The binomial and normal distributions
If n independent trials are carried out under the same conditions and in each of them event A occurs with probability p (q = 1 − p), this is called a Bernoulli scheme. The probability that A occurs exactly m times is P(n, m) = C(n, m) pᵐ qⁿ⁻ᵐ: there are C(n, m) suitable sequences, each of probability pᵐqⁿ⁻ᵐ. The table of P(n, m) for m = 0, 1, …, n is called the binomial distribution; its entries add up to (p + q)ⁿ = 1. A quantity that takes this or that value as a result of an experiment is a random variable (for example the points on a die, a height, the number of defective items). Quantities shaped by many small random factors (height, measurement error) are normally distributed: the density function is f(x) = (1/(σ√(2π))) · e^(−(x − a)²/(2σ²)); its graph is a bell-shaped curve symmetric about x = a, where a is the centre and σ the standard deviation. About 68 % of values lie within a ± σ and 95 % within a ± 2σ. For large n the binomial distribution approaches the normal distribution.
If a pupil marks every answer at random in a 5-question test with 4 options each, the probability of getting exactly 3 right is C(5, 3) · (1/4)³ · (3/4)² = 10 · 1/64 · 9/16 ≈ 0.088, less than 9 %. The Bernoulli scheme works for repeated independent trials like this.
Quality inspectors calculate the chance of exactly some number of faulty items in a batch, medical researchers estimate the chance that a treatment works for a given number of patients, and test writers choose how many questions are needed so that guessing does not pay off.
Imagine a 3-question test with 3 options each. List every way to get exactly 1 right and count them out of the 27 possible answer patterns. Compare the share with C(3, 1) · (1/3) · (2/3)².
“Rain of coins”: toss 6 coins together 40 times and record how many heads appear each time (0 to 6). Draw a histogram: does it look like a bell? What is the most frequent value? Compare it with the most probable theoretical value in the binomial distribution (3).