong success produces a perfectly computed answer to a different question.

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, pp. 237-238 — Try It 4.10 and Try It 4.11

11. Binomial or not? Find the failing condition

Sorting

Test each situation against the three characteristics. If it passes all three it is binomial; if not, name the first condition it breaks.

Sort into buckets

Sort each situation by whether it is binomial, and if not, which condition fails.

Binomial
Roll a fair die 10 times; X = number of ones; Survey 50 randomly chosen adults; X = number who prefer saving (60% do)
No fixed number of trials
Roll a die until the first six appears; X = number of rolls; Keep shooting free throws until you miss; X = number made
Not two outcomes per trial
Weigh 30 apples; X = their total weight; Roll a die 12 times; record the face, 1 to 6, on each roll
Trials not independent
Draw 5 cards from a deck without replacement; X = number of hearts; Draw 2 names from 6 students and 10 staff without replacement; X = number of students
bin
Ten independent rolls, each a one or not with p = 1/6; fifty randomly chosen adults, each a saver or not with p = 0.60. Fixed n, two outcomes, constant p.
n
Rolling until a six and shooting until a miss both decide the number of trials as you go. That is the geometric pattern of Section 4.4, not the binomial.
two
A weight is a measurement, not a success or failure, and a face from 1 to 6 has six outcomes. Recording whether each roll is a six would make it binomial.
ind
Without replacement, each draw changes what is left. After a heart the chance of another heart drops from 13/52 to 12/51; after a student the chance of another drops from 6/16 to 5/15.

Notice the fix for the dice record: change what you record from the face to whether it is a six, and the same experiment becomes binomial. Two outcomes is often a choice you make about what to count.

The last two rows are the most important. Sampling without replacement looks like repeated trials, but the chances shift after every draw. The next slide shows exactly how much.

The two rows about stopping rules are the other common near-miss. Rolling until a six sounds like repeated trials with a constant p, and it is; what it lacks is a number of trials fixed before you start. The count of rolls is the random quantity, not the count of successes, which is always exactly one.

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, p. 242 — Example 4.16 and Try It 4.16

12. Example 4.16: the committee draw is not binomial

Worked example

A student advisory committee has ten staff members and six students. Two names are drawn from a box without replacement: the first is chair, the second recorder. Is the number of students chosen a binomial random variable?

Test the fixed number of trials

Why: two draws, decided in advance.

\[ n = 2 \quad \checkmark \]

Test two outcomes

Why: each name is a student or a staff member.

\[ \text{student or staff} \quad \checkmark \]

So far it looks binomial. The third condition is where it breaks, and the tree shows why. Watch what happens to the chance of drawing a student on the second draw: it depends on which branch you are on, and in a binomial experiment it would not.

Figure (svg): A tree diagram for drawing two names without replacement from 6 students and 10 staff. First branches: student 6/16, staff 10/16. After a student the branches are student 5/15 and staff 10/15; after a staff member they are student 6/15 and staff 9/15. The student-student path is highlighted.

Look at the two branches that end in a student on the second draw: one says 5/15, the other 6/15. In a binomial tree those two numbers would have to be equal. Here the first draw changes them.

Probability of a student on the first draw

Why: 6 of the 16 names.

\[ P(\text{student first}) = \frac{6}{16} = 0.375 \]

On the second draw, after a student

Why: one student's name is gone: 5 of 15 remain.

\[ P(\text{student second} \mid \text{student first}) = \frac{5}{15} \approx 0.333 \]

On the second draw, after a staff member

Why: all six students remain among 15 names.

\[ P(\text{student second} \mid \text{staff first}) = \frac{6}{15} = 0.4 \]

The chance of a student on the second trial depends on the first, so the trials are not independent and p is not constant. The binomial's third condition fails. The fix is not a better formula for p; it is to stop using the binomial and multiply along the tree instead, as Chapter 3 taught for dependent events.

Compute the true probability of two students

Why: multiply along the highlighted path.

\[ P(\text{both students}) = \frac{6}{16} \cdot \frac{5}{15} = \frac{30}{240} = 0.125 \]

Check against what a binomial would wrongly give

Why: treating p as 6/16 on both draws squares it, and the answer is off by more than a tenth.

\[ \left(\frac{6}{16}\right)^2 = 0.140625 \ne 0.125 \]

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, p. 242 — Example 4.16

13. Trap: calling any repeated draw binomial

Trap

The trap

Try It 4.16: the names of all the seniors on a lacrosse team go in a hat, and the first three drawn become captains. A first attempt:

Three draws, fixed. Each captain plays the position or does not. So the number who play one position is binomial, and it is:

\[ X \sim B(3, p) \]

Wrong. It checks two conditions and skips the one that fails.

The fix

The names are not replaced, since one person cannot be captain twice. After the first draw the hat holds one fewer name, and the chance that the second captain plays a given position depends on who went first.

\[ \frac{k}{m} \;\longrightarrow\; \frac{k-1}{m-1} \;\text{or}\; \frac{k}{m-1} \]

That is exactly the committee problem again. Always check the third condition out loud: is anything used up, removed or learned between trials? If so, p moves, and the binomial formula gives a wrong answer.

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, p. 242 — Try It 4.16

14. When is sampling without replacement close enough?

Counterexample

Surveys always sample without replacement: nobody is asked twice. Yet Try It 4.14 treats 50 surveyed adults as binomial. Reconcile that with the committee.

Discussion prompt

The committee had 16 names; a national survey draws 50 people from millions. Compute how much the chance of the second person being a saver changes in each case, if 60 percent of a population of 1,000,000 are savers and the first person drawn was a saver. Then explain why one is binomial in practice and the other is not.

So the committee is a genuine counterexample, not a technicality: the population is tiny, and the dependence is large. A random sample from a large population is binomial to every decimal place that matters.

Example 4.14 in Part 4 sits in between: 100 pages drawn from 560 is almost a fifth of the catalogue, so the book's binomial answer there is an approximation. Keep that in mind when you meet it.

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, pp. 241-242 — Try It 4.14 and Example 4.16

15. Notation and setting up the question

Section

Part 2

16. Two numbers name the whole distribution

Concept

Once a situation passes the three tests, only two numbers matter: how many trials, and the probability of success on each. Everything else, every probability and the mean and the spread, follows from those two.

\[ X \sim B(n, p) \]

Read it as: X is a random variable with a binomial distribution, with n trials and success probability p. The tilde means is distributed as. The letter B names the family, and n and p pick one member of it.

parameters — The numbers that select one distribution from a family. For the binomial they are n, the number of trials, and p, the probability of success on each trial. q is not a separate parameter: it is 1 minus p.

Writing the distribution this way is the finish line of setting up a problem. Once you have written it, the rest of the problem is computation, so every setup in this section ends with a line of this form.

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, p. 239 — notation for the binomial

17. Reading X ~ B(n, p) and the question P(x = 15)

Notation

A setup has two lines: the distribution and the question. Every symbol in them has one job.

Annotate

On: \( X \sim B(20,\, 0.55), \quad P(x = 15) \)

  • The random variable, always defined in words first: the number of games you win out of 20. Capital X is the variable; lower-case x is one of its values.
  • Is distributed as. It says what pattern the probabilities follow, not that X equals anything.
  • n, the number of trials, fixed in advance. X can be any whole number from 0 to 20.
  • p, the probability of a success on each trial. The failure probability q = 0.45 is implied and never written in the brackets.
  • The question: the probability that the count comes out at exactly 15. The value inside can be an equation or an inequality.

Notice the case of the letters. The capital X names the random variable, the whole count before the games are played. The lower-case x inside the question is one particular value it might take. The same convention ran through Section 4.1, and it matters more here because you will be adding many P of x terms for different values.

A good habit: say the setup aloud as a sentence. X, the number of games won out of 20 played, is binomial with n equal to 20 and p equal to 0.55, and we want the probability that X is exactly 15.

If you cannot say the sentence, you do not yet know what is being counted, and that is the point to stop and reread the problem rather than reach for a formula.

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, p. 237 — Example 4.10

18. Example 4.10: setting up the game you win 55 percent of the time

Worked example

You play a game you can only win or lose, winning each game with probability 55 percent, independently. You play 20 times. Write the probability question for winning 15 of the 20 games.

Define X in words

Why: count the wins, so a success is a win.

\[ X = \text{number of games won out of 20} \]

Test the three conditions

Why: 20 games fixed in advance, each won or lost, independently with the same chance.

\[ n = 20 \text{ fixed}, \quad \text{win or lose}, \quad \text{independent} \]

The problem hands you independence directly: each game you play is independent. In a real game that is an assumption worth questioning, since players improve or tire, but in this example it is given, and without it nothing below would be valid.

Read off p and find q

Why: the complement of winning is losing.

\[ p = 0.55, \quad q = 1 - 0.55 = 0.45 \]

List the possible values

Why: anything from no wins to winning every game.

\[ x = 0, 1, 2, \ldots, 20 \]

Write the distribution

Why: the two parameters in the brackets.

\[ X \sim B(20,\, 0.55) \]

State the question

Why: exactly 15 wins.

\[ P(x = 15) \]

You now have a complete, unambiguous problem. Part 3 shows how to turn P(x = 15) into a number; the answer there is about 0.0365.

Check the setup against the story

Why: q = 0.45 matches the stated 45 percent chance of losing, and 15 is one of the allowed values 0 to 20.

\[ 0.55 + 0.45 = 1, \quad 0 \le 15 \le 20 \]

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, p. 237 — Example 4.10

19. Try It 4.10: the rescued dolphin

Fill the middle

A trainer teaches a rescued dolphin to catch live fish. It catches the fish 35 percent of the time. Out of 20 attempts, you want the probability that it succeeds 12 times. Fill in the setup.

