Probability Distributions: The Shapes of Randomness

Three things happen every day across India that all involve chance, but they don't behave the same way at all. Toss a coin five times and count the heads — the possible answers are 0, 1, 2, 3, 4, 5, and nothing in between; you can list every outcome on your fingers. Now check exactly when your Mumbai local actually departs, relative to its scheduled time — it could be 40 seconds early, 2.3 minutes late, 17.8 seconds late, anything at all in a continuous window; there's no "next" value after 2.3 minutes, because between any two times there's always another time in between. Now look at the marks of the roughly 20 lakh students who write the CBSE Class 10 board exam each year — plot how many students scored each mark, and you get a hump: few students at the very bottom, few at the very top, most bunched in the middle, roughly symmetric.

Three completely different shapes, all produced by randomness. This chapter is about the machinery that produces them — how to go from "here's a chance experiment" to "here's the exact probability of every possible result," not just wave your hands and say "it varies." That machinery has a name: a probability distribution. By the end of this chapter you'll be able to build one from first principles, compute probabilities from it, and, importantly, avoid the two mistakes that trip up almost everyone the first time they meet this topic.

From Outcomes to Numbers: What a Random Variable Actually Is

Every probability problem you've solved so far — dice, cards, coins — asked about events described in words: "the event that the sum is 7," "the event that both cards are red." A random variable is the shift from describing events in words to describing them with numbers. Formally, a random variable X is a rule that assigns a number to every outcome of a random experiment. Roll two dice and let X = the sum. The experiment has 36 equally likely outcomes like (3, 4), but X collapses each outcome to a single number: (3, 4) and (4, 3) both give X = 7. Once you have that number, you can ask numerical questions — What's the average value of X? How spread out is it? — that "the event of getting a red card" never let you ask.

Random variables come in two flavours, and the difference is exactly the coin-versus-train-timing contrast above:

• Discrete random variable: takes only countable, separated values — 0, 1, 2, 3, ... You can always ask "what comes right after this value?" and get a definite answer. Number of heads in 5 tosses, number of wrong answers in a 20-question quiz, number of sixes in an over.
• Continuous random variable: takes any value in an interval of real numbers — there is no "next" value, because between any two values there are infinitely many more. Exact arrival delay of a train, exact weight of a sack of rice, exact time between two UPI transaction pings.

The reason this split matters is that discrete and continuous random variables need genuinely different tools to describe their probabilities — you're about to see both.

The Probability Distribution of a Discrete Random Variable

A probability distribution for a discrete random variable X is simply the complete list of every value X can take, paired with the probability of each. Formally it's a function p(x) = P(X = x), called the probability mass function (pmf), and it must obey two rules that follow directly from the axioms of probability you already know:

1. 0 ≤ p(x) ≤ 1 for every value x (probabilities can't be negative or exceed 1)
2. The sum of p(x) over every possible x must equal exactly 1 (something has to happen)

Let's build one from scratch. Toss a fair coin 3 times and let X = number of heads. There are 2³ = 8 equally likely outcomes: HHH, HHT, HTH, THH, HTT, THT, TTH, TTT. Group them by X:

Check rule 2: 1/8 + 3/8 + 3/8 + 1/8 = 8/8 = 1. Good — that's not a coincidence, and you'll see exactly why it's guaranteed shortly.

Once you have this table, you can compute the expected value (mean) of X — the long-run average you'd get if you repeated this experiment thousands of times:

E[X] = Σ x·p(x) = 0(1/8) + 1(3/8) + 2(3/8) + 3(1/8) = (0 + 3 + 6 + 3)/8 = 12/8 = 1.5

That 1.5 makes sense: with 3 fair tosses, you "expect" half of them — 1.5 — to land heads, even though 1.5 heads can never actually happen on any single trial. The expected value describes the long-run average, not a possible single outcome — that's a distinction worth sitting with.

The Binomial Distribution: Counting Successes in Repeated Trials

The coin-toss table above is a special case of an enormously important pattern: repeat the same yes/no trial n independent times, each with the same success probability p, and count the number of successes. This is the binomial distribution, and it deserves a general formula rather than listing outcomes by hand every time.

