Interactive demo · Probability and Statistical Inference · Chapter 1

Long-run frequencies and the continuity of probability

Toss a coin with P(heads) = p and follow the proportion of heads p̂n after n tosses. The frequency interpretation says p̂n → p; the axioms say precisely in what sense. For a fixed n, the probability that p̂n is more than ε away from p is a binomial sum, and it falls to 0 (the weak law). Stronger: let Bn be the event that some later proportion p̂k, k ≥ n, is more than ε away. The events Bn shrink as n grows, so by the continuity of probability P(Bn) tends to the probability of their intersection, the event that the proportion leaves the band infinitely often, and that probability is 0.

Probability of heads p

Band half-width ε

Coins simulated K

Tosses per coin N

Try these

Each coin is an independent sequence of N tosses. The bottom-right panel can only look up to toss N, so it shows the fraction of coins outside the band at some toss between n and N; as N grows this rises to P(Bn) (continuity from below), and it can never be smaller than the fraction outside at toss n itself.

In the lecture notes