Experiments and Simulations
Expected versus observed
Probability says what to expect: a fair coin should land heads about half the time, a fair die should show each face about one time in six. But running an actual experiment gives observed results, and these rarely match the expectation exactly. Ten coin flips might give seven heads, not five. The difference between what is expected and what is observed is at the heart of this unit. Experiments and simulations let us see how that difference behaves, and why a short run can look so far from the chance that produced it.
A few trials look lumpy
With only a handful of trials, results are bumpy and uneven. Roll a die twelve times and the six faces almost never come up twice each; some appear four times, others not at all. This is not a faulty die, just chance at work over a small number of trials. A short experiment is easily swayed by luck, so its results can sit a long way from the expected even split. Seeing this lumpiness is the first lesson: a small experiment is a weak guide to the true chance behind it.
More trials settle down
As the number of trials grows, something steadying happens. Flip a coin ten times and the proportion of heads might be far from one half; flip it a thousand times and it settles close to one half. The more an experiment is repeated, the nearer the observed proportion creeps to the expected chance. This is why simulations run with many trials, often using digital tools to manage the count: a large number of trials smooths out the early lumpiness and reveals the underlying probability that a few trials hide.
Comparing observation and expectation
Putting observed counts beside the expected level shows the pattern directly. After a few rolls, the bars scatter widely around the expected line; after many, they cluster close to it. The gap between observed and expected, measured in proportion to the number of trials, shrinks as trials increase. This comparison is the central tool of the unit: it turns the vague idea that "more is better" into something visible, letting you see exactly how observation and expectation draw together as an experiment is repeated again and again.
Variation between runs
Even with the same number of trials, two runs of an experiment seldom give identical results. One set of twenty flips might give nine heads, the next thirteen. This run-to-run difference is called variation, and it is a normal feature of chance, not a mistake. Variation is larger when trials are few and smaller when they are many, which is why a single short run should never be trusted on its own. Understanding variation explains why repeating an experiment, and repeating it at length, gives a far more reliable estimate than any one run.
Trials and reliability
Bringing these ideas together gives a clear rule: the more trials an experiment or simulation runs, the closer its results come to the expected chance, and the less they vary from run to run. A few trials are lumpy and unreliable; many trials are smooth and dependable. This is why digital simulations, which can run thousands of trials in moments, are so useful for estimating probabilities. Matching the number of trials to the reliability you need is the practical skill at the centre of running chance experiments well.
Where experiments lead
Repeated experiments and simulations are how probability meets the real world. They show that a chance is not just a number on a scale but something that plays out over many trials, with observation drawing ever closer to expectation as the count grows. This idea, that more trials mean more reliable estimates and less variation, carries forward into the statistics and data science of later years, where simulations and large samples are the everyday tools for estimating chances that cannot be calculated directly.