AI Glossary / Statistics and probability / Beginner

Probability distribution

Definition

A probability distribution specifies how probability is assigned to the possible values of a random variable. It describes the variable's uncertainty, including which values are possible and how probability is spread among them. For a discrete variable, probabilities can be listed in a probability mass function. For a continuous variable with a density, interval probabilities are obtained from areas under that density. A cumulative distribution function provides a common representation across discrete, continuous and mixed cases.

What it means

A distribution contains more information than an average. Two processes can share the same mean while differing greatly in variability, asymmetry or the chance of extreme outcomes. The distribution can describe an assumed mechanism, a fitted statistical model or an empirical summary of observed data. These are related but different objects. An empirical distribution describes the sample in hand; applying it to future cases requires assumptions about how those cases relate to the sample.

How it works

Define the random variable and its support, meaning the values it can take. Assign nonnegative probability masses summing to one, or specify another valid representation such as a density integrating to one. Use the distribution to calculate event probabilities, expectations, quantiles and other summaries where they exist. Estimate unknown parameters from data when appropriate. Check whether the chosen distribution captures important features, and distinguish a marginal distribution for one variable from a joint distribution describing several variables together.

Three practical examples

Example 1: A shop models daily numbers of returns using probabilities for different counts. Example 2: A laboratory models the continuous variation in a measurement to assess tolerance limits. Example 3: A planner uses a joint distribution of demand and delivery time because simultaneous high demand and long delays matter. The third example requires dependence information: knowing the two separate distributions is generally insufficient to determine the probability of their joint extremes.

Think of it this way

Think of a distribution as a complete probability map for a quantity. The mean marks one center on the map, but it does not show every hill, tail or gap. A map with two separated peaks can have its mean in a region rarely visited. Reading only that center can therefore hide exactly the variation that matters for a decision.

Worked example

Suppose daily urgent jobs X can take values 0, 1, 2 or 3 with probabilities 0.10, 0.40, 0.30 and 0.20. The probabilities sum to one. The probability of at least two jobs is 0.30 + 0.20 = 0.50. The mean is 0 × 0.10 + 1 × 0.40 + 2 × 0.30 + 3 × 0.20 = 1.60 jobs. The second moment is 0 + 0.40 + 1.20 + 1.80 = 3.40, so variance is 3.40 - 1.60 squared = 0.84. Under the convention that the 50th percentile is the smallest value whose cumulative probability reaches 0.50, that percentile is 1. The mean of 1.60 is a useful average even though 1.60 jobs is not a possible daily count.

Illustrative case study

Fictional case study: Pine Services staffs urgent work using the average daily job count. An analyst presents the full illustrative count distribution and shows why a capacity of one job is exceeded on half the modeled days. Managers compare several capacity options using their costs and service requirements. They then test whether the historical distribution differs across weekdays. This case is fictional and demonstrates the value of distributional information, not a validated staffing recommendation or a claim that demand remains stationary.

Common misconceptions and mistakes

Common mistakes include assuming every distribution is normal, treating the mean as the most likely outcome, and confusing a density value with probability. A probability distribution need not have a finite mean or variance. Discrete quantiles can depend on convention, so software results should be interpreted carefully. Matching a few summary statistics does not establish that two distributions are the same. Separate marginal distributions also do not determine dependence unless additional assumptions are supplied.

Business use

Business use: Distributions support planning for variability rather than only averages. They help estimate overload frequencies, compare service guarantees and evaluate uncertain costs. Choose representations that respect the target: counts, bounded percentages and strictly positive measurements have different supports. Explain whether probabilities come from observed proportions, a fitted family or expert assumptions. Inspect the tails and subgroups that drive the decision, and revisit the model when operating conditions change.

Check your understanding

In the worked example, why does a mean of 1.60 not imply that most days have two jobs?

Answer: The mean is a probability-weighted average, while the most probable count is one job at probability 0.40.

What additional information is needed to combine demand and delivery-time risks?

Answer: Their joint dependence structure, not only their separate distributions.

Related terms

Random variables take values according to probability distributions

Learn next

Build a cumulative probability table from a small discrete distribution and calculate its mean, variance and selected quantiles. Compare it with another distribution having the same mean but different variability. Then study a continuous density and a mixed distribution to see why no single graphical representation fits every case. Keep support and dependence assumptions visible throughout the exercise.

Discrete distribution

Sources

Sources: NIST/SEMATECH e-Handbook, What is a Probability Distribution: https://www.itl.nist.gov/div898/handbook/eda/section3/eda361.htm ; The Aalto Dictionary of Machine Learning, probability distribution: https://aaltodictionaryofml.github.io/ADictML_Main.pdf ; Penn State STAT 414: https://online.stat.psu.edu/stat414/ . These university and institutional sources ground the probability framework. The urgent-job distribution, calculations and fictional staffing case are original illustrations.

Previous: Probability density function Next: Probability integral transform