Fill in the blanks

X \sim B(20,\, 0.35), \quad q = 0.65, \quad P(x = 12)

Why: The dolphin makes 20 attempts, so n is 20. A success is a catch, with probability 0.35, and the failure probability is 1 minus 0.35, which is 0.65. The question asks for exactly 12 successes.

The success is a catch because catches are what the question counts. The 65 percent in the problem is q, and it is already implied by p; you would get it by subtraction even if the problem did not say it.

If you typed 0.65 for p, you counted the misses instead of the catches. That is a legitimate binomial too, but for it the question would become exactly 8 misses, P(x = 8), with the same answer. Swapping the success swaps p with q and x with n minus x, all together.

Part 3 finishes this one: the probability comes out at about 0.0136. Before you get there, notice that 12 catches is well above the 7 you would expect from 20 attempts at 35 percent, so the answer should be small.

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, p. 237 — Try It 4.10

20. Turning words into inequalities

Concept

Most binomial questions do not ask for exactly one count. They ask for a range, stated in words, and each phrase decides whether the boundary value is in or out.

Figure (svg): Four number lines from 36 to 44. At least 40: dots filled from 40 upward. More than 40: filled from 41 upward. At most 40: filled from 40 downward. Fewer than 40: filled from 39 downward. Each row is labelled with its inequality.

Look only at the dot on 40. It is filled in the two rows that say at least and at most, and empty in the two that say more than and fewer than. Every translation error in this section is a mistake about that one dot.
wordsinequalityfor n = 50, the counts included
at least 40x ≥ 4040, 41, …, 50
more than 40x > 4041, 42, …, 50
at most 40x ≤ 400, 1, …, 40
fewer than 40x < 400, 1, …, 39

At least and at most include the boundary; more than and fewer than exclude it. Because x is a whole number, the strict ones can always be rewritten with the next count in: more than 40 is the same as at least 41.

Write the list of included counts, not just the symbol, every time you translate. It takes three seconds and it is exactly what you will need when you add the probabilities up.

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, p. 239 — at least translates as greater than or equal to

21. Match each phrase to the counts it includes

Matching

A class of 50 students, X the number who hand homework in on time. Match each phrase to its list of counts. Watch the boundary value.

Match the pairs

  • a. at least 40 hand it in
  • b. more than 40 hand it in
  • c. no more than 40 hand it in
  • d. fewer than 40 hand it in
  • w. x = 40, 41, …, 50
  • x. x = 41, 42, …, 50
  • y. x = 0, 1, …, 40
  • z. x = 0, 1, …, 39

Why: At least 40 includes 40 and everything above it. More than 40 starts at 41. No more than 40 is the same as at most 40 and includes 40. Fewer than 40 stops at 39.

Read each phrase from the point of view of the boundary student, the fortieth. Does the event still happen if exactly 40 hand it in? For at least 40 and no more than 40, yes; for more than 40 and fewer than 40, no. That single question settles every translation.

No more than is the phrase that trips people. It sounds like a strict limit, but it means the count may reach 40 and go no higher, so 40 is in. Read it as at most.

Two of the lists are complements of each other: at least 40 and fewer than 40 between them cover every count exactly once. So do more than 40 and at most 40. That pairing is what makes the subtraction trick in Part 5 work.

22. Example 4.12: setting up the homework question

Worked example

About 70 percent of statistics students hand in homework on time, each independently. In a class of 50 students, what is the probability that at least 40 do? The book walks through the setup as questions (a) to (f).

(a) Why binomial

Why: each student either does or does not hand in on time, there are a fixed 50 students, and the chance is 0.70 for each.

\[ \text{success or failure}, \quad n = 50, \quad p = 0.70 \]

(b) Define X

Why: count the students who hand in on time.

\[ X = \text{number of students who hand homework in on time} \]

(c) The values X can take

Why: from nobody to the whole class.

\[ x = 0, 1, 2, \ldots, 50 \]

Listing the values is not busywork. It tells you where the distribution ends, which is exactly what you need when a range like at least 40 has to stop somewhere, and it reminds you that X counts people, so it is always a whole number.

(d) Failure, in words

Why: the outcome not counted.

\[ \text{failure} = \text{a student who does not hand in on time} \]

(e) Find q

Why: the complement of p.

\[ q = 1 - 0.70 = 0.30 \]

(f) Translate at least 40

Why: at least includes 40 itself.

\[ P(x \ge 40) \]

Written as one line, the whole setup is below. It took six small steps, and none of them was arithmetic beyond one subtraction. This is typical: the setup is where the thinking is, and the computation, done in Part 5, is mechanical.

\[ X \sim B(50,\, 0.70), \quad P(x \ge 40) = P(x = 40) + \cdots + P(x = 50) \]

Check the translation

Why: the question's counts are 40 up to 50, eleven values, and 40 is included because at least is not strict.

\[ 50 - 40 + 1 = 11 \text{ counts} \]

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, pp. 238-239 — Example 4.12

23. Try It 4.11: which inequality is more than three times?

Prediction

A fair six-sided die is rolled ten times, independently. You want the probability of rolling a one more than three times. X is the number of ones, and X follows B(10, 1/6).

Predict first

Which statement of the question is correct?

  • P(x > 3), the same as P(x ≥ 4)
  • P(x ≥ 3)
  • P(x < 3)
  • P(x = 3)

Correct: P(x > 3), the same as P(x ≥ 4)

Why: More than three excludes 3, so the counts included are 4, 5, and so on up to 10. Because x is a whole number, x greater than 3 and x at least 4 describe exactly the same counts.

The version with the boundary included is the common slip: it quietly adds P(x = 3), which here is about 0.155, more than twice the correct answer of about 0.070. The boundary bar is often one of the largest bars in the tail, so getting it wrong is not a rounding matter.

Part 5 computes this probability. For now, the translation alone is the skill: a strict phrase, a strict inequality, and a list of counts starting at 4.

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, p. 238 — Try It 4.11

24. Try It 4.12: which are genuine reasons it is binomial?

Discrimination

Sixty-five percent of people pass the state driver's exam on the first try. A group of 50 people who have taken it is chosen at random, and X is the number who passed first time. The book asks for two reasons this is binomial. Sort these statements.

Sort into buckets

Sort each statement: a genuine reason the problem is binomial, a true statement that is not a reason, or a false statement.

A genuine reason
There is a fixed number of trials, the 50 people; Each person either passed first time or did not; The people are chosen at random, so each has the same 0.65 chance, independently
True, but not a reason
65 percent is more than half; The expected number who passed is 32.5
False
Every person in the group must have passed
r
These are the three characteristics: a fixed number of trials, two outcomes each, and independent trials with the same probability. Any two of them answer the book's question; a complete answer gives all three.
t
The size of p has nothing to do with being binomial; any p between 0 and 1 works. And 50 times 0.65 is 32.5, the mean, which you can only compute once you already know it is binomial.
x
The random variable counts how many passed first time; it can be anything from 0 to 50. Nothing requires everyone to have passed.

The third genuine reason is the one the random selection supplies. Because the 50 are chosen at random from a large population of test takers, one person's result tells you nothing about the next, and each has the same 0.65 chance. Without random selection, say 50 people from the same driving school, independence would be doubtful.

The trap here is offering a correct calculation as a reason. The mean is a consequence of the model, so it cannot be evidence for the model. A reason must be one of the three characteristics, checked against the story.

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, p. 239 — Try It 4.12

25. The binomial formula

Section

Part 3

26. The probability of one particular sequence

Concept

Start small, with Joe answering three true-false questions, p equal to 0.6. Ask first about one specific sequence of results: right, right, wrong.

\[ P(\text{SSF}) = p \cdot p \cdot q \]

Independent trials multiply, exactly as in the warm-up. Each success contributes a factor of p and each failure a factor of q, in whatever order they come.

\[ P(\text{SSF}) = (0.6)(0.6)(0.4) = 0.144 \]

Now rearrange the letters: S F S, or F S S. The factors are the same three numbers in a different order, and multiplication does not care about order, so each of these sequences also has probability 0.144.

\[ P(\text{SFS}) = (0.6)(0.4)(0.6) = 0.144, \quad P(\text{FSS}) = (0.4)(0.6)(0.6) = 0.144 \]

That is the key observation: every sequence with the same number of successes has the same probability. With x successes and n minus x failures, it is p multiplied x times and q multiplied n minus x times.

\[ P(\text{one particular sequence}) = p^x q^{\,n-x} \]

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, p. 237 — Joe's p = 0.6 and q = 0.4

27. All eight sequences for three guesses

Picture it

Draw every possible sequence for Joe's three guesses as a tree. Each fork splits into a success, with 0.6, and a failure, with 0.4, and each path's probability is the product along it.

Figure (svg): A tree with three levels of branching for three true-false guesses. Each upper branch is a success with 0.6, each lower branch a failure with 0.4. The eight leaves read SSS 0.216, SSF 0.144, SFS 0.144, SFF 0.096, FSS 0.144, FSF 0.096, FFS 0.096, FFF 0.064; the three leaves with exactly two successes are highlighted.

The three highlighted paths each have two successes and one failure, and all three have the same probability, 0.144. Only where the failure sits changes from path to path.

Group the eight leaves by how many successes they have. One leaf has three, three leaves have two, three leaves have one, one leaf has none. The three highlighted leaves are the ways to get exactly two right, and each carries 0.144.

\[ P(x = 2) = 3 \times 0.144 = 0.432 \]

This is the whole binomial idea in one picture: the probability of a count is the number of sequences that give that count, times the probability of any one of them. The only hard part is counting the sequences when the tree is too big to draw, which it is by n equal to 10, with 1,024 leaves.

