Law of Large NumbersStatLens

Toss a coin over and over and the proportion of heads settles down near 0.5. Roll a die over and over and the average roll settles near 3.5. Add trials a few at a time and watch the line stop thrashing — not because anything corrects it, but because each new trial counts for less.

Experiment:

The last few tosses

Law of Large Numbers — Help

The long-run proportion (or average) of independent trials settles toward the value the probability model predicts. This page runs the trials and draws that settling.

Why it settles. Nothing corrects a run of tails. The line calms down because the denominator grows: one more toss changes a proportion out of 20 by up to 0.05, and a proportion out of 2000 by at most 0.0005. The early trials are never cancelled — they are outvoted.

“Spread out the early trials” puts the trial count on a log scale, so the first ten trials get as much width as the last nine hundred. The wild early stretch is where the law is doing its visible work, and on an ordinary axis it is squeezed into a sliver.

“Also show the running total” draws heads minus tails (or the total roll minus 3.5 per roll). This is the half that gets left out: the proportion converges while the total wanders further and further from zero. A coin is never “due”.

Starting over. The circular arrow gives a different run each time, so you can ask “does that happen every time?” and find out. A link with ?seed= always opens on the same run — reload the page to get it back after resetting.

Instructors: there is a printable guide with talking points, and ?mode=present hides the student prompt.

Keyboard