← Back to the Introduction to Statistics outline The Course Maker
← Introduction to Statistics outline
Week 7 · Chapter

Chapter 7 — The Binomial Distribution

Introduction to Statistics Generic evergreen edition

Course: Introduction to Statistics (18-week generic edition)
The module's primary reading · ~30–40 minutes · Objective 4 · Read this before (or alongside) the week's lecture — everything else in the module builds on it.

Why you already care about this

Three questions that sound like they come from three different lives. A shop emails 5 customers, and each one clicks the link with a 1-in-5 chance — how likely is it that exactly two click? A basketball player who makes half her free throws takes 4 shots — how often does she make all 4? You didn't study, so you guess on every true/false question — how many will you get right?

Here is the secret this chapter is built on: those are all the same problem. Each one has a fixed number of tries. Each try ends in a plain yes or no. And all we care about is the count of yeses. That pattern — repeat a yes/no trial a set number of times and count the successes — shows up so relentlessly, from marketing campaigns to seed packets to test-taking, that statisticians long ago solved it once and for all. The solution is called the binomial distribution, and it is the first famous random variable you'll meet: it comes with its own recognition checklist, its own probability formula, its own shortcut for the mean and standard deviation, and its own spreadsheet function.

By the end of this chapter you will be able to: recognize a binomial setting in the wild (and refuse the impostors that only look like one); compute the exact probability of any number of successes; use a one-subtraction shortcut for "at least one"; state what count to expect, give or take how much; compute all of it with technology; and describe the shape of a binomial distribution — including the strange, wonderful fact that when the number of trials grows, a bell-shaped silhouette starts to emerge. That bell is next week's whole story.

Where we are

You've spent two weeks preparing for this one, whether you noticed or not. Week 5 gave you the probability rules — especially multiplying probabilities across independent events and the complement rule, P(not A) = 1 − P(A). Week 6 taught you what a random variable is: a numeric outcome with a probability attached to each possible value, summarized by its expected value E(X) and its standard deviation. Back then we built every distribution by hand, one row of the table at a time. This week, one hugely common random variable gets the star treatment: its entire distribution — every row of that table — comes from just two numbers, n and p.

1. The binomial setting: does it fit in the B·I·N·S?

Before any formula, a gatekeeping question: does the binomial model even apply? It applies only when the situation passes a four-part checklist.

  • B — Binary. Each trial has exactly two outcomes. Call one "success" and the other "failure." A crucial warning: success is just the thing you're counting — it doesn't mean something good. If you're counting missed appointments, a missed appointment is a "success."
  • I — Independent. The trials don't affect each other. One customer clicking an email tells you nothing about whether the next one will.
  • N — Number fixed. The number of trials, written n, is set before the trials begin. "Keep trying until the first win" is not a fixed n.
  • S — Same p. Every trial has the same probability of success, written p.

If all four hold, then X = the number of successes in the n trials is a binomial random variable. The memory hook: B·I·N·S — if it fits in the BINS, it's binomial.

Worked example. An online shop sends a promotional email to 5 customers. Past campaigns show each recipient clicks the link with probability 0.2, independently of the others. Let X = the number who click. Run the checklist: Binary? Click or no click — yes. Independent? One customer's click doesn't change another's — yes, as stated. Number fixed? n = 5, decided before sending — yes. Same p? p = 0.2 for every recipient — yes. So X is binomial with n = 5 and p = 0.2, and its possible values are 0, 1, 2, 3, 4, 5. This example will follow us through the whole chapter.

Just as important is recognizing what fails the checklist. Three impostors to refuse on sight:

  • "Count the emails you must send until the first click." There is no fixed n — the number of trials is itself the random thing. (That setting has its own distribution with its own name, but it's outside this course. You only need to recognize that it isn't binomial.)
  • "Draw 4 names from a box of 15 staff names, without replacement, and count the part-timers." Each draw changes the mix left in the box, so the success probability shifts from draw to draw and the trials aren't independent — the I and the S both fail.
  • "Record how long each seed takes to sprout." That's a measurement, not a yes/no count. There's no "number of successes" to distribute.

Misconception check. "Binomial problems need p = 0.5, like a fair coin." No — p can be anything between 0 and 1: clicks at 0.2, germination at 0.9, anything. What matters is not the value of p but that p is the same on every trial.

2. The binomial formula: ways × wins × losses

Now the payoff question for our email example: what is P(X = 2), the probability that exactly two of the five customers click? We'll build the answer in three moves, because the formula makes perfect sense once you've built it — and no sense if it's just handed to you.