28. Counting the orderings

Concept

How many sequences of n trials contain exactly x successes? Each such sequence is fixed by choosing which x of the n trials are the successes, so the question is how many ways there are to choose x positions out of n.

Figure (svg): Ten rows of five circles, each row a sequence of five coin flips with exactly two heads highlighted: heads on flips 1 and 2, 1 and 3, and so on through 4 and 5. A note says each row has probability 0.25 squared times 0.75 cubed.

Choosing an ordering is choosing which two of the five flips are heads. Listing them methodically, first head on flip 1, then on flip 2, and so on, gives 4, 3, 2 and 1 rows: 10 in all.

For two heads in five flips, list them methodically, and you get 10. The general count has a name and a formula.

\[ \binom{n}{x} = \frac{n!}{x!\,(n-x)!} \]

Read the left side as n choose x. The exclamation mark is factorial: 5 factorial is 5 times 4 times 3 times 2 times 1, which is 120. Check it on the picture.

\[ \binom{5}{2} = \frac{5!}{2!\,3!} = \frac{120}{2 \cdot 6} = 10 \]

And on Joe's tree, where three leaves had two successes:

\[ \binom{3}{2} = \frac{3!}{2!\,1!} = \frac{6}{2} = 3 \]

Why does the formula divide? The n factorial on top counts every way to line up all n trials as if each were different. But the x successes are interchangeable with each other, and so are the failures, so each genuinely different sequence has been counted x factorial times n minus x factorial times. Dividing removes the repeats.

Many calculators have this as the nCr key. The book leaves it to the calculator; the counting is what it means.

29. Pascal's triangle holds every count

Picture it

Line the counts up by row and you get Pascal's triangle. Row n lists the number of orderings with 0, 1, 2 and so on up to n successes.

Figure (svg): Pascal's triangle for rows n equals 0 to 5: 1; 1 1; 1 2 1; 1 3 3 1; 1 4 6 4 1; 1 5 10 10 5 1. The entries 3 in row 3, and 10 and 5 in row 5, are highlighted and labelled with their n and x.

Each entry is the number of orderings for its row n and position x, counting from x equals 0 at the left. Each entry is also the sum of the two above it: an ordering either starts with a success or it does not.

Row 3 is 1, 3, 3, 1: one way to get no successes, three ways to get one, three to get two, one to get three. That is exactly how Joe's eight leaves split, and 1 plus 3 plus 3 plus 1 is 8.

\[ \binom{n}{0} = \binom{n}{n} = 1, \qquad \binom{n}{x} = \binom{n}{n-x} \]

Two facts are visible without any formula. Each row starts and ends with 1, because there is only one way to have no successes and one way to have all of them. And each row is symmetric, because choosing which trials succeed is the same as choosing which fail.

Each entry is also the sum of the two above it. That is not a coincidence: a sequence of n trials with x successes either starts with a success, leaving x minus 1 to place in the rest, or starts with a failure, leaving x to place.

30. The binomial formula, factor by factor

Notation

Put the two pieces together: the number of sequences times the probability of one of them. This is the formula printed under the book's Figure 4.2.

Annotate

On: \( P(x) = \binom{n}{x}\, p^{x}\, q^{\,n-x} \)

  • The probability that the count of successes is exactly x, for x from 0 to n.
  • How many different sequences contain exactly x successes: which x of the n trials are the successes.
  • One factor of p for each of the x successes in a sequence.
  • One factor of q for each of the remaining trials, the failures. The two exponents always add to n.

The last two factors together are the probability of one particular sequence; the first factor counts how many such sequences there are. Every binomial probability is that product, and there is no other idea in it.

Say the formula as a sentence and it stops being a string of symbols: the number of ways to place the successes, times the chance of the successes, times the chance of the failures. Each factor answers one question, and a missing factor means an unanswered question.

Two quick checks catch most errors before you compute. The exponents must add to n, because every trial is a success or a failure. And p goes with the successes: if you are counting failures, you have chosen the wrong success, and should swap p and q.

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, p. 238 — the formula with Figure 4.2

31. Your own: Joe's three guesses, the whole distribution

Worked example

Build the full distribution of X ~ B(3, 0.6), the number of Joe's three guesses that are right, from the formula. Then check it against the tree.

x = 0

Why: one ordering, three failures.

\[ P(0) = \binom{3}{0}(0.6)^0(0.4)^3 = 1 \cdot 1 \cdot 0.064 = 0.064 \]

Each row uses the same three-factor recipe. Only the count out front and the split of the exponents change as x moves along, and the exponents always add to 3.

x = 1

Why: three orderings, one success and two failures.

\[ P(1) = \binom{3}{1}(0.6)^1(0.4)^2 = 3 \cdot 0.6 \cdot 0.16 = 0.288 \]

x = 2

Why: three orderings, two successes and one failure.

\[ P(2) = \binom{3}{2}(0.6)^2(0.4)^1 = 3 \cdot 0.36 \cdot 0.4 = 0.432 \]

Compare the rows for one and two. Both have three orderings, so the count factor cannot tell them apart; the difference comes entirely from swapping one q for one p. Since p is the larger number, two rights beats one right.

x = 3

Why: one ordering, three successes.

\[ P(3) = \binom{3}{3}(0.6)^3(0.4)^0 = 1 \cdot 0.216 \cdot 1 = 0.216 \]

Anything to the power zero is 1, so the end rows are just one factor each. The table is a genuine Section 4.1 distribution, and the picture shows its shape.

Figure (svg): A probability histogram of B(3, 0.6): bars at 0, 1, 2 and 3 of heights 0.064, 0.288, 0.432 and 0.216. A dashed line at 1.8 marks the mean.

Two correct is the most likely count, and the bars lean left of 3: with p above one half, successes are more common than failures, so the heavy end is the high end.

Check that the probabilities add to one

Why: every possible count is listed exactly once, so they must; the leaves of the tree add to the same total.

\[ 0.064 + 0.288 + 0.432 + 0.216 = 1 \]

32. Example 4.11: the altered coin, more than 3 heads

Worked example

A coin is altered so that heads comes up with probability 0.25, and it is flipped 5 times, independently. Find the probability of more than 3 heads. The book calls it the fair coin in one line; it is not, and p is 0.25 throughout.

Set up

Why: X is the number of heads, a success is a head.

\[ X \sim B(5,\, 0.25), \quad q = 0.75 \]

Translate more than 3

Why: a strict inequality: 4 or 5 heads.

\[ P(x > 3) = P(x = 4) + P(x = 5) \]

The book develops the whole distribution first and reads the answer from it. By hand you only need the two bars the question asks for, because the counts 4 and 5 are the only ones above 3.

Compute P(x = 4)

Why: 5 orderings, four heads, one tail.

\[ P(4) = \binom{5}{4}(0.25)^4(0.75)^1 = 5(0.00390625)(0.75) \]

Multiply

Why: keep full precision until the end.

\[ P(4) = 0.0146484 \approx 0.0146 \]

Five orderings: the single tail can be any of the five flips. Each ordering has probability 0.0029, and five of them make 0.0146.

Compute P(x = 5)

Why: one ordering, five heads.

\[ P(5) = \binom{5}{5}(0.25)^5(0.75)^0 = 0.0009766 \approx 0.0010 \]

Add the two

Why: the addition rule: the counts 4 and 5 cannot both happen.

\[ P(x > 3) = 0.0146484 + 0.0009766 = 0.015625 \]

In context: with this coin, more than three heads in five flips happens only about 1.6 percent of the time, roughly once in 64 sets of five flips.

Figure (svg): Figure 4.2 redrawn: a probability histogram of B(5, 0.25) with bars at 0 to 5 of heights 0.2373, 0.3955, 0.2637, 0.0879, 0.0146 and 0.0010. The two bars at 4 and 5 are highlighted in pink. A dashed line at 1.25 marks the mean.

The question asks for the two pink bars, which are barely visible. The mass sits at 0, 1 and 2 and the bars trail off to the right: skewed right, because p is well below one half.

Check the book's figures

Why: the book prints P(x = 5) as 0.0007 and the sum as 0.0153; 0.25 to the fifth is 0.0009766, so the correct values are 0.0010 and 0.0156. The whole table does add to one.

\[ 0.2373 + 0.3955 + 0.2637 + 0.0879 + 0.0146 + 0.0010 = 1.0000 \]

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, p. 238 — Example 4.11 and Figure 4.2

33. Four attempts at P(x = 4) for B(5, 0.25)

Error analysis

The book's own formula line for this example has a typo, and students make four characteristic slips here. The correct value is 0.0146. Read each attempt and say what went wrong.

Annotate

On: \( \text{correct: } \binom{5}{4}(0.25)^4(0.75)^1 = 0.0146 \)

  • Computed 0.25 to the 4th times 0.75 and stopped. That is the probability of one sequence, such as H H H H T. The factor 5, the number of orderings, is missing.
  • Computed 5 times 0.25 times 0.75 to the 4th. The exponents are swapped, and this is P(x = 1), one head and four tails. Four heads means p to the 4th.
  • Used 0.25 to the 4th times 0.25, putting p on the failure too. The failure factor must be q, 0.75.
  • Correct number, wrong question: at least 4 also needs P(x = 5), giving 0.0156.

Attempts two and three are errors of bookkeeping: which probability goes with which outcome, and how many times. Say it in words and they vanish: four heads at 0.25 each, one tail at 0.75, and five places the tail could be.

The first attempt is by far the most common. It feels complete because it uses every number in the problem. The test that catches it: ask how many sequences have four heads. If the answer is more than one, the formula needs a count out front.

