Randomization vs. Random Sampling: Why Do Researchers Keep Confusing Them?

When Researchers Say Participants Were “Randomized,” What Actually Happened?

A paper states that 100 students were “randomly selected and divided into two groups.” Another says participants were “randomly sampled into the experimental and control groups.” A third describes an experiment as randomized because respondents were selected using probability sampling.

These statements may sound similar because they all contain the word “random.” Methodologically, however, they can describe very different procedures.

Random sampling concerns how units are selected from a population into a sample. Random assignment concerns how units already participating in a study are allocated to experimental conditions.

Confusing the two is not merely a vocabulary problem. It can lead researchers to claim population representativeness from random assignment or causal inference from random sampling, neither of which follows automatically.

Sampling Selects the Participants; Assignment Allocates the Conditions

The Same Word “Random” Is Doing Two Different Jobs

The confusion is understandable. Both procedures deliberately use chance rather than researcher discretion. But they occur at different stages of research and address different sources of bias.

A useful way to remember the distinction is to ask two questions:

• Who gets into the study?
• Once they are in the study, what condition do they receive?

The first is a sampling question. The second is an assignment question.

Random sampling is about selecting units from a population

Suppose your target population is all 10,000 undergraduate students enrolled at a university. You want a sample of 500 students.

A probability sampling design specifies a mechanism through which units have known probabilities of selection. In simple random sampling, for example, each possible sample of a specified size has an equal probability of selection, which implies equal inclusion probabilities for units under that design.

The purpose is to connect the observed sample to the population from which it was probabilistically selected.

Random sampling is therefore primarily a sampling-design issue. It concerns selection into the study rather than treatment allocation.

Random assignment happens after the study units have been recruited or selected

Now suppose those 500 students enter an experiment comparing two instructional approaches.

The researcher uses a random allocation procedure to assign participants to approach A or approach B.

That is random assignment.

The students are already part of the study. The random process now determines which experimental condition they receive.

The American Psychological Association describes random assignment as a technique used after participants have been selected, in contrast with random sampling, which determines who will be studied.

The population-to-sample-to-condition sequence makes the distinction easier

Once you visualize the process this way, “randomly sampling participants into the control group” becomes easier to diagnose. Sampling and assignment refer to different transitions.

Random sampling and random assignment support different kinds of inference

Random sampling is closely connected to inference from a sample to the population represented by the sampling design. When probability sampling is properly implemented, it provides a principled basis for estimating population quantities and quantifying sampling uncertainty.

Random assignment addresses a different problem. Because treatment allocation is determined by a random mechanism rather than participants' characteristics or researcher choice, the assigned conditions are comparable in expectation with respect to baseline characteristics. This provides a strong basis for attributing systematic outcome differences to the assigned interventions, subject to the assumptions and conduct of the experiment.

Random assignment does not make your sample randomly selected

Imagine recruiting 100 volunteers through a social-media advertisement. Every volunteer then has an appropriate random chance of being assigned to an intervention or control condition.

The experiment can be randomized even though the participants were volunteers rather than a probability sample of the wider population.

Random assignment addresses the comparability of assigned conditions. It does not change the process through which those 100 volunteers entered the study.

You should therefore be cautious about claiming that random assignment makes a study sample representative of all students, teachers, patients, employees, or other members of a target population.

Random sampling does not create a randomized experiment

Now reverse the situation.

Suppose 500 students are selected from a university using a probability sampling procedure. Researchers measure their social-media use and academic performance but assign no exposure or intervention.

The sample may support population estimates under the sampling design. The study has not become a randomized experiment.

Random sampling does not randomly distribute social-media use, prior achievement, motivation, socioeconomic characteristics, or other exposures across conditions because there are no randomly assigned conditions.

The observed relationship between social-media use and performance remains observational.

A study can use both random sampling and random assignment

Suppose a university uses probability sampling to select students from its enrollment records and then randomly assigns consenting sampled students to two instructional interventions.

Both random procedures are present.

The sampling design helps connect the participating sample to the target population, while random assignment creates the experimental comparison. The precise scope of generalization still depends on issues such as eligibility, consent, nonresponse, attrition, implementation, and the target population being claimed.

A study can also use neither

A researcher might recruit a convenience sample of students from one class and compare those who voluntarily use an educational application with those who do not.

No probability sampling occurred. No random assignment occurred.

That does not automatically make the research worthless. It simply means the study cannot claim the inferential advantages that arise specifically from those random procedures.

The design might still provide useful descriptive or associational evidence if its limitations are understood and reported accurately.

Random assignment does not guarantee identical groups

Random assignment makes the groups comparable in expectation. It does not guarantee that every characteristic will have exactly the same mean or proportion in a particular experiment.

Chance imbalance can occur, especially in smaller samples.

This connects directly to the problem of baseline differences between groups. A baseline difference after proper randomization does not automatically mean that the assignment procedure failed.

Randomization is more than casually dividing participants

Researchers sometimes write that participants were “randomly assigned” when they actually alternated participants between groups, assigned the first half to one condition and the second half to another, used birth dates, or allowed participants to choose.

Those procedures should not automatically be described as random assignment.

True random assignment requires an appropriate chance mechanism. In formal trials, additional safeguards such as allocation concealment may be needed to prevent foreknowledge of upcoming assignments from influencing enrollment or allocation.

The methodological question is therefore not whether the word “random” appears in the methods section, but whether the assignment mechanism was genuinely random and properly implemented.

Random sampling is broader than simple random sampling

Another source of confusion is treating “random sampling” as synonymous with only one probability sampling design.

Simple random sampling is one form of probability sampling. Depending on the population and research objectives, researchers may instead use stratified, cluster, multistage, or other probability sampling designs.

These designs do not all give every unit the same selection probability. What makes them probability samples is that selection follows a probability mechanism with known, nonzero selection probabilities for the relevant units under the design.

This nuance matters because the familiar phrase “everyone has an equal chance” accurately describes some random procedures but is not a universal definition of all probability sampling.

Four Studies That Use the Word “Random” Very Differently

Study B and Study C demonstrate why the two procedures cannot substitute for one another.

Study B may provide a strong basis for estimating the effect of assigned treatment within the experimental context even though its volunteers were not randomly sampled from all university students. Study C may support population-oriented estimates under its sampling design, but merely observing naturally occurring exposure does not give the exposure comparison the causal protection of random assignment.

The word “random” is therefore incomplete unless you specify what was randomized.

Common Ways Random Sampling and Random Assignment Get Mixed Up

Always Specify What Was Randomized

The easiest way to prevent confusion in your own research is to stop writing “participants were randomized” without an object.

Were participants randomly selected from a population? Were enrolled participants randomly assigned to conditions? Did both happen?

Say exactly which procedure occurred.

If your primary concern is what treatment allocation accomplishes after participants enter the study, the next issue is what random assignment actually does and what it does not do. Keeping that separate from sampling prevents a surprising number of methods-section headaches.

Before Using the Word “Random” in Your Methods Section

Frequently Asked Questions About Random Sampling and Random Assignment

Ask Whether Chance Selected the Sample or Assigned the Condition

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.