Move 1 — the probability of one specific story. What's the chance the first two customers click and the last three don't? The trials are independent, so multiply:

0.2 × 0.2 × 0.8 × 0.8 × 0.8 = 0.2² × 0.8³ = 0.04 × 0.512 = 0.02048.

Tempting as it is to stop here, 0.02048 is not the answer. It's the probability of one particular arrangement: click, click, no, no, no.

Move 2 — count the ways. "Exactly two click" doesn't care which two. The clicks could come from customers 1 and 2, or 1 and 3, or 2 and 5… The number of ways to choose which 2 of the 5 trials are the successes is called the binomial coefficient, written C(5, 2) and read "5 choose 2." Here C(5, 2) = 10 — you can list all ten pairs of positions and count them. Every one of those ten arrangements has the same probability, 0.02048.

Move 3 — assemble. Ten equally likely ways, each with probability 0.02048:

P(X = 2) = C(5, 2) × 0.2² × 0.8³ = 10 × 0.04 × 0.512 = 0.2048 — about a 20% chance.

That construction is the general formula. For a binomial random variable with n trials and success probability p:

P(X = k) = C(n, k) × p^k × (1 − p)^(n − k)

Read it in words, left to right: (the number of ways to place the k successes among the n trials) × (the probability the successes happen) × (the probability the failures happen). Or shorter: ways × wins × losses.

The complement trick — "at least one." Questions about "at least one success" look like they need five separate formula runs (k = 1, 2, 3, 4, 5). They don't. The only outcome excluded from "at least one" is "none," so subtract:

P(at least one click) = 1 − P(X = 0) = 1 − 0.8⁵ = 1 − 0.32768 = 0.67232 — about a 67% chance.

Memory hook: "At least one = 1 − P(none)." One subtraction replaces five computations.

Run the formula for every k from 0 to 5 and you get the complete distribution — exactly the kind of table you built by hand in Week 6, except now a formula generates it. The table below lists every possible click count for the email example (n = 5, p = 0.2), with probabilities rounded to 4 decimal places.

k (clicks) 0 1 2 3 4 5
P(X = k) 0.3277 0.4096 0.2048 0.0512 0.0064 0.0003

Three things worth noticing. The probabilities sum to 1 — Week 6's law, still in force. The most likely count is 1 click, not 0. And going 5-for-5 is a 3-in-10,000 long shot.

Misconception check. "P(exactly 2 of 5) = 0.2² × 0.8³ — done." This is the single most common binomial error: forgetting the ways factor. Without the ×C(5, 2), you've computed the chance of one particular arrangement, and your answer is too small by a factor of ten. If a computed probability feels suspiciously tiny, ask: did I count the ways?

Misconception check. "'At least 2' and 'exactly 2' are basically the same question." Read probability questions like a lawyer. Exactly 2 is one bar of the histogram. At least 2 is a whole tail of bars (2, 3, 4, and 5). And "at least one" has the express lane: 1 − P(0).

3. Mean and standard deviation: np, give or take

Often you don't need the probability of an exact count — you just want to know what to expect, give or take how much. For a binomial random variable, both answers are one-line formulas.

Start with the intuition. If each of 100 tries succeeds 25% of the time, how many successes do you expect? A quarter of 100 — 25. That intuition is exactly right, and it is the theorem:

  • Mean (expected value): μ = np.
  • Standard deviation: σ = √( np(1 − p) ).

The SD formula is less guessable, and we won't derive it — Week 6 already taught you what the standard deviation of a random variable means (the typical distance of outcomes from the mean); this is simply the binomial shortcut for it.

Worked example. A packet of 100 prairie wildflower seeds; each germinates with probability 0.25, independently (a tough native species).

  • Mean: μ = np = 100 × 0.25 = 25 seeds.
  • Variance: np(1 − p) = 100 × 0.25 × 0.75 = 18.75.
  • SD: σ = √18.75 ≈ 4.33 seeds.

Now say it in words, because the words are the point: "Expect about 25 sprouts, give or take about 4." The SD turns the mean into a usable range. A tray with 22 or 29 sprouts? Unremarkable — within about one give-or-take of expectation. A tray with only 5 sprouts? That's 20 below the mean — more than four SDs — and four-SD accidents essentially don't happen by luck. Something is wrong with the seeds, the soil, or the watering. This is the week's quiet superpower: the mean-and-SD pair lets you tell an ordinary run from a real signal.