And the book's typo: the line under Figure 4.2 reads 25 to the x times 75 to the 5 minus x, with the decimal points lost. Those are 0.25 and 0.75; 25 to any power is not a probability.

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, p. 238 — the formula under Figure 4.2

34. Where is the tallest bar of B(5, 0.25)?

Prediction

Before looking at the next slide's numbers again, reason about the shape of X ~ B(5, 0.25) from p alone. One head in four flips on average means the count should cluster low.

Predict first

Which count of heads is most likely with this coin?

  • 1 head
  • 0 heads
  • 2 heads
  • all counts are equally likely

Correct: 1 head

Why: The mean is 5 times 0.25, which is 1.25, and the tallest bar sits next to the mean. P(1) is about 0.3955, higher than P(0) at 0.2373 and P(2) at 0.2637.

Zero heads is tempting, because no single flip is likely to be a head. But there are five different ways to get exactly one head and only one way to get none, and that factor of five outweighs the smaller single-sequence probability.

\[ \frac{P(1)}{P(0)} = \frac{5(0.25)(0.75)^4}{(0.75)^5} = \frac{5(0.25)}{0.75} \approx 1.67 \]

That ratio is the counting factor fighting the probability factor, and here counting wins. The same competition decides the shape of every binomial distribution.

Push the same comparison one more step, from one head to two. There are 10 orderings instead of 5, but each swaps a 0.75 for a 0.25, a factor of one third. Doubling the count and then taking a third leaves two thirds, so P(2) is smaller than P(1), and the bars fall from 1 onward.

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, p. 238 — Figure 4.2

35. Example 4.10 finished: exactly 15 wins out of 20

Worked example

Back to the game you win 55 percent of the time. The setup was X ~ B(20, 0.55) and the question P(x = 15). The book stops at the setup; here is the number.

Write the formula

Why: 15 successes, 5 failures, and the exponents add to 20.

\[ P(15) = \binom{20}{15}(0.55)^{15}(0.45)^{5} \]

Count the orderings

Why: which 15 of the 20 games are wins.

\[ \binom{20}{15} = \frac{20!}{15!\,5!} = 15{,}504 \]

Over fifteen thousand different win-loss sequences each give exactly 15 wins. That huge count is why this probability is not tiny even though any one sequence is.

One sequence

Why: fifteen factors of 0.55 and five of 0.45.

\[ (0.55)^{15} \approx 0.00012748, \quad (0.45)^{5} \approx 0.018453 \]

Keep several significant figures in these powers. They are small, and rounding 0.00012748 to 0.0001 would throw the final answer off by more than a fifth. Round only at the last line.

Multiply the three factors

Why: count times one sequence.

\[ P(15) \approx 15{,}504 \times 0.00012748 \times 0.018453 \approx 0.0365 \]

In context: winning exactly 15 of 20 games happens about 3.6 percent of the time, roughly once in 27 sets of twenty games.

Check with the mean and spread

Why: the expected number of wins is np = 11 with standard deviation about 2.22, so 15 is almost two standard deviations above the mean, where a bar should be small but visible.

\[ \frac{15 - 11}{\sqrt{20(0.55)(0.45)}} = \frac{4}{2.225} \approx 1.8 \]

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, p. 237 — Example 4.10

36. Try It 4.10 finished: the dolphin catches 12 of 20

Worked example

The setup was X ~ B(20, 0.35), with q = 0.65, and the question P(x = 12).

Write the formula

Why: 12 successes, 8 failures.

\[ P(12) = \binom{20}{12}(0.35)^{12}(0.65)^{8} \]

Count the orderings

Why: which 12 of the 20 attempts are catches.

\[ \binom{20}{12} = \frac{20!}{12!\,8!} = 125{,}970 \]

By the symmetry of Pascal's triangle this is the same as 20 choose 8: choosing the 12 catches is the same as choosing the 8 misses. Cancel the 12 factorial and you only multiply eight numbers on top, 20 down to 13, and divide by 8 factorial.

One sequence

Why: twelve factors of 0.35 and eight of 0.65.

\[ (0.35)^{12} \approx 3.3792 \times 10^{-6}, \quad (0.65)^{8} \approx 0.031864 \]

Twelve factors of 0.35 is a very small number, about three in a million. The count of orderings, over a hundred thousand, pulls it back up.

Multiply the three factors

Why: count times one sequence.

\[ P(12) \approx 125{,}970 \times 3.3792 \times 10^{-6} \times 0.031864 \approx 0.0136 \]

In context: a dolphin that catches fish 35 percent of the time will catch exactly 12 in 20 attempts only about 1.4 percent of the time. If the trainer saw 12, it would be a sign the dolphin has improved.

Check with the mean

Why: np = 7 and the standard deviation is about 2.13, so 12 is well over two standard deviations above the mean, consistent with a probability near one percent.

\[ \frac{12 - 7}{\sqrt{20(0.35)(0.65)}} = \frac{5}{2.133} \approx 2.3 \]

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, p. 237 — Try It 4.10

37. The shape depends on p

Picture it

Hold n at 10 and change p. The formula is the same; the shape moves.

Figure (svg): Three probability histograms for n equals 10: p equals 0.2 with the bars piled at the left and a long right tail; p equals 0.5 symmetric about 5; p equals 0.8, the mirror image of the first, piled at the right. A dashed line in each marks the mean at 2, 5 and 8.

The first and last panels are mirror images: swapping p and q swaps success and failure, which reads the count from the other end. Only p equal to one half gives a symmetric distribution.

When p is small, successes are rare, the bars pile up at the low counts and a thin tail runs right: skewed right, like the altered coin. When p is one half, the counting factor alone sets the shape, and Pascal's row makes it symmetric. When p is large the picture flips.

The dashed line in each panel is the mean, at 2, 5 and 8, which is ten times p. In the skewed panels the tallest bar sits right at the mean, but the long tail sits on the side away from the pile, just as the skewed data sets of Chapter 2 did.

The book notes this for Figure 4.2: the distribution is skewed right because p is 0.25, not 0.5. Knowing the shape in advance lets you sanity-check a table. If the bars of a distribution with p equal to 0.2 were tallest at the right, you would know something was wrong before checking a single number.

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, p. 238 — the skew of Figure 4.2

38. The mean and standard deviation

Section

Part 4

39. Why the mean is np

Concept

You could find the mean of any binomial by building the Section 4.2 table and adding the value-times-probability column. For B(5, 0.25) that gives the following.

\[ \mu = 0(0.2373) + 1(0.3955) + 2(0.2637) + 3(0.0879) + 4(0.0146) + 5(0.0010) \]

\[ \mu = 0 + 0.3955 + 0.5273 + 0.2637 + 0.0586 + 0.0049 = 1.25 \]

Using unrounded probabilities the total is exactly 1.25. And 1.25 is 5 times 0.25. That is no accident. A binomial count is the sum of n Bernoulli trials, and each one has mean p, as you saw in Part 1. Adding n trials adds n means.

\[ \mu = \underbrace{p + p + \cdots + p}_{n \text{ trials}} = np \]

The long table and the short formula must agree, because they compute the same thing. The table is still the definition of the mean; np is a shortcut that holds for every binomial, proved once so that you never need the table again.

This matches intuition, too. If one flip gives a quarter of a head on average, five flips give five quarters. If Joe gets 0.6 of a question right on average, over 10 questions he averages 6.

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, p. 238 — mu = np = 1.25 for Example 4.11

40. Why the variance is npq

Concept

Find the variance of one Bernoulli trial with the Section 4.2 table. The values are 0 and 1, and the mean is p.

\[ \sigma^2_{\text{one}} = (0 - p)^2 q + (1 - p)^2 p \]

Replace one minus p by q, since they are equal.

\[ \sigma^2_{\text{one}} = p^2 q + q^2 p \]

Both terms contain pq; factor it out, and what is left is p plus q, which is one.

\[ \sigma^2_{\text{one}} = pq(p + q) = pq \]

For independent trials, variances add, just as means do, so n trials have n times the variance of one. The standard deviation is the square root, as always.

\[ \sigma^2 = npq, \qquad \sigma = \sqrt{npq} \]

Check on the coin: 5 times 0.25 times 0.75 is 0.9375, and the long table computation gives the same 0.9375. Two numbers, n and p, have replaced a six-row table.

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, p. 237 — mu = np and sigma squared = npq

41. Reading σ = √(npq)

Notation

Three letters under a root, each doing a job you can see on a picture.

Annotate

On: \( \mu = np, \qquad \sigma = \sqrt{n\,p\,q} \)

  • The long-run average count of successes, in the units of the problem: workers, heads, games won. It need not be a whole number.
  • More trials, more spread in the count. Doubling n doubles the variance, so it multiplies σ by the square root of 2, not by 2.
  • The variance of one trial. It is largest, 0.25, when p is one half, and it shrinks to zero as p approaches 0 or 1, where the outcome becomes certain.
  • Brings the variance back to the units of the count. Forgetting it is the most common error in this part.

Notice the unequal growth. The mean grows in proportion to n, but the spread grows only with the square root of n. So as n grows, the count gets relatively more predictable: its typical distance from the mean shrinks compared with the mean itself.

That observation is the seed of Chapter 7, where the same square root makes sample averages settle down.