Here's a sharper example than coins. A 5-question MCQ test has 4 options per question, and a student guesses every answer completely at random, so the chance of getting any single question right is p = 0.25 (and wrong is q = 1 − p = 0.75). Let X = number of correct guesses out of 5. What's P(X = k) for each k?

Build the formula in two steps, the way you'd build any counting argument:

Step 1 — probability of one specific arrangement. Suppose the student gets exactly the 1st, 3rd, and 4th questions right and the rest wrong: that's a specific sequence like (right, wrong, right, right, wrong). Because the questions are independent, you multiply: p·q·p·p·q = p³q². Any other specific sequence with exactly 3 rights and 2 wrongs also has probability p³q² — the order of multiplication doesn't matter, only how many p's and q's appear.

Step 2 — how many such sequences are there? This is exactly the combinatorics you already know: choosing which 3 of the 5 questions are the "right" ones is C(5, 3) = 10 ways. Each of those 10 arrangements has the same probability p³q², so:

P(X = 3) = C(5, 3)·p³·q² = 10 × 0.25³ × 0.75² = 10 × 0.015625 × 0.5625 = 0.0879

Generalising this argument for any k successes out of n trials gives the binomial probability formula:

P(X = k) = C(n, k)·pᵏ·q^n−k, for k = 0, 1, 2, ..., n

Computing all six values for our 5-question example:

from math import comb
n, p = 5, 0.25
for k in range(n + 1):
    prob = comb(n, k) * p**k * (1 - p)**(n - k)
    print(k, round(prob, 4))

Tracing this line by line: for each k from 0 to 5, comb(5, k) computes C(5, k), and the formula multiplies in p to the power k and (1−p) to the power (5−k). Running it produces:

0 0.2373    1 0.3955    2 0.2637    3 0.0879    4 0.0146    5 0.001

These six numbers are exactly what the bar chart below plots.

Misconception 1: "The Probabilities Only Need to Roughly Add to 1"

Look again at those six numbers: 0.2373 + 0.3955 + 0.2637 + 0.0879 + 0.0146 + 0.001. Add them and you get exactly 1.000000 (to floating-point precision) — not approximately, not "close enough," but algebraically exactly 1, every single time, for every valid n and p. This isn't a lucky coincidence you should just trust; it's provable using something you already know: the Binomial Theorem.

Recall that (a + b)ⁿ = Σ C(n, k)·aᵏ·b^n−k, summed over k = 0 to n. Now set a = p and b = q, remembering that q = 1 − p was defined exactly so that p + q = 1:

Σ P(X = k) = Σ C(n, k)·pᵏ·q^n−k = (p + q)ⁿ = 1ⁿ = 1

The sum-to-1 property of the binomial distribution and the Binomial Theorem's expansion of (p+q)ⁿ are literally the same algebraic statement viewed two ways. If you ever compute a distribution table and the probabilities don't sum to exactly 1, you have a genuine arithmetic error — not an acceptable rounding artifact to shrug off.

How the Shape Changes: Mean, Variance, and Skew

A binomial distribution has two parameters, n and p, and both control its shape. Its mean and variance follow from a short, clean argument. Write X as a sum of n independent 0/1 indicator variables, X = X₁ + X₂ + ... + Xₙ, where Xᵢ = 1 if trial i succeeds and 0 otherwise (each Xᵢ is called a Bernoulli variable). Since Xᵢ only takes values 0 and 1:

E[Xᵢ] = 1·p + 0·q = p

By linearity of expectation (the expectation of a sum is the sum of expectations, regardless of independence), E[X] = E[X₁] + ... + E[Xₙ] = np.

For variance, notice that because Xᵢ ∈ {0, 1}, Xᵢ² = Xᵢ always (0² = 0 and 1² = 1), so E[Xᵢ²] = E[Xᵢ] = p. Then Var(Xᵢ) = E[Xᵢ²] − (E[Xᵢ])² = p − p² = p(1 − p) = pq. Because the trials are independent, variances add: Var(X) = npq.

