Can a sample space be infinite?

Yes; it can be countably infinite or continuous.

Is an event the same as one outcome?

An event can contain one outcome or many.

Can two different sample-space descriptions represent the same process?

Yes, at different levels of detail, provided the probability model and questions are handled consistently.

AI Glossary / Statistics and probability / Intermediate

Sample space

Definition

A sample space is the set of all possible outcomes represented by a probability model for an experiment or random process. It is commonly denoted S or Omega. Events are measurable collections of those outcomes, and probabilities describe how likely the events are under the model. The sample space itself specifies possibilities, not their probabilities.

What it means

Before asking how likely something is, define what could happen and what detail counts as an outcome. For three tests, an outcome might record each test result in order, or a coarser representation might record only the number of failures.
The choice should preserve the information needed for the question. A sample space is not the dataset already collected, a survey sampling frame, or a list of only the outcomes observed so far. Unobserved possibilities can still belong to it.

How it works

Specify the experiment and observation rule. Drawing one item, drawing several items with replacement, and drawing several without replacement generate different possibilities. Likewise, recording order gives more detailed outcomes than recording an unordered collection.
Construct mutually exclusive elementary outcomes at the chosen level of detail. For a finite experiment, list them or use a systematic counting rule. Include every outcome the model permits so probabilities are not assigned to an incomplete set by accident.
Define events as collections of outcomes relevant to a question. At least one failure, exactly one failure, and failure on the first trial are different events even when they overlap.
Add a probability model. A finite sample space does not imply equally likely outcomes. Counting favorable outcomes and dividing by the total works only when the elementary outcomes being counted have equal probabilities.
Use independence only when justified. For successive tests, multiplying marginal probabilities requires an appropriate independence assumption; otherwise use joint or conditional probabilities. Listing all combinations does not itself establish independence.
For continuous quantities, the sample space may be an interval or another uncountable set. Probabilities are assigned to suitable events such as intervals. An individual point can have probability zero while still being a possible value under a continuous model.
A random variable maps outcomes to numbers. Different detailed outcomes can produce the same value of that variable. Its range is therefore often smaller or less detailed than the original sample space, and probabilities must be aggregated over outcomes that map to the same value.
Review the model boundary. If real outcomes can include missing, delayed, invalid, or interrupted results, decide how those possibilities are represented. A model that omits important operational outcomes may give a mathematically correct answer to an inadequate version of the business question.

Three practical examples

Quality testing: a model distinguishes sequences of pass and fail results across several checks.
Reliability analysis: analysts define possible operating and failure states of a system before assigning probabilities.
AI evaluation: a test protocol specifies whether a result can be correct, incorrect, abstained, or unavailable, rather than silently ignoring some outcomes.

Think of it this way

Think of the sample space as the menu of outcomes the model allows. The probabilities are the portions of likelihood assigned to items or collections on that menu. Writing a longer or shorter menu changes how outcomes are described; it does not automatically make every listed item equally likely.

Worked example

Three fictional independent checks each pass with probability 0.8 and fail with probability 0.2. Let P mean pass and F mean fail, preserving the order of checks. The sample space has eight sequences: PPP, PPF, PFP, FPP, PFF, FPF, FFP, and FFF.
Exactly one failure corresponds to PPF, PFP, and FPP. Each has probability 0.8²×0.2 = 0.128, giving total probability 3×0.128 = 0.384.
It would be wrong to use 3/8 because the eight sequences are not equally likely. For example, PPP has probability 0.8³ = 0.512, while FFF has probability 0.2³ = 0.008.
At least one failure is the complement of PPP, so its probability is 1−0.512 = 0.488. If the recorded random variable is the number of failures, its possible values are only 0, 1, 2, and 3; the detailed sequences have been grouped, not made equiprobable.

Illustrative case study

Fictional case: Maple Automation models the outcomes of three inspection checks. Its first report counts successful and failed checks but omits cases where a sensor produces no valid reading.
The team revises the outcome definitions to represent unavailable readings and decides how they affect the operational decision. It then estimates a probability model supported by actual testing, instead of assuming that the simplified independent-check illustration describes the real equipment.

Common misconceptions and mistakes

Possible outcomes are not necessarily equally likely.
A sample space is not the same as a sample of observed data.
An event may contain many elementary outcomes, and events need not be mutually exclusive.
Zero probability does not always mean impossibility in a continuous model.
The count of outcomes depends on whether order and other details are retained. Changing that representation requires consistent probability aggregation, not arbitrary reassignment of equal probabilities.

Business use

Use explicit sample spaces to make risk calculations, simulations, and evaluation rules understandable. Listing what can happen often reveals missing outcomes or ambiguous definitions before a model is built. For an AI service, include operational states such as abstention or failure to return an answer when those states affect the decision.

Check your understanding

Why is exactly one failure not assigned probability 3/8 in the example?

Answer: The eight sequences have unequal probabilities.

Does observing only PPP in a short dataset remove FFF from the model's sample space?

Answer: No; lack of observation does not make it impossible under the stated model.

Related terms

Learn next

Write the ordered sample space for a short repeated experiment. Define two overlapping events and compute their probabilities using a stated model rather than assuming equal likelihood.

Sources

Penn State STAT 414, Properties of Probability: https://online.stat.psu.edu/stat414/Lesson02 OECD Glossary of Statistical Terms, Sample space: https://www.oecd.org/content/dam/oecd/en/publications/reports/2008/09/oecd-glossary-of-statistical-terms_g1gh9ad7/9789264055087-en.pdf The three-check probability model and operational case are original fictional illustrations. OpenStax, Introductory Statistics 2e, Terminology (outcomes, events, and equal-likelihood counting): https://openstax.org/books/introductory-statistics-2e/pages/3-1-terminology

Previous: Sample size disparity Next: Sample Standard Deviation