42. Judging an answer by how many σ away it is

Concept

Every check step in this deck asks the same question: how far is the count in the question from the mean, measured in standard deviations? That one number tells you roughly what size of answer to expect, before you trust a calculator.

\[ \text{distance} = \frac{\text{count} - \mu}{\sigma} \]

For a binomial with a hump-shaped distribution, most of the probability lies within about two standard deviations of the mean. So a rough guide works well enough to catch errors.

boundary sitsa tail beyond it holds roughlyexample from this section
about 0.5 σ from μa third or moremore than 50 of 80 shots: 0.37
about 1.5 σ from μseveral percentat least 40 of 50: 0.079
about 2.5 σ from μaround one percent or lessexactly 12 dolphin catches: 0.014
3 σ or more from μa fraction of a percentat most 20 of 50 savers: 0.0034

This is a guide, not a formula: skewed distributions, like the cancer example with its tiny p, have uneven tails. But it is enough to catch the big mistakes, a swapped exponent or a complement taken on the wrong side, which usually miss by a factor of ten or land on the wrong side of one half.

Chapter 6 makes this idea exact for the normal distribution, where the distance in standard deviations gets a name, the z-score.

43. Example 4.13 (part 2): how many workers to expect

Worked example

About 41 percent of adult workers have a high school diploma but no further education. Twenty workers are chosen at random. How many would you expect to have only a diploma, and how far from that would a typical sample be?

Set up

Why: X is the number of the 20 with only a diploma.

\[ X \sim B(20,\, 0.41), \quad q = 1 - 0.41 = 0.59 \]

The mean

Why: n times p.

\[ \mu = np = (20)(0.41) = 8.2 \]

In context: in samples of 20 workers, about 8.2 on average have only a high school diploma. No sample has 8.2 people; this is the long-run average across many samples.

Compare this with Section 4.2. There, a mean needed a full table with a row for every value, here 21 rows. The binomial formula replaces all of that with one multiplication, because the table is already known to follow the binomial pattern.

The variance

Why: n times p times q.

\[ \sigma^2 = npq = (20)(0.41)(0.59) = 4.838 \]

The standard deviation

Why: the square root of the variance.

\[ \sigma = \sqrt{4.838} \approx 2.20 \]

Figure (svg): A probability histogram of B(20, 0.41) for x from 0 to 20. The bars rise to a peak of about 0.179 at 8 and fall away on both sides. A dashed line at 8.2 marks the mean, and a blue band from 6.0 to 10.4 shows one standard deviation either side.

With n equal to 20 the shape is already close to a hump. Most of the probability lies within about two workers of 8.2, and almost none lies beyond 14.

The blue band runs from about 6 to about 10.4 workers. Counts in that band are typical; a sample with only 2 or as many as 15 such workers would be unusual.

Check against the picture

Why: the tallest bar should sit near μ, and it does: the bar at 8 is the highest, at about 0.179, with 9 close behind.

\[ P(8) \approx 0.1790, \quad P(9) \approx 0.1658 \]

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, pp. 239-240 — Example 4.13 and Figure 4.3

44. Example 4.14: signature-artist pages

Worked example

A 560-page art catalogue has 8 pages featuring signature artists. A random sample of 100 pages is taken; X is the number featuring signature artists. The book uses a binomial model with p equal to 8 over 560.

(b) The distribution

Why: each sampled page is a signature page with probability 8/560.

\[ X \sim B\left(100,\, \tfrac{8}{560}\right), \quad p \approx 0.014286 \]

On (a) the book lists x from 0 to 8, which is right for the real catalogue, since only 8 such pages exist. But the binomial model allows any count from 0 to 100. The model is an approximation here; its probabilities beyond 8 are so small, together about 0.00001, that it does not matter.

(i) Exactly two pages

Why: 4,950 orderings of two successes in 100 trials.

\[ P(2) = \binom{100}{2}\left(\tfrac{8}{560}\right)^2\left(\tfrac{552}{560}\right)^{98} \approx 4950(0.000204)(0.2441) \approx 0.2466 \]

That power of 98 looks alarming, but it is just the probability that 98 particular pages are all ordinary. Each is ordinary with probability 552 over 560, very close to one, so even 98 of them multiply to about a quarter.

(iii) More than three

Why: complement of at most three, which has only four terms.

\[ P(x > 3) = 1 - \left[P(0) + P(1) + P(2) + P(3)\right] \]

Add the four terms

Why: each from the formula.

\[ 0.2372 + 0.3438 + 0.2466 + 0.1168 = 0.9443 \]

Subtract from one

Why: the complement.

\[ P(x > 3) = 1 - 0.9443 = 0.0557 \]

In context: in about 5.6 percent of random samples of 100 pages, more than three would feature signature artists. With a mean below one and a half, four or more is unusual.

(ii) At most six

Why: one minus the tiny tail from 7 up.

\[ P(x \le 6) = 1 - P(x \ge 7) \approx 1 - 0.0006 = 0.9994 \]

(c) Mean and standard deviation

Why: the two formulas.

\[ \mu = 100 \cdot \tfrac{8}{560} \approx 1.4286, \quad \sigma = \sqrt{100 \cdot \tfrac{8}{560} \cdot \tfrac{552}{560}} \approx 1.1867 \]

Figure (svg): A probability histogram of B(100, 8/560) for x from 0 to 8: heights 0.2372, 0.3438, 0.2466, 0.1168, 0.0410, 0.0114, 0.0026, 0.0005 and 0.0001. The bar at 2 is yellow; the bars at 4 and above are pink. The mean 1.43 is marked.

The model allows up to 100 such pages, but the bars beyond 8 are too small to draw. The yellow bar is part (i); the pink bars together are part (iii).

Check the answers against each other and the book

Why: the mean of 1.43 sits between the two tallest bars at 1 and 2, as it should, and all five values agree with the book's printed answers to four places.

\[ 0.2466, \; 0.9994, \; 0.0557, \; 1.4286, \; 1.1867 \]

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, pp. 240-241 — Example 4.14

45. Example 4.15: the lifetime risk of cancer

Worked example

The lifetime risk of developing cancer is about 1 in 67, or 1.5 percent. A random sample of 200 people is taken; X is the number who will develop cancer.

(a) The distribution

Why: 200 independent people, each with p = 0.015.

\[ X \sim B(200,\, 0.015) \]

(b)(i) The mean

Why: n times p.

\[ \mu = 200(0.015) = 3 \]

(b)(ii) The standard deviation

Why: square root of npq.

\[ \sigma = \sqrt{200(0.015)(0.985)} = \sqrt{2.955} \approx 1.719 \]

In context: samples of 200 people will contain 3 future cancer cases on average, typically give or take about 1.7.

(c) At most eight

Why: add P(0) through P(8); the book uses the calculator's binomcdf.

\[ P(x \le 8) = P(0) + P(1) + \cdots + P(8) \approx 0.9965 \]

Eight is about three standard deviations above the mean of 3, so nearly every sample falls at or below it. The remaining 0.0035 is the chance that a sample of 200 contains nine or more future cases.

(d) Five people

Why: the formula with x = 5.

\[ P(5) = \binom{200}{5}(0.015)^5(0.985)^{195} \approx 0.1011 \]

(d) Six people

Why: the formula with x = 6.

\[ P(6) = \binom{200}{6}(0.015)^6(0.985)^{194} \approx 0.0500 \]

So five is about twice as likely as six. A sentence that answers the question with the numbers: it is more likely that five people develop cancer, probability about 0.101, than six, probability about 0.050.

Figure (svg): A probability histogram of B(200, 0.015) for x from 0 to 10. Heights rise from 0.0487 at 0 to about 0.2257 at 3 and fall to 0.0007 at 10. The bars at 5 and 6 are highlighted, labelled 0.1011 and 0.0500. A dashed line at 3 marks the mean.

Five and six are both on the falling side of the peak, and each step to the right roughly halves the bar there. Past the mean, the next count is always less likely than the one before.

Check the ratio without the big coefficients

Why: dividing P(6) by P(5), almost everything cancels, leaving one short product that agrees with 0.0500 over 0.1011.

\[ \frac{P(6)}{P(5)} = \frac{195}{6} \cdot \frac{0.015}{0.985} \approx 0.495 \]

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, p. 241 — Example 4.15

46. Try It 4.15: estimate before you compute

Estimation

One season, a Los Angeles Clippers player scored with 61.3 percent of his shots. You choose a random sample of 80 of his shots; X is the number that scored. Estimate before touching a calculator.

Predict first

About how many of the 80 shots would you expect to score, and roughly how many either side counts as typical?

  • about 49, give or take about 4
  • about 61, give or take about 6
  • about 49, give or take about 19
  • about 40, give or take about 4

Correct: about 49, give or take about 4

Why: Sixty percent of 80 is 48, and 61.3 percent is a little more, 49.04. The variance is 80 times 0.613 times 0.387, about 19, and its square root is about 4.4. The 19 is the variance, not the spread.

The 61 option confuses the percentage with the count; it would be right only for 100 shots. The give or take 19 option stops at the variance and forgets the square root. The 40 option is half the shots, which ignores p entirely.

Rough mental arithmetic like this, about 60 percent of 80 and about the square root of 20, is the fastest defence against a calculator slip. The next slide computes it properly and uses these two numbers to judge a probability.

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, pp. 241-242 — Try It 4.15

47. Try It 4.15: the shooter's 80 shots

Worked example

X is the number of the 80 sampled shots that scored, with each shot scoring with probability 0.613.

(a) The distribution

Why: 80 shots, success = scoring.

\[ X \sim B(80,\, 0.613), \quad q = 0.387 \]

(b)(i) The mean

Why: n times p.

\[ \mu = 80(0.613) = 49.04 \]

(b)(ii) The standard deviation

Why: square root of npq.

\[ \sigma = \sqrt{80(0.613)(0.387)} = \sqrt{18.978} \approx 4.356 \]

Both numbers match the estimate from the previous slide, about 49 give or take about 4. That agreement is the point of estimating first: if the calculator had said 61 or 19, you would know at once that something was typed wrong.

