Notion of chance

Random experiments and outcomes. A random experiment is an action whose exact result cannot be predicted in advance, even though every possible result is already known. Tossing a coin, throwing a die, and drawing a card from a shuffled pack are all random experiments — you cannot say in advance which face will show, but you can list every face that could show. Each possible result is called an outcome, and the complete set of all outcomes is the sample space, written $S$. A single coin toss gives $S = \{\text{Head}, \text{Tail}\}$, with 2 outcomes; a single die throw gives $S = \{1, 2, 3, 4, 5, 6\}$, with 6 outcomes. Listing the sample space correctly is the first and most important step in reasoning about chance — get it wrong, and every judgement built on top of it is wrong too.

Equally likely outcomes. Outcomes are equally likely when nothing about the experiment favours one over another. A fair coin has no reason to prefer Head over Tail; a fair, evenly balanced die has no reason to prefer one face over another. This assumption of fairness is what allows chance to be reasoned about through simple counting rather than by running an experiment thousands of times first. It is also an assumption that must be checked, not assumed automatically — a bag holding 5 red balls and 1 blue ball does not give red and blue an equal chance, because there are more ways for red to be drawn.

The scale of chance. Every event, once you fix a sample space, can be placed somewhere on a scale running from impossible to certain:

• An impossible event can never occur — a normal die showing a 7. Its chance sits at 0.
• A certain (sure) event always occurs — a coin landing Head or Tail. Its chance sits at 1.
• An event that is equally likely to happen or not sits near the middle of the scale, close to one half.
• Events can also be simply unlikely (chance close to 0 but not impossible) or likely (chance close to 1 but not certain).

Why repeated trials matter. This scale is not a guess — it is anchored to what actually happens when an experiment is repeated many times. Toss a fair coin a handful of times and almost anything can happen; toss it a thousand times and Heads comes up remarkably close to half the time. No single toss "remembers" earlier tosses, yet the proportion of Heads across many trials settles down and stabilises. That stabilising proportion is called the relative frequency of an event, and it is the experimental bridge between vague chance and an exact number:

where: $f$ is the number of trials in which the event actually occurred, and $n$ is the total number of trials performed. As $n$ grows large, this relative frequency settles down close to the event's true chance — which is exactly why 60 throws of a die give a far more trustworthy picture of fairness than 6 throws do.

There is no single counting formula behind the notion of chance itself — that comes later, once outcomes are known to be equally likely. What makes the scale of chance trustworthy is a piece of reasoning about repeated trials, and it is worth walking through carefully.

Step 1 — an event's chance is really about proportion. When a sample space has several equally likely outcomes, an event's chance is governed by what fraction of the sample space favours it. A bag with 5 red balls and 1 blue ball out of 6 total makes drawing red far more of the sample space than drawing blue, so red is judged likely and blue unlikely — long before any exact number is calculated. This proportion-based reasoning is exactly how the formula for classical probability will later be built, but the notion of chance itself only needs the comparison, not the division.

Step 2 — a single trial is a poor witness. One coin toss, or one die throw, tells you almost nothing about the underlying chance. A fair coin can easily show three Heads in a row purely by luck; that run says nothing about whether the coin is biased. Judging chance from a single trial is the single biggest reasoning error at this stage.

Step 3 — many trials reveal the pattern. As the number of trials $n$ grows, the relative frequency $f/n$ from equation (1) fluctuates less and less, and settles close to the event's true chance. This is not because the coin or die develops a memory — each trial is still independent of the last — but because random fluctuations that push the count up in some trials are increasingly balanced out by fluctuations that push it down in others, once there are enough trials for balancing to happen.

Step 4 — this is why fairness is judged by big experiments, not small ones. A teacher who wants to check if a die is fair does not throw it 3 times; she throws it 60 or 600 times and compares each face's count to the expected value, because only a large number of trials lets random noise average out and the true pattern of chance show through.

Meera throws an ordinary die 60 times in a class activity. She records that "6" comes up 9 times and "1" comes up 11 times. Comment on whether the die appears fair, and describe the chance of getting a "6" in both qualitative and relative-frequency terms.

Step 1 — expectation for a fair die. The sample space for one throw is $S = \{1, 2, 3, 4, 5, 6\}$, with 6 equally likely outcomes if the die is fair. Over 60 throws, each face is expected to appear close to $\dfrac{60}{6} = 10$ times — not exactly 10 every time, since real experiments fluctuate around the expected value.

Step 2 — compute the relative frequencies. Using equation (1) with $n = 60$: for "6", $f = 9$, so relative frequency $= \dfrac{9}{60} = 0.15$. For "1", $f = 11$, so relative frequency $= \dfrac{11}{60} \approx 0.183$. The expected relative frequency for a fair die on any face is $\dfrac{1}{6} \approx 0.167$.

Step 3 — compare. Both $0.15$ and $0.183$ sit close to the expected $0.167$ — the gap is small enough to be ordinary trial-to-trial fluctuation, not evidence of bias.

Step 4 — conclusion. Since the counts are near the expected value, the die appears fair: no face is dramatically over- or under-represented. Getting a "6" on any throw is reasonably close to equally likely with the other five faces, exactly as expected for a fair die.

Competitive stretch

Suppose Meera repeats the experiment with 600 throws instead of 60, and "6" now comes up 108 times. Its relative frequency is $\dfrac{108}{600} = 0.18$ — almost the same fraction as before, but now backed by ten times as much data. This is the stronger, more trustworthy verdict on fairness, because a larger $n$ shrinks the effect of chance fluctuation, exactly as Step 3 above explains.

• Thinking that if a coin has shown Heads several times in a row, Tails is "due" next — this is the gambler's fallacy. Each toss is independent; the coin has no memory of past tosses.
• Confusing "impossible" with merely "unlikely." An impossible event, like a die showing 8, can never happen, while an unlikely event can happen, just rarely.
• Assuming outcomes are automatically equally likely without checking. In a bag of 5 red and 1 blue ball, drawing red and drawing blue are not equally likely, even though there are only two colours.
• Expecting a small number of trials, say 6 throws of a die, to exactly match the theoretical expectation. Real experiments fluctuate; the underlying pattern of chance only becomes clear over many repeated trials.

Ask the tutor: Want to try classifying the chance of events for a spinner, a pack of playing cards, or a bag of mixed coloured beads — or see how relative frequency behaves as trial counts grow even larger? Ask the AI tutor to work through more notion-of-chance examples with you.