Does or include both events?

In standard probability notation, yes, unless the question explicitly asks for exactly one.

Can two event percentages be added directly?

Only when the intended union has no overlap, or when overlap has already been handled appropriately.

Do repeated trials automatically become independent?

No. Shared conditions, sampling without replacement, or persistent system states can make trial outcomes dependent.

AI Glossary / Data science and analytics / Beginner

Compound Events

Definition

Compound events are events formed from combinations of simpler events, often using and, or, and not. In introductory probability, the term also describes an event containing more than one elementary outcome. The underlying task is to specify exactly which outcomes satisfy the combined condition, then calculate their probability without double-counting or assuming independence without justification.

What it means

A business may care whether a request is late or incorrect, whether two required checks both pass, or whether at least one of several services fails. These are different combinations of events. In standard probability usage, or is inclusive: A or B includes cases where both occur. And requires both conditions. Clear language is essential because a small change in the condition can produce a very different result.

How it works

Represent the conditions as sets within the same sample space. For an inclusive union, P(A or B) = P(A) + P(B) − P(A and B). Subtracting the overlap prevents double-counting. For an intersection, P(A and B) = P(A) × P(B given A), when the conditional probability is defined. This becomes P(A) × P(B) only when the events are independent. Tables, tree diagrams, or explicitly listed outcomes can help verify what is being counted.

Three practical examples

Service quality: a team measures the proportion of requests that are late, incorrect, or both, without counting the same request twice.
AI release checks: a model version must satisfy a quality criterion and an operational-readiness criterion before it qualifies for a defined release decision.
Customer analysis: analysts examine customers who bought one product or another, distinguishing that inclusive group from customers who bought exactly one of them.

Think of it this way

Think of two overlapping circles on a sheet, one for red objects and one for round objects. The overlap contains objects with both properties, while the combined area contains anything with either property. The picture clarifies logical combinations. It does not establish probabilities from the drawn areas unless the diagram has been constructed to represent them quantitatively.

Worked example

In a fictional set of 100 evaluated AI responses, 30 are late and 20 are factually incorrect. Twelve are both late and incorrect. The event late or incorrect contains 30 + 20 − 12 = 38 responses, so its observed proportion is 38%. The event late and incorrect contains 12%, and exactly one of those problems occurs in (30 − 12) + (20 − 12) = 26%.
If an analyst multiplied 0.30 by 0.20, the result would be 6%, not the observed 12% intersection. The multiplication would have assumed independence, which these proportions do not support. The conditional proportion incorrect given late is 12/30 = 40%, illustrating why the general multiplication rule uses the conditional value: 0.30 × 0.40 = 0.12.

Illustrative case study

In a fictional operations dashboard, separate charts show failed requests and timed-out requests. A manager adds the chart totals to estimate the number of affected customers. The data team finds that some requests appear in both groups and some customers made several requests. It creates a clearly defined request-level union, then separately counts affected customers using an appropriate customer identifier. The example shows two distinct risks: overlapping events and changing the unit of analysis. Correct compound-event arithmetic cannot fix an undefined denominator.

Common misconceptions and mistakes

Mutually exclusive and independent are different properties. Mutually exclusive events cannot occur together, while independent events do not change one another's probabilities. Nonzero-probability mutually exclusive events are not independent. Do not assume or means exactly one unless that is specified. Do not multiply marginal probabilities simply because the sentence contains and. Also, a condition combining two tests is not automatically a fair or useful decision rule; the probability calculation describes the rule rather than justifying its purpose.

Business use

Businesses use compound events to understand combined risks, eligibility conditions, and quality outcomes. Specify the reporting unit, period, and handling of missing observations. For AI systems, keep joint outcome information instead of only separate percentages for correctness, speed, and other properties. Without overlap data, the exact combined failure rate may be unknown. A transparent report can state that limitation rather than inventing independence. Decisions based on several model checks should likewise consider whether the checks share data or common failure mechanisms.

Check your understanding

Forty requests fail a format check and 25 fail a content check, with 10 failing both. How many fail at least one check?

Answer: Forty plus 25 minus 10 gives 55. Adding the two totals without subtracting the overlap would count those 10 requests twice.

Related terms

Learn next

construct a two-by-two table for two binary checks and calculate the union, intersection, and conditional proportions directly from it. Compare inclusive or with exactly one, then determine whether the observed table is consistent with independence before using a simplified multiplication rule.

Sources

Sources: OpenStax's original teaching material explains compound events built with or and the addition rule: https://openstax.org/books/contemporary-mathematics/pages/7-8-the-addition-rule-for-probability . Its companion section explains conditional probability and the multiplication rule for and: https://openstax.org/books/contemporary-mathematics/pages/7-9-conditional-probability-and-the-multiplication-rule . Relevant textbook excerpts were consulted. All service counts and numerical examples here are original illustrations, not observations from a real AI system.

Previous: Compositional semantics Next: Compound symmetry