(c) Exactly 60 scored

Why: the formula; the coefficient is enormous, so a calculator does the product.

\[ P(60) = \binom{80}{60}(0.613)^{60}(0.387)^{20} \approx 0.0036 \]

Sixty is about 2.5 standard deviations above the mean, which is why exactly 60 is so unlikely. On any sample of 80 shots, something near 49 is what you would see.

(d) More than 50

Why: complement of at most 50.

\[ P(x > 50) = 1 - P(x \le 50) \approx 1 - 0.6282 = 0.3718 \]

In context: in about 37 percent of random samples of 80 of his shots, more than 50 would have scored. That is plausible, since 50 is only a little above the mean of 49.04.

Check with the spread

Why: 51 and up starts less than half a standard deviation above the mean, so a probability a bit under one half is right; 60 sits 2.5 standard deviations out, so a fraction of a percent is right.

\[ \frac{50.5 - 49.04}{4.356} \approx 0.34, \quad \frac{60 - 49.04}{4.356} \approx 2.5 \]

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, pp. 241-242 — Try It 4.15

48. Which p gives the most spread?

Prediction

Fix n at 20. For which success probability is the count most spread out, that is, the standard deviation largest?

Predict first

For X with n = 20, which p makes σ largest?

  • p = 0.5
  • p = 0.9
  • p = 0.1
  • σ is the same for every p

Correct: p = 0.5

Why: Sigma is the square root of 20 times p times q. The product pq is largest when p and q are equal, at 0.5 times 0.5, which is 0.25. At p = 0.1 or 0.9 the product is only 0.09.

Figure (svg): A graph of p times q, that is p times 1 minus p, for p from 0 to 1. It is an upside-down parabola, zero at both ends and highest, 0.25, at p equals 0.5. Points are marked at p equals 0.41 with value 0.2419 and p equals 0.015 with value 0.0148.

The variance of one trial, pq, is what npq multiplies by n. It is largest when success and failure are equally likely and shrinks to zero as the outcome becomes certain either way.

The graph of pq explains it: a hill, zero at both ends, peaking at one half. When success is nearly certain or nearly impossible, almost every trial goes the same way and the count hardly varies. When it is a coin flip, the count varies most.

The cancer example sits near the left foot of this hill, with pq about 0.015, which is why 200 trials gave a standard deviation of only 1.7. The worker example sits near the top, so 20 trials gave 2.2.

49. Slide n and p and watch the distribution move

Tweak it

The graph draws the binomial probabilities, one bar per count, for the n and p you choose. Start with n = 5 and p = 0.25, the altered coin, and change one thing at a time.

Parameter explorer

Set n = 20 and p = 0.41, the workers. Read the height of the tallest bar. Then move p to 0.5 and to 0.9 and describe what happens to the peak and the spread.

\[ X \sim B({n}, {p}), \quad \mu = np, \quad \sigma = \sqrt{npq} \]

  • n — from 1 to 30: number of trials n
  • p — from 0.05 to 0.95: probability of success p

With n = 20 and p = 0.41 the tallest bar is at 8, about 0.18 high, next to the mean 8.2. At p = 0.5 the bars centre on 10 and spread slightly more: the top of the pq hill. At p = 0.9 they crowd against 20, the spread shrinks and the peak climbs.

Try one more setting on purpose: n = 30 with p = 0.05. The bars crowd against zero, the tallest at 1, and there is a long thin tail to the right. That is the shape of the cancer example, rare events in many trials, and it is why its tails were uneven.

Now hold p at 0.5 and raise n from 5 to 30. The centre moves right in proportion to n, but the bars spread only slowly, and the peak gets lower because the same total probability of one is shared among more counts.

50. Trap: forgetting the square root

Trap

The trap

Example 4.15, cancer in samples of 200 at 1.5 percent. A first attempt at the standard deviation:

\[ \sigma = npq \]

\[ = 200(0.015)(0.985) = 2.955 \]

Then: a typical sample has 3 cases, give or take about 3. It sounds reasonable, which is what makes it dangerous.

Wrong. 2.955 is the variance.

The fix

npq is the variance, in squared units: cases squared. The standard deviation takes the square root to bring it back to cases.

\[ \sigma = \sqrt{2.955} \approx 1.72 \]

A quick sense check catches it whenever npq is bigger than 1: the root must be smaller than the variance. For B(50, 0.70), npq is 10.5 but σ is about 3.24, and a spread of 10.5 students around a mean of 35 would be absurd.

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, p. 237 — sigma is the square root of npq

51. Mean and spread for five distributions

Comparison

Fill in the missing means and standard deviations. Round σ to two decimal places. Each row is from this section.

Comparison matrix

problemnpμ = npσ = √(npq)
Ex 4.12, homework500.70353.24
Try It 4.14, savers500.60303.46
Try It 4.13, volunteers300.329.62.55
Ex 4.13, workers200.418.22.20
Ex 4.15, cancer2000.01531.72

\[ \sqrt{50(0.7)(0.3)} = \sqrt{10.5} \approx 3.24, \quad \sqrt{50(0.6)(0.4)} = \sqrt{12} \approx 3.46 \]

Compare the two rows with n equal to 50. Moving p from 0.70 to 0.60 moves it toward one half, so the spread grows a little, from 3.24 to 3.46, even though the mean falls. The mean and the spread respond to p differently.

The worker row is a good one to do by hand: 20 times 0.41 is 8.2, and 8.2 times 0.59 is 4.838, whose square root is just under 2.2. Each blank needs at most two multiplications and one root, never a table.

Compare the cancer row with the others: the largest n by far, but the smallest spread relative to n, because p is so close to zero. Every row confirms the hill in the pq graph.

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, pp. 238-242 — Examples 4.12-4.15 and Try Its 4.13-4.14

52. Cumulative probabilities

Section

Part 5

53. A range of counts is a sum of bars

Concept

A probability for a range of counts, such as at most 12, is the total of the bars for every count in the range. The counts are mutually exclusive, since the count cannot be 11 and 12 at once, so the addition rule applies.

\[ P(x \le 12) = P(0) + P(1) + P(2) + \cdots + P(12) \]

The book's calculator does this with binomcdf, the cumulative function, which always gives at most a value. The pdf, binompdf, gives exactly a value. Every range question has to be rewritten in terms of at most, or computed by hand.

questionrewritecalculator
P(x = 12)one barbinompdf(20, 0.41, 12)
P(x ≤ 12)bars 0 to 12binomcdf(20, 0.41, 12)
P(x > 12)1 − P(x ≤ 12)1 − binomcdf(20, 0.41, 12)
P(x ≥ 12)1 − P(x ≤ 11)1 − binomcdf(20, 0.41, 11)

The last row is the one to slow down on. At least 12 includes 12, so the part you remove is at most 11, not at most 12.

You do not need the calculator to follow any of this. Each row is a statement about which bars to add, and the bars come from the formula. The calculator names are worth knowing only because the book's answers are printed with them, so you can match your arithmetic to its lines.

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, pp. 239-240 — binompdf and binomcdf

54. The complement: add the short side

Concept

Every count from 0 to n is somewhere in the distribution, and all the probabilities add to one. So any range and its opposite add to one as well.

\[ P(x \le 12) + P(x > 12) = 1 \]

That gives you a choice. To find P(x at most 12) for B(20, 0.41), you could add the 13 bars from 0 to 12, or add the 8 bars from 13 to 20 and subtract from one.

Figure (svg): The same B(20, 0.41) histogram, with the bars from 0 to 12 in blue and the bars from 13 to 20 in red. A dashed divider at 12.5 separates them. Labels: blue total 0.9738, red total 0.0262.

The question asks for the whole blue region. Adding thirteen blue bars is tedious; the red tail has only eight bars, most of them almost zero, so subtract it from one instead.

\[ P(x \le 12) = 1 - P(x \ge 13) \]

Always add whichever side has fewer bars, or smaller ones. Here the right side has fewer terms, and most of them are almost zero, so the whole sum is dominated by two or three bars you can compute by hand.

The same rule gave the answer to the warm-up: at least one right is one minus all wrong. The complement is the one tool from Chapter 3 you will use more than any other in this part.

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, p. 240 — use 1 − binomcdf for more than

55. Example 4.13 (part 1): at most 12 of 20 workers

Worked example

Twenty workers, each with only a high school diploma with probability 0.41. Find the probability that at most 12 of them have only a diploma.

Set up and translate

Why: at most 12 includes 12.

\[ X \sim B(20,\, 0.41), \quad P(x \le 12) \]

Choose the shorter side

Why: 0 to 12 is thirteen bars; 13 to 20 is eight.

\[ P(x \le 12) = 1 - P(x \ge 13) \]

Before computing, note that the tail is far from the mean of 8.2, so its bars will be small and will shrink quickly. Most of the tail's total will come from its first two or three bars.

The largest tail bar

Why: the formula at x = 13.

\[ P(13) = \binom{20}{13}(0.41)^{13}(0.59)^{7} \approx 0.0178 \]

The rest of the tail

Why: each from the formula; they shrink fast.

\[ P(14) \approx 0.0062, \; P(15) \approx 0.0017, \; P(16) \approx 0.0004, \; P(17 \text{ to } 20) \approx 0.0001 \]

Add the tail

Why: the unrounded sum.

\[ P(x \ge 13) \approx 0.02622 \]

Subtract from one

Why: the complement.

\[ P(x \le 12) \approx 1 - 0.02622 = 0.9738 \]

In context: the probability that at most 12 of the 20 workers have only a high school diploma is about 0.974. The calculator line binomcdf(20, 0.41, 12) gives the same 0.9738.