For the 5-MCQ example: E[X] = 5(0.25) = 1.25 correct guesses on average, Var(X) = 5(0.25)(0.75) = 0.9375.

Now watch what p does to the shape. Picture an 8-match T20 league stretch where a team's chance of winning any single match is a constant p (a simplification — real form varies match to match, but it isolates what p alone does to the distribution of X = number of wins). Compare p = 0.2 (a struggling side), p = 0.5 (evenly matched), and p = 0.8 (a dominant side):

All three panels use the identical vertical scale, so the heights are directly comparable. p = 0.2 and p = 0.8 are mirror images of each other (skewed toward 0 and toward 8 respectively) — that's not a visual coincidence either: P(X = k; p = 0.8) always equals P(X = 8−k; p = 0.2), because "8 successes at 0.8" is the same combinatorial statement as "8 failures at 0.2" with the roles of success/failure swapped. Only p = 0.5 is symmetric on its own, because success and failure are equally likely.

A Bridge to Continuous Random Variables

Everything so far worked because you could list outcomes: 0, 1, 2, ..., n. A continuous random variable — like the exact delay of a train — has no list to make. Between a 2-minute delay and a 2.01-minute delay there are infinitely many possible delays, and in fact P(X = exactly 2.000000... minutes) is mathematically 0 for any single exact value, because there's no way to spread positive probability across infinitely many points and still sum to 1. So "the probability of each value" — the pmf idea — breaks down completely for continuous variables. What we use instead is a probability density function (PDF), and the questions we ask shift from "probability of exactly this value" to "probability of landing in this range."

One honest note before going further: what follows briefly uses the language of integration — the ∫ symbol — which is formally taught with limits and antiderivatives in Class 11–12. You don't need that machinery yet to follow the ideas here, because every example in this chapter has a density that's a flat, constant height over an interval — which means every probability calculation is literally just the area of a rectangle: base (the length of the interval) times height (the density value). Read "∫f(x)dx between a and b" simply as "the area under the density curve between a and b" — geometry you already know, not new machinery you're missing.

For a continuous random variable X with density f(x), probability over an interval is that area:

P(a ≤ X ≤ b) = ∫_a^b f(x) dx = (area under f between a and b)

and the density must satisfy its own version of the two pmf rules: f(x) ≥ 0 everywhere, and the total area under f across all possible x must equal exactly 1.

Misconception 2: "A Density Value Can't Be More Than 1, Because Probabilities Can't Exceed 1"

This is the single most common continuous-distributions mistake, and it's worth killing with two concrete, contrasting numbers rather than just asserting it's false.

The simplest continuous distribution is the uniform distribution: X is equally likely to land anywhere in an interval [c, d], so its density is flat — f(x) = 1/(d − c) for c ≤ x ≤ d, and 0 outside. Consider a school science-fair reflex-test buzzer that sounds at a uniformly random instant somewhere in a 0.5-second window after the light turns green: X ~ Uniform(0, 0.5). Here d − c = 0.5, so:

f(x) = 1/0.5 = 2, for 0 ≤ x ≤ 0.5

That's a density of 2 — genuinely greater than 1 — and there is nothing wrong with it. Check the total area: base × height = 0.5 × 2 = 1. Total probability is still exactly 1, as required; it's just packed into a narrow interval, so the height has to be tall to make the area work out. Now compute an actual probability with it: what's the chance the buzzer sounds within the first 0.1 seconds of the window?

P(X ≤ 0.1) = (0.1 − 0) × 2 = 0.2

Now contrast this with your Mumbai local: suppose your train's arrival is uniformly random somewhere in a 10-minute window, X ~ Uniform(0, 10). Here d − c = 10, so f(x) = 1/10 = 0.1 for 0 ≤ x ≤ 10 — a density well under 1. Total area check: 10 × 0.1 = 1, correct. Probability of waiting at most 2 minutes:

P(X ≤ 2) = (2 − 0) × 0.1 = 0.2

Notice both examples gave probability 0.2, from completely different-looking densities (2 versus 0.1). That's exactly the point: f(x) is a height, not a probability. A probability is an area (height × width), and only areas are capped at 1 — heights can be anything non-negative, and narrower intervals force taller heights to keep the area equal to 1.

