AI Glossary / Statistics and probability / Beginner
Probability
Definition
Probability is a numerical measure assigned to an event within a defined model of uncertainty. Its values lie between zero and one, with larger values indicating greater probability. The probability of the complete set of possible outcomes is one, and probabilities of mutually exclusive events add. These rules make probability statements coherent. The interpretation and evidence supporting a particular value still depend on the situation, such as a random experiment, an observed population or a model-based judgment.
What it means
A probability of 0.20 can be written as 20%. It does not mean that an event must occur exactly once in every five attempts. In repeated comparable trials, observed proportions can provide evidence about probabilities, but individual sequences fluctuate. Probability also requires a clearly defined event and information set. The probability of a delay for all deliveries can differ from its probability for deliveries during a storm, because those statements refer to different conditions.
How it works
Define the possible outcomes and the event of interest. Assign probabilities from a justified design, empirical model or other stated basis. Use the addition rule to combine alternatives, subtracting overlap when events can occur together. Use conditional probability when information changes the relevant group of outcomes. Multiply probabilities directly for independent events; otherwise account for conditional dependence. Check that the resulting values remain coherent and that the assumptions match the real question rather than a convenient substitute.
Three practical examples
Example 1: A quality team estimates the probability that a randomly selected item from a specified production period has a defect. Example 2: A weather forecast assigns a probability to rainfall at a defined location and time. Example 3: A planning model calculates the probability that demand exceeds a stated capacity. Each statement needs its target and conditions. A percentage without a population, time period or event definition can look precise while remaining operationally ambiguous.
Think of it this way
Think of a probability model as distributing a fixed budget of one unit of belief or long-run frequency across possible outcomes. An event collects the share assigned to the outcomes it includes. Overlapping events can collect some of the same share, so adding their probabilities without adjustment can count it twice. This analogy helps with arithmetic, but the justification for the allocation must come from the actual model and evidence.
Worked example
Suppose 100 equally likely records describe past service requests. Thirty involve a delay, 25 involve a correction, and 10 involve both. A randomly selected record has probability 0.30 of delay and 0.25 of correction. The probability of either is (30 + 25 - 10)/100 = 0.45. Among corrected requests, the delay probability is 10/25 = 0.40, so the two events are not independent: independence would imply overlap 0.30 × 0.25 = 0.075 rather than 0.10. Separately, if two future components each fail with probability 0.10 independently, the probability that neither fails is 0.90 squared = 0.81. The independence assumption is essential to that final calculation.
Illustrative case study
Fictional case study: Willow Services publishes a dashboard showing a single delay percentage. Managers initially apply it to every request, but an analyst separates urgent and routine work and finds different historical rates. The team rewrites the dashboard labels to identify the relevant request group and period. It also separates observed proportions from model forecasts. This fictional case illustrates clearer probability communication; it does not establish that historical rates remain constant or that the dashboard changes future service outcomes.
Common misconceptions and mistakes
Common mistakes include confusing probability with odds, adding overlapping event probabilities without adjustment, and multiplying dependent probabilities as if the events were independent. A rare event is not impossible, and a likely event is not guaranteed on a particular trial. Probability zero need not mean logical impossibility in a continuous model: any single exact value can have zero probability. Observed frequencies also contain sampling variation and may poorly represent a different population.
Business use
Business use: Probability makes uncertainty explicit in forecasts, quality monitoring and operational planning. Use it alongside the consequences of outcomes rather than ranking decisions by probability alone. A low-probability event can matter greatly when its consequences are large, while a frequent minor event may justify a different response. State the evidence, conditions and uncertainty behind estimates. Check whether the assumed events and time horizon match the decision actually being made.
People also ask
Are probability and odds the same?
No. For probability p, odds in favor are p/(1 - p), when p is below one. Q: Does a 70% prediction become wrong if the event fails to occur once? A: Not necessarily; probability predictions need evaluation over appropriate collections. Q: Are equally possible outcomes always equally probable? A: No. Equal weighting needs a justification, not just a list of possibilities.
Check your understanding
Why subtract the ten requests with both delay and correction?
Answer: They were counted in both individual totals but should appear only once in the union.
Does a 10% failure probability imply exactly one failure in the next ten components?
Answer: No. The actual count varies; the expected count under comparable conditions is one, which is a different statement.
Related terms
- An event is a set of possible outcomes
Learn next
Draw a two-way table and calculate a joint probability, two marginal probabilities and both directions of a conditional probability. Compare independent and dependent examples. Then practice distinguishing an observed sample proportion from an unknown population probability and a model forecast. Pay particular attention to the event definition before performing any arithmetic.
More in Statistics and probability
- Poisson distribution - The Poisson distribution is a probability distribution for a count taking the values 0, 1, 2 and so on.
- Policy capturing - Policy capturing is a research method for describing how people combine information when making judgments or decisions.
- Poly-Weibull distribution - The poly-Weibull distribution is a continuous time-to-event distribution constructed by adding several Weibull-type hazard functions.
- Pólya urn model - A Pólya urn model is a probability model in which sampling changes the probabilities of future samples through a specified reinforcement rule.
- Polynomial regression - Polynomial regression models an outcome using powers of one or more numerical predictors.
- Polynomial trend coefficient - A polynomial trend coefficient is a numerical weight used with other such weights to test a specified trend across ordered group means.
Sources
Sources: Penn State, STAT 414, probability foundations and conditional probability: https://online.stat.psu.edu/stat414/ ; The Aalto Dictionary of Machine Learning, probability and probability distribution entries: https://aaltodictionaryofml.github.io/ADictML_Main.pdf . These university sources provide the underlying probability framework. The service-request table, component calculation and fictional dashboard case are original illustrations.
Previous: Probabilistic Topic Model Next: Probability based disclosures