Check by where 12 sits

Why: 12 is about 1.7 standard deviations above the mean of 8.2, so almost all the probability should lie at or below it: a value near 0.97 fits.

\[ \frac{12 - 8.2}{2.20} \approx 1.7 \]

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, pp. 239-240 — Example 4.13

56. Example 4.12 finished: at least 40 of 50 hand in on time

Worked example

The setup was X ~ B(50, 0.70) and the question P(x at least 40). The book gives the setup only; here is the probability.

Choose the side

Why: 40 to 50 is eleven bars; 0 to 39 is forty. Add the eleven directly.

\[ P(x \ge 40) = P(40) + P(41) + \cdots + P(50) \]

The first bar

Why: 10,272,278,170 orderings of 40 successes.

\[ P(40) = \binom{50}{40}(0.7)^{40}(0.3)^{10} \approx 0.0386 \]

Ten billion orderings, each with a probability of about four in a billion: the two huge numbers nearly cancel, leaving a bar of a few percent. This is why nobody computes these by hand in practice, and also why it is worth doing once, to see that the formula is all there is.

The next bars

Why: each from the formula.

\[ P(41) \approx 0.0220, \; P(42) \approx 0.0110, \; P(43) \approx 0.0048, \; P(44) \approx 0.0018 \]

The far tail

Why: 45 up to 50 together.

\[ P(45) + \cdots + P(50) \approx 0.0007 \]

Add all eleven

Why: the unrounded sum.

\[ P(x \ge 40) \approx 0.0789 \]

Figure (svg): A probability histogram of B(50, 0.7) drawn from x equals 22 to 50. The bars peak near 35. Bars from 40 to 50 are pink; bars up to 39 are blue. A dashed divider at 39.5. Labels: at least 40, 0.0789.

Forty is well above the mean of 35, so the pink region is a thin right tail. Notice that the bar at 40 itself is pink: at least includes it.

In context: in a class of 50 such students, there is only about an 8 percent chance that 40 or more hand their homework in on time. On a calculator this is 1 minus binomcdf(50, 0.7, 39); note the 39.

Check by where 40 sits

Why: the mean is 35 and σ is about 3.24, so 40 is about 1.5 standard deviations above the mean: a right tail of several percent is about right.

\[ \frac{40 - 35}{3.24} \approx 1.5 \]

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, pp. 238-239 — Example 4.12

57. Try It 4.11 finished: more than three ones in ten rolls

Worked example

A fair die is rolled ten times. X is the number of ones. Find the probability of rolling a one more than three times.

Set up and translate

Why: more than 3 is 4 or more.

\[ X \sim B\left(10,\, \tfrac16\right), \quad P(x > 3) = 1 - P(x \le 3) \]

P(0)

Why: no ones: ten failures.

\[ P(0) = \left(\tfrac56\right)^{10} \approx 0.1615 \]

Four bars against seven on the other side, so the complement wins. Each of the four needs its own count of orderings from row 10 of Pascal's triangle.

P(1)

Why: 10 orderings.

\[ P(1) = 10\left(\tfrac16\right)\left(\tfrac56\right)^{9} \approx 0.3230 \]

P(2)

Why: 45 orderings.

\[ P(2) = 45\left(\tfrac16\right)^2\left(\tfrac56\right)^{8} \approx 0.2907 \]

P(3)

Why: 120 orderings.

\[ P(3) = 120\left(\tfrac16\right)^3\left(\tfrac56\right)^{7} \approx 0.1550 \]

Add the four

Why: the short side.

\[ P(x \le 3) \approx 0.1615 + 0.3230 + 0.2907 + 0.1550 = 0.9303 \]

Subtract from one

Why: the complement.

\[ P(x > 3) \approx 1 - 0.9303 = 0.0697 \]

Figure (svg): A probability histogram of B(10, 1/6) for x from 0 to 10. Heights 0.1615, 0.3230, 0.2907, 0.1550 for 0 to 3 in blue, then 0.0543, 0.0130 and smaller values for 4 upward in pink. Label: more than 3, 0.0697.

The four blue bars hold 93 percent of the probability. The pink region has seven bars, the blue only four, so subtract the blue from one.

In context: in about 7 percent of sets of ten rolls you will see a one four or more times.

Check the coefficients against Pascal

Why: row 10 of Pascal's triangle begins 1, 10, 45, 120, the four counts used above; and the mean 10 times one sixth, about 1.67, sits between the two tallest bars at 1 and 2.

\[ \binom{10}{0}, \binom{10}{1}, \binom{10}{2}, \binom{10}{3} = 1, 10, 45, 120 \]

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, p. 238 — Try It 4.11

58. Try It 4.14: savers in a sample of 50

Worked example

A Gallup poll found 60 percent of American adults prefer saving to spending. X is the number of savers in a random sample of 50.

(a) The distribution

Why: 50 adults, success = prefers saving.

\[ X \sim B(50,\, 0.60) \]

(c) The mean and standard deviation first

Why: they tell you which answers to expect.

\[ \mu = 50(0.6) = 30, \quad \sigma = \sqrt{50(0.6)(0.4)} = \sqrt{12} \approx 3.46 \]

Working the mean and spread out before the probabilities is a deliberate reordering of the book's parts. With 30 and 3.46 in hand, each of the three probabilities can be judged the moment it appears.

(b)(i) Exactly 25

Why: the formula; 25 is about 1.4 standard deviations below the mean.

\[ P(25) = \binom{50}{25}(0.6)^{25}(0.4)^{25} \approx 0.0405 \]

(b)(ii) At most 20

Why: a far left tail: 20 is almost three standard deviations below the mean.

\[ P(x \le 20) = P(0) + \cdots + P(20) \approx 0.0034 \]

(b)(iii) More than 30

Why: complement of at most 30.

\[ P(x > 30) = 1 - P(x \le 30) \approx 1 - 0.5535 = 0.4465 \]

In context: in a random sample of 50 adults, there is about a 4 percent chance that exactly 25 prefer saving, a third of a percent chance that 20 or fewer do, and about a 45 percent chance that more than 30 do.

Why is more than 30 a bit under one half, when 30 is the mean? Because more than 30 excludes 30 itself, and the bar at 30 alone holds about 0.115. The two sides of the mean plus that middle bar make one.

Check the sizes against the spread

Why: 20 is nearly three standard deviations below 30, so a tail of a third of a percent fits; above 30 must be a little under one half, since the bar at the mean is removed.

\[ \frac{20 - 30}{3.46} \approx -2.9 \]

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, p. 241 — Try It 4.14

59. Trap: the off-by-one in a complement

Trap

The trap

Example 4.12 asks for at least 40 of 50. A first attempt with the calculator's at most function:

\[ P(x \ge 40) = 1 - P(x \le 40) \]

\[ = 1 - 0.9598 = 0.0402 \]

Wrong. The answer is about half what it should be.

The fix

One minus at most 40 removes the bar at 40 along with everything below it, but at least 40 includes 40. The part to remove is at most 39.

\[ P(x \ge 40) = 1 - P(x \le 39) \]

\[ = 1 - 0.9211 = 0.0789 \]

Check it with the listing: the counts wanted are 40 to 50; the counts removed must then be 0 to 39.

The missing 0.0386 is exactly P(40), the largest bar in the tail. To avoid this, list the counts you want, 40 to 50, then list the counts you are removing, 0 to 39. The two lists must not overlap and must not leave a gap.

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, p. 240 — the NOTE on using 1 − binomcdf

60. Add directly, or use the complement?

Sorting

For each question, decide the fastest way to compute it by hand: add the bars you want, or add the other side and subtract from one.

Sort into buckets

Sort each question by the side with fewer bars to add.

Add the bars directly
B(5, 0.25): P(x > 3); B(50, 0.7): P(x ≥ 40)
Use one minus the other side
B(20, 0.41): P(x ≤ 12); B(10, 1/6): P(x > 3); B(100, 8/560): P(x > 3); B(3, 0.6): P(x ≥ 1)
dir
More than 3 of 5 is just the two bars 4 and 5; at least 40 of 50 is eleven bars against forty on the other side.
comp
At most 12 of 20 is thirteen bars against eight. More than 3 of 10 is seven bars against four. More than 3 of 100 is ninety-seven bars against four. At least 1 of 3 is three bars against the single bar for zero.

The last item is the warm-up in disguise. At least one is always one minus none, and none is a single bar: q to the n. Whenever you see at least one, reach for the complement.

On a calculator the choice is made for you: binomcdf only gives at most, so any at least or more than question becomes a complement. By hand, count the bars first; it can turn ninety-seven terms into four.

61. Is the machine still making 2 percent defects?

Real world

A factory's machine makes a defective part 2 percent of the time. An inspector samples 100 parts and finds 7 defective. Should she worry? These numbers are an illustration, not from the book.

Discussion prompt

Model the number of defects in the sample, find its mean and standard deviation, and decide whether 7 defects is surprising if the machine is working as usual. Use numbers.

Figure (svg): A probability histogram of B(100, 0.02) for x from 0 to 10, peaked at 2. A blue band from 0.6 to 3.4 marks one standard deviation either side of the mean 2. The bar at 7 is red and labelled observed.

Seven defects sits more than three and a half standard deviations above the mean. If the machine really made 2 percent defects, a count that high would come up about 4 times in 1,000 samples.

Look at where the red bar sits in the picture: far outside the blue band, in a region where the bars are too short to see. The observed count is not just above average; it is in the part of the distribution the model says almost never happens.

The reasoning is exactly what you did in the check steps of this deck, turned around. There you used μ and σ to judge whether an answer was plausible. Here you use them to judge whether the model is still true. This is the logic of the hypothesis tests in Chapter 9.

