Does random mean equal chance?

Not necessarily. Unequal chances are compatible with probability sampling when they are known and positive.

Does a random sample guarantee representativeness?

No. Chance imbalance and nonsampling errors remain possible.

Can a volunteer survey be analyzed as a simple random sample?

Not merely because many people volunteered; its selection mechanism differs.

AI Glossary / Statistics and probability / Beginner

Random (probability) sample

Definition

A random, or probability, sample is selected through a chance-based design in which every unit in the target population covered by the design has a known, positive probability of inclusion. These probabilities support estimation of population quantities and sampling uncertainty. They need not be equal. Simple random sampling is one probability design, while stratified, cluster, and unequal-probability designs provide other approaches. A casually chosen or convenient set of observations is not a probability sample merely because its selection feels unpredictable.

What it means

The essential feature is a specified selection mechanism that makes inclusion probabilities calculable. Randomness is introduced by the design, rather than inferred from how representative the resulting dataset happens to look.

A well-designed probability sample provides a basis for generalizing beyond the observed units. That basis still depends on the sampling frame, response, measurement quality, and correct analysis. If the frame omits a group entirely, random selection within the frame does not create evidence about that excluded group. A selected sample and the subset that actually responds must also be distinguished.

How it works

Define the population, its units, the reference period, and the quantities to estimate. Construct or obtain a sampling frame with adequate coverage and resolve duplicate or ineligible records. Choose a design suited to the information needed and the cost of collecting it.

Use a documented random mechanism to select units. Record inclusion probabilities and any strata, clusters, or selection stages needed for analysis. For simple random sampling without replacement, each possible sample of the chosen size is equally likely. Other probability designs can intentionally sample some groups more heavily.

Estimate population quantities using methods that reflect the design. Inverse inclusion probabilities often provide starting weights, while nonresponse adjustment and calibration may require additional information and assumptions. Estimate uncertainty with the correct design, not an automatic formula for independent observations when the sample is clustered. Preserve selection records so the procedure can be audited and reproduced.

Three practical examples

1. A retailer randomly selects customer accounts from a complete active-account frame to estimate satisfaction during a defined quarter.

2. A manufacturer samples items from production lots using known selection probabilities to estimate the frequency of defects.

3. A city survey samples addresses within geographic strata, deliberately allocating enough observations to smaller districts for useful local estimates.

Think of it this way

Think of a transparent lottery whose rules are written before the draw. Every eligible ticket has a calculable chance, even if some categories deliberately receive different chances. The fairness of the inference comes from knowing and using those rules, not from assuming that every realized draw will look perfectly balanced.

Worked example

Consider an original stratified sampling example with 100 customer accounts. There are 80 standard accounts and 20 premium accounts. Select 8 standard accounts and 10 premium accounts by simple random sampling within each group, with complete response assumed.

A standard account has inclusion probability 8/80 = 0.10 and initial weight 10. A premium account has inclusion probability 10/20 = 0.50 and weight 2. Suppose 4 sampled standard accounts and 8 sampled premium accounts report satisfaction.

The unweighted sample proportion is 12/18, approximately 66.7%. It overrepresents premium accounts relative to their population share. The design-weighted estimated satisfied total is 4 times 10 plus 8 times 2 = 56 accounts. The weighted denominator is 8 times 10 plus 10 times 2 = 100. The estimated population satisfaction proportion is therefore 56%.

This calculation illustrates unequal probabilities and weighting, not a complete uncertainty analysis. A standard error should respect the stratified design and sampling fractions. The assumptions of accurate frame membership and complete response are stated because failures in either would require further work.

Illustrative case study

In a fictional subscription business, a team surveys every tenth person who happens to contact support and calls the result a random customer sample. A reviewer observes that customers who never contact support cannot enter the sample.

The team defines a population of active customer accounts and draws a probability sample from that frame. It monitors response by account type and documents remaining coverage limitations. The revised report distinguishes selection uncertainty from possible nonresponse and question-wording problems. Management gains a clearer basis for interpreting the estimate without claiming that random selection has removed every source of error.

Common misconceptions and mistakes

Probability sampling does not require equal selection probabilities, and equal probabilities alone do not make every possible sample equally likely. Do not confuse random sampling with random assignment to treatments: one concerns population selection, the other allocation within a study.

Large convenience samples do not become probability samples through size alone. A random split of a biased dataset preserves its underlying coverage problems. Replacing unavailable selected units with convenient substitutes can alter inclusion probabilities. Finally, a sample can depart from population proportions by chance; this alone does not show that the random mechanism failed.

Business use

Probability samples help businesses estimate customer experiences, inventory characteristics, and operational performance without measuring every unit. Their value depends on planning selection and analysis together. Budget for follow-up, maintain a usable frame, and publish the population definition, response information, weighting approach, and uncertainty. Use oversampling deliberately when smaller groups need reliable attention, then account for it in overall estimates.

Check your understanding

A group has inclusion probability 0.25. What is its basic inverse-probability weight?

Answer: The weight is 1/0.25 = 4, before any additional adjustments. It reflects the selection design, not a guarantee that each respondent perfectly represents three specific nonrespondents.

Related terms

Stratified sampling
Cluster Sampling
Nonresponse
Sampling error
Sampling frame
simple random sample
inclusion probability
sampling weight

Learn next

Study simple random sampling first, then compare stratified and cluster designs. Learn why weighting and variance estimation must follow the actual selection process.

Sources

Australian Bureau of Statistics, Statistical terms and concepts glossary, Random (probability) sample: https://www.abs.gov.au/statistics/understanding-statistics/statistical-terms-and-concepts/statistical-terms-and-concepts-glossary

The official definition supplies the core selection-probability requirement. The numerical design and fictional case are original teaching illustrations.

Previous: Random policy Next: Random projection