The left rectangle clears the "density = 1" reference line by a wide margin; the right one sits far beneath it. Both have area exactly 1. That's the whole misconception, corrected with numbers instead of just a rule to memorise.

The Cumulative Distribution Function: A Running Total

Alongside the density, every random variable — discrete or continuous — has a cumulative distribution function (CDF), F(x) = P(X ≤ x): the probability of landing at or below x. For the reflex-buzzer example, F(x) is the area under f from 0 up to x, and since f is flat at height 2, that area is just a growing rectangle: F(x) = 2x for 0 ≤ x ≤ 0.5 (and F(x) = 0 before 0, F(x) = 1 after 0.5, since probability can't exceed 1). Check: F(0.5) = 2(0.5) = 1 — by x = 0.5 you've accumulated all the probability there is, which matches the window ending there. F(0.1) = 2(0.1) = 0.2, exactly the P(X ≤ 0.1) computed earlier — the CDF is just a running total of the density.

For the Mumbai-local example, F(x) = x/10 for 0 ≤ x ≤ 10, and F(2) = 0.2, again matching. Notice something: F(x) here is a straight ramp, and its steepness (rise over run) is exactly the density height — the ramp for the buzzer example rises with slope 2, the ramp for the train example rises with slope 0.1. "Steeper ramp = higher density packed into that stretch = more probability accumulating faster" is the entire intuition behind the relationship F′(x) = f(x), which you'll meet formally as a consequence of the Fundamental Theorem of Calculus once integration is introduced in Class 11–12. You don't need calculus to see it here — for a straight ramp, slope is just rise divided by run, arithmetic you've known since Class 8.

A Third Shape: The Normal Distribution, Qualitatively

Go back to the CBSE board-marks example from the opening: plot how many of the lakhs of students who sit an exam score each mark, and — for a well-designed exam attempted by a large, varied population — you tend to get roughly the symmetric hump this chapter opened with, called the normal (or Gaussian) distribution. Its defining features, without needing its formula (which does require calculus and is genuinely Class 12+ material, so it's stated here only for completeness, not for computation): it's perfectly symmetric about its mean, and mean = median = mode all coincide at the centre of the hump. A useful rule of thumb called the empirical rule says that for a normal distribution, about 68% of values fall within 1 standard deviation of the mean, about 95% within 2 standard deviations, and about 99.7% within 3 — a fact you'll prove using the normal density's integral in Class 12, but can use qualitatively right now to reason about spread: if the average mark on a test is 70 with a standard deviation of 8, and marks are roughly normal, then roughly 95% of students scored somewhere between 54 and 86. The point of including this third shape here isn't to compute with it yet — it's so you recognise, when you meet its formula later, that you already understand exactly what shape it's describing and why that shape shows up so often (many small independent random effects adding up tend to produce this hump, a deep result called the Central Limit Theorem that's well beyond this chapter).

Where This Shows Up in Exams

Classical, single-event probability — the kind CBSE Class 10 boards test directly — gets noticeably easier once you're fluent in random-variable thinking, because you start seeing "count the favourable outcomes and divide" as a special case of building a distribution table rather than a one-off trick per question. The binomial formula P(X = k) = C(n, k)pᵏq^n−k, its mean np, and its variance npq are a recurring, direct-application topic in JEE Main, JEE Advanced, and BITSAT — questions there often just require you to identify n and p correctly and plug in, which is exactly the muscle the 5-MCQ and 8-match examples above built. Combinatorics-heavy probability problems — including the binomial-theorem sum-to-1 identity from Misconception 1 — are also a classic technique in olympiad-style contests such as RMO/INMO. None of that requires you to have already learned integration: everything computable in this chapter uses only combinatorics and rectangle areas. What you've built here is the conceptual foundation — random variables, distributions, density-as-height-not-probability — that Class 11–12 Applied Mathematics later formalises with calculus; you are not seeing a preview you'll have to re-learn from scratch, you're seeing the ideas the formal tools will eventually be attached to.