Notice that the rate alone would not settle it: 7 percent against 2 percent sounds bad, but in a sample of 10 parts even one defect would be 10 percent. It is the count compared with np and σ that carries the evidence.

62. Order the steps of a binomial problem

Ranking

Put the steps in the order you would carry them out on a new problem. Several orderings are forced.

Put in order

Order the steps for solving a binomial probability problem.

  1. Define X in words and say what counts as a success
  2. Check the three characteristics: fixed n, two outcomes, independent trials with constant p
  3. Write X ~ B(n, p) and find q = 1 − p
  4. Translate the words into an inequality and list the counts included
  5. Choose the shorter side and add bars with the formula, or use the complement
  6. Check the answer against μ = np and σ = √(npq), then state it in context

Why: You cannot check the conditions until you know what is being counted, and you cannot write the distribution until the conditions pass. The inequality needs n to know where the counts end. The computation needs the list of counts, and the sanity check needs the answer.

The step most often skipped is the second one. A problem that fails a condition, like the committee draw, still produces a number if you push it through the formula; it is just the wrong number, and nothing in the arithmetic warns you.

The step most often rushed is the fourth. Writing the list of counts, not just the symbol, is what prevents the off-by-one error.

And the last step is the one that turns a number into an answer. A probability of 0.0789 means nothing on its own; the probability that at least 40 of 50 students hand in on time is 0.0789 is a statement someone can act on.

63. Putting it together

Section

Part 6

64. Pattern: recognise, set up, count, check

Pattern

Every problem in this section follows the same four moves. The formula is only the third of them.

  1. Recognise: fixed number of trials, two outcomes, independent trials with constant p. If any fails, stop: it is not binomial.
  2. Set up: X in words, the success in words, then X ~ B(n, p), q = 1 − p, and the question as an inequality with its list of counts.
  3. Count: each bar is (n choose x) pˣ qⁿ⁻ˣ, the number of orderings times the probability of one; a range is a sum of bars, done on the shorter side.
  4. Check: μ = np and σ = √(npq) say where the bulk of the distribution sits, so an answer can be judged by how many σ its boundary is from μ.
phrasecountsby complement
at least kk, k + 1, …, n1 − P(x ≤ k − 1)
more than kk + 1, …, n1 − P(x ≤ k)
at most k0, 1, …, k1 − P(x ≥ k + 1)
fewer than k0, 1, …, k − 11 − P(x ≥ k)

Read the complement column once more: every entry pairs a range with the opposite range, and the two never share a count. At least k pairs with at most k minus 1; more than k pairs with at most k. If your pair overlaps or leaves a gap, one of the two translations is wrong.

The final sentence always names the evidence and the context: a probability, the count it is about, and the population it describes. The number alone answers only half the question.

65. Check: a standard deviation

Check

Two formulas, one root. Keep variance and standard deviation apart.

Check your understanding

X ~ B(60, 0.25). What is the standard deviation of X?

  • A. about 3.35 (correct)
  • B. 11.25
  • C. about 3.87
  • D. 15

Answer: A

Why: The variance is 60 times 0.25 times 0.75, which is 11.25, and the standard deviation is its square root, about 3.35 successes.

Why B tempts people
That is npq, the variance. The standard deviation is its square root, back in units of successes.
Why C tempts people
That is the square root of np, 15. The formula needs q as well: the root of n times p times q.
Why D tempts people
That is the mean, np. It says where the distribution is centred, not how spread out it is.

\[ \sigma = \sqrt{60(0.25)(0.75)} = \sqrt{11.25} \approx 3.35 \]

An estimate rules out two answers at once: the spread must be much smaller than the mean of 15, since a count typically lands within a few of its mean. That leaves the two roots, and only one of them includes q.

If you chose the variance, go back to the trap in Part 4: whenever npq is bigger than one, σ must come out smaller than it.

66. Check: one binomial probability

Check

Now the formula itself. Count the orderings before you multiply.

Check your understanding

A fair coin is flipped 4 times. What is the probability of exactly 2 heads?

  • A. 0.375 (correct)
  • B. 0.0625
  • C. 0.5
  • D. 0.6875

Answer: A

Why: There are 6 orderings of two heads in four flips, and each has probability 0.5 to the fourth, one sixteenth. Six sixteenths is 0.375.

Why B tempts people
That is the probability of one particular ordering, such as heads, heads, tails, tails. There are six such orderings, so the count factor of 6 is missing.
Why C tempts people
That reasons that 2 is half of 4. The most likely count is 2, but it still has to share the probability with 0, 1, 3 and 4 heads.
Why D tempts people
That is the probability of at most 2 heads, 1 plus 4 plus 6 over 16. The question asks for exactly 2.

\[ P(2) = \binom{4}{2}(0.5)^2(0.5)^2 = 6 \cdot \tfrac{1}{16} = 0.375 \]

With a fair coin every ordering has the same probability whatever the count, so the whole distribution is just the counts of orderings divided by the total number of sequences, 2 to the fourth, which is 16.

Row 4 of Pascal's triangle, 1, 4, 6, 4, 1, gives every count for four flips at once, and with a fair coin each is divided by 16. The middle entry, 6, is the one you need.

67. Check: at least, with the complement

Check

Finally a range. Translate carefully, then take the short side.

Check your understanding

X ~ B(10, 0.2). Which value is P(at least 2)?

  • A. 1 − P(x ≤ 1) ≈ 0.6242 (correct)
  • B. 1 − P(x ≤ 2) ≈ 0.3222
  • C. P(x ≤ 1) ≈ 0.3758
  • D. P(x = 2) ≈ 0.3020

Answer: A

Why: At least 2 is every count from 2 to 10, so the part to remove is 0 and 1. Those two bars are 0.1074 and 0.2684, adding to 0.3758, and one minus that is about 0.6242.

Why B tempts people
That removes the bar at 2 as well, which at least 2 includes. It is the probability of more than 2, off by the whole bar at 2.
Why C tempts people
That is the side you remove, not the answer. At least 2 is its complement.
Why D tempts people
That is just the single bar at 2. At least 2 also includes 3, 4 and every count up to 10.

\[ P(x \ge 2) = 1 - \left[(0.8)^{10} + 10(0.2)(0.8)^9\right] = 1 - (0.1074 + 0.2684) \approx 0.6242 \]

Listing the counts protects you here. You want 2 through 10, which is nine bars; the other side is 0 and 1, which is two. Add the two, subtract from one.

If you chose the answer that removed the bar at 2, revisit the off-by-one trap: the counts you want and the counts you remove must not overlap.

68. Explain the counting factor

Explain it to yourself

The binomial formula has three factors, and the first is the one students most often drop. Explaining why it is there is the best test of whether the formula makes sense to you.

Discussion prompt

In two or three sentences, explain why the formula for P(x) needs the factor n choose x, using Joe's three guesses and exactly two right answers as your example.

A strong answer names both halves: one factor for the probability of a sequence and one for the number of sequences. It also says why they multiply: the sequences are mutually exclusive and equally likely, so adding them is the same as multiplying one by the count.

An extension worth trying: explain why the factor is 1 when x is 0 or n. There is only one sequence with no successes, and only one with all of them.

69. Exit ticket: Try It 4.13, volunteers

Exit ticket

About 32 percent of students take part in a community volunteer program outside school. Thirty students are chosen at random.

Discussion prompt

Set up the distribution, find the mean and standard deviation, and find the probability that at most 14 of them volunteer. Explain whether your answer is plausible using the mean and standard deviation, and state it in a sentence in context.

Set up first, in words: X is the number of the 30 who volunteer, and a success is a volunteer. At most 14 includes 14, so the complement removes 15 and up.

The plausibility sentence is what earns full marks. It names the evidence, how far 14 sits from the mean in standard deviations, and connects it to the size of the answer. This one mirrors Example 4.13 almost exactly, and the answer is close to its 0.9738 for the same reason.

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, p. 240 — Try It 4.13

70. Recap

Recap

A binomial experiment is a fixed number of independent trials, each a success or a failure with the same probability. The count of successes has a distribution fixed entirely by n and p.

ideaformularead it as
the distributionX ~ B(n, p), q = 1 − pn trials, success probability p on each
one barP(x) = (n choose x) pˣ qⁿ⁻ˣnumber of orderings × probability of one
the meanμ = npn trials, each averaging p successes
the spreadσ = √(npq)largest when p = 0.5; grows like √n
a rangeP(x ≤ k) = P(0) + … + P(k)add the shorter side, use the complement

\[ P(x \ge k) = 1 - P(x \le k - 1) \]

Test all three conditions before any formula: sampling without replacement from a small group, as in the committee draw, breaks independence. Translate every phrase into a list of counts. And judge every answer by where its boundary sits relative to np, measured in standard deviations. The business statistics text by the same authors treats the same distribution with quality-control examples, if you want more practice. Section 4.4 turns the question round: instead of counting successes in fixed trials, count the trials until the first success.

OpenStax Introductory Business Statistics 2e, §4.2 Binomial Distribution §4.2 — the binomial distribution in a business setting

OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution §4.3, pp. 237-242 — the whole section

Sources

  1. OpenStax Introductory Statistics 2e, §4.3 Binomial Distribution — Illowsky & Dean, OpenStax / Rice University, CC BY 4.0, pp. 236-242
  2. OpenStax Introductory Business Statistics 2e, §4.2 Binomial Distribution — Illowsky & Dean, OpenStax / Rice University, CC BY 4.0

Want this taught 1-on-1? Alexander tutors Statistics — $55/session, free consultation.

Book on Wyzant · Text (657) 465-8108