A second example, closer to home. Guess on all 16 questions of a true/false quiz: n = 16, p = 0.5. Mean: 16 × 0.5 = 8. SD: √(16 × 0.5 × 0.5) = √4 = 2. Pure guessing delivers about 8 right, give or take 2 — so a guesser typically lands between 6 and 10. Luck is not a study strategy, and now you can prove it with two multiplications and a square root.

Closing the loop with Week 6. Week 6's method for an expected value was E(X) = Σ k · P(k) — multiply each value by its probability and add. Does the shortcut really agree? Take a small binomial with n = 4 and p = 0.5, whose probabilities (for k = 0, 1, 2, 3, 4) are 0.0625, 0.25, 0.375, 0.25, 0.0625:

E(X) = 0(0.0625) + 1(0.25) + 2(0.375) + 3(0.25) + 4(0.0625) = 0 + 0.25 + 0.75 + 0.75 + 0.25 = 2,

and np = 4 × 0.5 = 2. Same answer — one multiplication instead of five products and a sum. That's what formulas are: the long way, pre-done.

Misconception check. "μ = 25 means 25 seeds will germinate." The mean is a long-run average, not a promise — expected, not guaranteed. Individual runs wander, and the SD measures exactly how far a normal wander goes.

Misconception check. "She's missed three free throws in a row — she's due to make the next one." The I in B·I·N·S says otherwise: independent trials don't remember. If p = 0.5, it is 0.5 on the fourth shot whether the shooter is 3-for-3 or 0-for-3. (Whether real shooters are truly independent trials is a genuinely interesting question — it's this week's discussion.)

4. Computing with technology

Hand computation is how you own the machinery, and it's realistic for small n. Nobody hand-computes C(100, 37) — for anything bigger, the spreadsheet takes over.

In practice (your tools). The workhorse is =BINOM.DIST(k, n, p, cumulative) — same function in Google Sheets and Excel. The last argument is the one that bites people:

  • =BINOM.DIST(2, 5, 0.2, FALSE) → 0.2048. FALSE means "exactly k" — one bar of the histogram. This is our worked example, confirmed by machine.
  • =BINOM.DIST(1, 5, 0.2, TRUE) → 0.73728. TRUE means cumulative — P(X ≤ 1), "at most 1," the first two bars added together (0.32768 + 0.4096).
  • "At least one" by machine: =1 - BINOM.DIST(0, 5, 0.2, TRUE) → 0.67232, matching the complement trick by hand.

To build a whole distribution in 30 seconds: put k = 0 through 5 in column A, fill =BINOM.DIST(A2, 5, 0.2, FALSE) down column B, and insert a column chart — the table from Section 2, drawn. A Desmos-class stats tool will plot the same distribution and let you drag n upward, which is the best possible preparation for the next section.

A habit worth keeping all term: whenever you compute a binomial probability, you now have three independent weapons — the formula, the course's mini-table (in the tutorial), and =BINOM.DIST. They must agree. When a chatbot hands you a binomial answer, check it the same way: AIs routinely drop the ways factor or quietly return the cumulative probability when you asked for exactly. The tool drafts; you judge.

5. Shape: from lopsided to bell

Every binomial distribution has a histogram — one bar for each possible k from 0 to n. Three facts about its shape, all visible the moment you plot:

  • When p = 0.5, the histogram is symmetric. Successes and failures trade places evenly. The n = 4, p = 0.5 distribution (0.0625, 0.25, 0.375, 0.25, 0.0625) is a perfect little pyramid.
  • When p is far from 0.5 and n is small, it's skewed. Our email histogram (n = 5, p = 0.2) piles up at 0 and 1 clicks and trails off to the right. Mirror-image p values (0.2 and 0.8) give mirror-image histograms.
  • As n grows, a bell emerges. This is the astonishing one. Even at p = 0.2, by n = 100 the bars form a smooth, symmetric-looking mound centered at np = 20. Try it with the spreadsheet column trick or a Desmos-class plot: the lopsidedness melts away as n climbs.

Sit with that last fact for a second. A machine built entirely from yes/no trials — nothing smooth about it — keeps producing the same soft bell silhouette whenever n gets large. Statisticians noticed the same silhouette appearing from dozens of other unrelated machines, too. Next week that shape finally gets its name — the normal distribution — and one smooth curve will stand in for a thousand histograms. This week's job is only to see the bell coming.