Test Yourself

1. A discrete random variable X has P(X=1) = 0.2, P(X=2) = 0.3, P(X=3) = k, P(X=4) = 0.1, and these are the only values X can take. Find k.
2. Using the 5-MCQ guessing model (n = 5, p = 0.25), find P(X ≥ 4) — the probability of getting at least 4 of the 5 guesses right.
3. For the n = 8, p = 0.5 cricket example, compute E[X] and Var(X), and give the standard deviation to 2 decimal places.
4. For the reflex-buzzer distribution X ~ Uniform(0, 0.5), find P(0.2 ≤ X ≤ 0.4).
5. True or false, with justification: "A probability density function can never exceed 1."

Answers. (1) The four probabilities must sum to 1: 0.2 + 0.3 + k + 0.1 = 1, so k = 0.4. (2) P(X=4) + P(X=5) = C(5,4)(0.25)⁴(0.75) + (0.25)⁵ = (1/256)(15/4) + (1/256)(1/4) — more directly, factor (0.25)⁴ out: (0.25)⁴[5(0.75) + 0.25] = (1/256)(4) = 4/256 = 1/64 = 0.015625. (3) E[X] = np = 8(0.5) = 4; Var(X) = npq = 8(0.5)(0.5) = 2; SD = √2 ≈ 1.41. (4) P(0.2 ≤ X ≤ 0.4) = (0.4 − 0.2) × 2 = 0.4. (5) False — the reflex-buzzer example above has f(x) = 2 for 0 ≤ x ≤ 0.5, a density greater than 1, with total area (probability) still exactly 1; density is a height, and only areas — not heights — are bounded by 1.

Summary

• A random variable maps outcomes of a chance experiment to numbers, and comes in two kinds: discrete (countable, separated values) and continuous (any value in an interval).
• A discrete distribution is a pmf p(x) = P(X = x) satisfying 0 ≤ p(x) ≤ 1 and Σp(x) = 1 exactly — provably exactly, not approximately, as the binomial case shows via the Binomial Theorem.
• The binomial distribution P(X=k) = C(n,k)pᵏq^n−k models counting successes across n independent identical trials; it has mean np and variance npq, both derivable from writing X as a sum of independent 0/1 Bernoulli variables.
• A continuous distribution is described by a density f(x) ≥ 0 whose total area is 1; probabilities are areas (P(a≤X≤b) = area under f from a to b), computable as base×height for uniform (flat) densities without needing formal integration.
• Density is a height, not a probability — it can exceed 1 when the interval it's spread over is narrow, as long as the total area still equals 1.
• The CDF F(x) = P(X≤x) is a running total of the density; for flat densities it's a straight ramp whose slope equals the density height, previewing the calculus relationship F′(x) = f(x).
• The normal distribution is a third shape — symmetric, bell-shaped, governed qualitatively by the 68-95-99.7 empirical rule — that shows up whenever many small independent effects add together, such as exam marks across a large student population.

Think About It

Think about this: How would you explain probability distributions: the shapes of randomness to a friend who has never seen a computer? What real-world analogy would you use? Imagine you had to build a system using these concepts — what would be your first step? Try this: before moving on, write down three things you learned and one question you still have.

Practice Exercises

Now it is time to practice! Complete these challenges to solidify your understanding:

• Exercise 1: Write a short program that demonstrates the core concept from this chapter. Test it with at least 3 different inputs.
• Exercise 2: Find a real-world example where probability distributions: the shapes of randomness is used in an Indian company (like TCS, Infosys, Flipkart, or ISRO). Write a paragraph explaining the connection.
• Exercise 3: Create a mind-map connecting probability distributions: the shapes of randomness to at least 3 other topics you have studied.