What fair means in maths
In probability, a fair coin is a model: two outcomes, heads and tails, each with probability exactly 1/2, with every toss independent of the others. It is the starting point for many examples because it is simple and symmetric. It is called an assumption because no physical object can be guaranteed to match it exactly.
The fair coin model has two separate parts. One is the 50-50 split. The other is independence, meaning earlier results have no effect on later ones. A coin could in principle fail either: it might favour one side, or the way it is tossed might link one result to the next. When people say a streak shows a coin is “not fair”, they are usually mixing these up with the gambler’s fallacy.
Is a real coin exactly 50-50?
For a coin that is caught in the air or lands on a hard surface, the answer from physics is “very close, but perhaps not exactly”. A tossed coin follows ordinary mechanics. The outcome is decided by how fast it is launched, how quickly it spins and how long it stays in the air. If those were known exactly, the result would in principle be predictable. The randomness comes from tiny differences between tosses that nobody can control or measure by eye.
A well-known research study led by the mathematician Persi Diaconis, together with colleagues, analysed the physics of coin tossing and also tested real tosses with high-speed cameras. In general terms, their conclusion was that a typical flip is close to fair, with a small tendency for the coin to land on the same face it started on. The size of that tendency was reported as only slightly above one half, around 51 percent. I mention that as an approximate figure from the research, not a precise property of any coin you might hold.
Two things are worth noting. First, an effect that small would be nearly impossible to notice in everyday use. Second, it depends on the toss technique. A deliberate flip made by a machine or a very controlled hand could give very different results from a casual flick.
Why a small bias is hard to detect
Suppose a coin really did land heads 51 percent of the time. How many tosses would you need to notice? With 100 tosses, the typical random wobble in the number of heads is about 5 (the square root of 100 x 0.5 x 0.5 = 5). The 51 percent bias predicts one extra head, which is well inside the noise. With 10,000 tosses, the typical wobble is about 50 heads, and the bias would add about 100. Only at that kind of scale does the pattern start to separate from chance, and even then it is evidence, not proof.
That is a useful lesson about small sample sizes. A short experiment cannot tell a very slightly unfair coin from a fair one. Conversely, a coin that gives 7 heads in 10 tosses is not shown to be unfair, since results like that arise fairly often by chance.
Other ways to use a coin
Not every coin action is a toss.
- Spinning a coin on a table is generally considered less fair than tossing it. Coins often have uneven edges and slightly uneven weight, which can make one face more likely to end up down. Reports of strong bias in spun coins exist, though how strong depends on the coin.
- Standing a coin on its edge and knocking the table can bias outcomes in a similar way.
- Landing on the edge is possible when a coin falls on a soft or uneven surface. It is rare enough to ignore in most examples, but it is a reminder that the model with two outcomes is a simplification.
Fairness and method
A flip can be made fairer by good habits, such as letting the coin spin well in the air and catching or letting it land on a flat surface. Even so, in everyday decision-making, a coin flip is accepted as fair because no one can reliably steer it, not because the physics is perfect. For practical purposes, such as deciding who starts a game, it works. For serious uses, such as large lotteries or draws, organisers tend to use better-controlled methods than a hand toss.
Updating your view of a coin
If you did want to test whether a coin favours a side, the sensible way is to count results over many tosses and then work out how unusual the outcome would be for a fair coin. Another approach is belief updating, where you begin with an assumption that the coin is probably fair and adjust it as evidence arrives. Neither approach gives certainty, and both depend on the toss method staying the same.
Why it matters for gambling
Casino games do not rely on hand-tossed coins, but the ideas carry across: a model may be an idealisation, and any tiny departure from fairness is far smaller than the built-in house edge in most games. Being slightly unfair in the coin’s favour does not make a game a good bet. Gambling carries a risk of harm, and anyone struggling should reach out to a local support service.