Key terms

  • Binomial setting — a situation passing the B·I·N·S checklist: Binary outcomes, Independent trials, Number of trials fixed in advance, Same p on every trial.
  • Trial — one repetition of the yes/no process (one email sent, one shot taken).
  • Success / failure — the two outcomes of a trial; "success" is simply the outcome being counted, good or not.
  • n — the fixed number of trials.
  • p — the probability of success on any single trial (the same for all trials).
  • Binomial random variable — X = the number of successes in the n trials; possible values 0, 1, …, n.
  • Binomial coefficient C(n, k) — the number of ways to choose which k of the n trials are the successes ("n choose k"); the formula's "ways" factor.
  • Binomial formula — P(X = k) = C(n, k) × p^k × (1 − p)^(n − k): ways × wins × losses.
  • Complement rule for "at least one" — P(at least one success) = 1 − (1 − p)^n.
  • Mean of a binomial — μ = np; the long-run average count of successes.
  • Standard deviation of a binomial — σ = √( np(1 − p) ); the give-or-take around np.
  • Cumulative probability — P(X ≤ k), the histogram's bars added up through k; what =BINOM.DIST(…, TRUE) returns.

Summary

  • The binomial model applies only to settings that pass B·I·N·S: Binary outcome, Independent trials, Number fixed, Same p. The classic impostors: no fixed n ("until the first success"), and without-replacement draws (p drifts, trials linked).
  • P(X = k) = C(n, k) p^k (1 − p)^(n−k) — ways × wins × losses. The ways factor C(n, k) is the piece everyone forgets; skipping it made our email answer ten times too small.
  • Read questions like a lawyer: exactly k is one bar; at least / at most are tails. The express lane: at least one = 1 − P(none).
  • μ = np and σ = √(np(1−p)): what to expect, give or take. Expected, not guaranteed — and the SD is what separates an ordinary run from a real signal.
  • =BINOM.DIST(k, n, p, FALSE) gives exactly k; TRUE gives the cumulative P(X ≤ k). Formula for small n, technology for big n — and all your weapons must agree.
  • Shape: symmetric at p = 0.5, skewed when p is extreme and n small — and as n grows, the histogram reaches for a bell. Next week the bell gets its name.

Check your understanding

Work these before looking at the answers — they're ungraded, and the answers (with reasoning) are at the bottom of the page.

  1. A dental clinic knows from its records that 20% of booked patients don't show up, independently of one another. Tomorrow morning it has 5 patients booked. Let X = the number of no-shows. Is X binomial? Run the B·I·N·S checklist explicitly, and if it passes, state n, p, and what counts as a "success."
  2. For the clinic in Question 1, compute P(exactly one no-show) with the binomial formula, showing the ways factor.
  3. A bakery's sourdough process gives each loaf a perfect crust with probability 0.75, independently. Tonight's batch is 48 loaves. Find the mean and standard deviation of the number of perfect-crust loaves, then say both in plain words.
  4. A teacher picks 3 students from a class of 12 by drawing names from a jar, without replacement, and counts how many of the 3 walk to campus. A classmate says, "That's binomial — three trials, walk or don't walk." Explain, in one or two sentences, why it is not binomial, naming the checklist letter(s) that fail.

Answers (worked reasoning — read after attempting)

  1. Yes — X is binomial. Binary: each patient either no-shows or shows (and "success" = a no-show, the thing being counted — success isn't good!). Independent: stated in the problem. Number fixed: n = 5, booked in advance. Same p: p = 0.2 for every patient. So X is binomial with n = 5, p = 0.2, values 0 through 5.
  2. P(X = 1) = C(5, 1) × 0.2¹ × 0.8⁴ = 5 × 0.2 × 0.4096 = 0.4096 — about a 41% chance of exactly one empty chair. (The ways factor is C(5, 1) = 5: the single no-show could be any one of the five patients. Without it you'd get 0.08192, five times too small.)
  3. Mean: μ = np = 48 × 0.75 = 36 loaves. Variance: np(1 − p) = 48 × 0.75 × 0.25 = 9; SD: σ = √9 = 3 loaves. In words: the bakery should expect about 36 perfect crusts, give or take about 3. A batch with 33–39 perfect loaves is business as usual; a batch with 20 means the starter, the oven, or the recipe has a real problem — that's more than five SDs below expectation, not bad luck.
  4. Without replacement, each draw changes who's left in the jar, so the probability that the next drawn student walks to campus shifts depending on who has already been drawn. The trials are not independent and p is not the same on every trial — the I and the S in B·I·N·S both fail. (With a large population, or drawing with replacement, the binomial would be back in business.)

Next up in this module: the slide deck if you want the lecture's visual walk-through, then the Readings & Resources page for optional deeper dives — and when you're ready to practice, Lecture Tutorial 7 turns your chatbot into your personal binomial tutor, mini-table included.