Probability Distributions: Mapping Odds to Outcomes

Why it exists

To move from simple guessing to formally quantifying uncertainty. We need a rigorous way to describe all the things that could happen in a random process and how likely each one is. Without this, we can't build predictive models or make informed decisions when outcomes aren't guaranteed.

The mental model

Think of a probability distribution as a complete rulebook for a game of chance. For a fair six-sided die, the rulebook says: a 1 has a 1/6 chance, a 2 has a 1/6 chance, and so on for all six faces. The distribution lists every possible outcome and its exact probability. It's not just a list of possibilities; it's a list with their weights, and the total weight for all possible outcomes must sum to 100%.

How it works

A probability distribution is a function that assigns a probability (a number between 0 and 1) to every possible outcome of a random experiment. For discrete events, like a die roll, you can list each outcome and its probability. For continuous events, like a person's exact height, you can't list every possibility. Instead, the distribution describes the probability of the outcome falling within a certain range (e.g., the chance of being between 175cm and 180cm tall). The core axiom is that the probabilities of all possible, non-overlapping outcomes must sum to 1.

When to use it

Use a probability distribution whenever you need to model or reason about a random process. This is fundamental in many fields. For example, in finance, to model asset returns; in operations, to model customer arrival times at a service desk; in machine learning, to represent the uncertainty in a model's predictions; and in A/B testing, to determine if a result is statistically significant.

When not to use it

A probability distribution is for random phenomena, not deterministic processes where the outcome is certain. More importantly, applying the wrong type of distribution is a major error. Forcing a normal distribution (bell curve) onto data that is clearly skewed, like income levels or website traffic, will produce nonsensical results and bad forecasts. The choice of distribution must match the underlying nature of the process being modeled.

One canonical example

The simplest example is the Bernoulli distribution, which models a single event with two possible outcomes, like one coin flip. Let's label "success" (e.g., heads) as having probability 'p' and "failure" (tails) as having probability '1-p'. For a fair coin, p = 0.5. The distribution simply states: P(Heads) = 0.5 and P(Tails) = 0.5. This is the foundational building block for more complex distributions that model multiple trials, like the Binomial distribution.