How to Calculate Probability: The Basic Formulas and Where Intuition Gets It Wrong

Working out a probability, in the simplest case, means counting two numbers and turning them into a ratio. OpenStax, in its introductory statistics textbook, defines the classical probability of an event as the ratio between the number of favorable outcomes and the total number of possible outcomes, provided every outcome in the sample space is equally likely. With a fair six-sided die, the probability of rolling a number of at least five is 2/6: the favorable outcomes are two, the 5 and the 6, out of six possible in total.

The classical probability formula

The classical calculation doesn’t need a single roll: it only needs counting. There is one condition to check first, though, true for a fair die but not for every situation: each possible outcome must be exactly as likely as the others. When that condition holds, the formula is:

Worked example: rolling a fair die, the probability of an even number (2, 4, or 6) follows the same formula: three favorable outcomes out of six possible, so 3/6.

This is also how to find probability in any situation with a finite, equally likely set of outcomes — count what counts as a win, count everything that could happen, and divide.

Relative frequency and the law of large numbers

There’s another route to the same number, one that goes through observation instead of counting on paper. Relative frequency is the proportion of times an event actually occurs over a large number of repeated trials. OpenStax connects the two definitions by showing that, rolling a die many times in practice, the observed frequency of “a number of at least five” gets closer to the theoretical 2/6 value worked out with classical probability. For a broader look at how single ratios like this one fit into collecting and summarizing data, probability and statistics is a useful starting point.

The law of large numbers, as the textbook presents it, says that the more times an experiment gets repeated, the closer the observed relative frequency gets to the theoretical probability calculated beforehand. With few trials, the frequency bounces around quite a bit; with many trials, it settles down. The two definitions of probability — classical, worked out ahead of time, and relative frequency, observed in the field — converge on the same value as trials pile up.

Independent events: the multiplication and addition rules

Two events are independent when knowing that one of them occurred doesn’t change the probability that the other occurs. OpenStax formalizes independence with three equivalent conditions, including the multiplication rule: the probability that both events A and B occur is the product of their individual probabilities, P(A)×P(B).

In the textbook’s example, with P(G)=0.6, P(H)=0.5, and P(G and H)=0.3, the product of the two individual probabilities (0.6×0.5=0.3) matches the probability of the intersection — confirming that events G and H are independent.

When what’s needed instead is the probability that at least one of the two events occurs, the addition rule comes in. OpenStax states it as the sum of the individual probabilities minus the probability of their intersection, P(A)+P(B)−P(A and B): the subtraction keeps the cases where both events happen together from being counted twice.

The gambler’s fallacy: when the past doesn’t count, but feels like it does

Someone who flips a fair coin and gets heads five times in a row often expects tails to be “more likely” on the sixth flip. According to OnlineStatBook, this error comes from believing that the outcome of an event independent of the past is influenced by earlier results in the same sequence — the gambler’s fallacy. In reality, even after five consecutive heads, the probability that the next flip comes up tails stays at 50%, because flips are independent events and there’s no quota a result has to fill.

The source uses a two-flip case to show it plainly: the outcome of the first flip doesn’t change the probability of the second one at all. A run of five heads leaves no trace on the coin; the coin isn’t keeping score.

The birthday problem: why a single classroom is enough

The birthday problem shows the same intuition missing the mark on the low side. Harvey Mudd College lays it out this way: with 23 people in a room, the probability that at least two of them share a birthday already passes 50%. The proof first works out the probability that all 23 people have different birthdays, multiplying the shrinking probabilities that each new person added doesn’t repeat a day already taken by the others, then gets the probability of at least one match by subtracting that value from 1. With 23 people, this probability already passes 50%. With 57 people, Harvey Mudd College reports that the probability of at least one shared birthday climbs past 99%.

Picturing the wrong comparison is what throws people off: one person against the other 22, a matchup that would leave little room for a coincidence. The real count runs over every possible pair inside the group: with 23 people, the number of pairs to check is already large, far larger than 23 itself. The more pairs get checked, the higher the chance that at least one of them matches.

Base rate fallacy: when the description beats the numbers

The third case is about how new information gets combined with starting proportions. A study published on PubMed Central defines the base rate fallacy, or base rate neglect, as people’s tendency to not properly combine prior probabilities — the base rate of an event in a population — with new information received, the way Bayes’ theorem would require.

The classic problem used to measure this error, drawn from Kahneman and Tversky’s experiment and reported in that PubMed Central study, describes a person picked at random from a group made up of, say, 30 engineers and 70 lawyers: the description matches the stereotype of the smaller group, the engineers. Participants in the experiment tend to ignore the 30-against-70 split and judge only from how closely the description fits the stereotype — the same snap judgment that stereotypes and prejudice produce far beyond a probability classroom.

Worked example: in a group of 30 engineers and 70 lawyers, even a description that sounds “like an engineer” still leaves it more likely, by sheer proportion, that the person picked at random is a lawyer — unless the description is specific enough to outweigh that starting split.

The underlying habit in all three cases is the same one at the heart of the scientific method: check a claim against numbers gathered beforehand instead of trusting whichever story feels most convincing in the moment.