03 · What You Need to Know
How Criterion Sampling Works and When It Strengthens Qualitative Research
What Exactly Is Criterion Sampling?
Criterion sampling is a purposive strategy in which researchers select cases that satisfy predefined conditions relevant to the research question.
The criteria may concern a particular experience, event, role, behavior, institutional condition, or other defining characteristic.
For example, a study examining the experiences of students who received inaccurate AI-generated academic feedback might recruit only students who have encountered that situation within a specified period.
The purpose is to ensure that selected cases have a meaningful connection to the phenomenon.
Criterion sampling is therefore more specific than broadly recruiting people who appear interested in the research topic.
However, a criterion should not be chosen merely because it makes recruitment easier. Its relevance must follow from the intended inquiry.
When Is Criterion Sampling Particularly Useful?
Criterion sampling is useful when a phenomenon is defined by a particular experience or condition.
Examples include teachers who discontinued an educational technology, students who experienced a specific assessment procedure, administrators who participated in policy development, or institutions that implemented a defined intervention.
In each case, the criterion establishes that the case belongs to the phenomenon being studied.
For example, a study of why teachers abandon AI-supported assessment requires participants who have actually discontinued such use.
Recruiting teachers who merely hold opinions about AI adoption would answer a different question.
Criterion sampling is most valuable when this distinction between experiencing the phenomenon and merely having views about it is central to the research.
How Is Criterion Sampling Different from General Purposive Sampling?
Criterion sampling is one form of purposive sampling in qualitative research.
Purposive sampling is the broader approach of deliberately selecting cases according to their analytical relevance.
Criterion sampling specifically requires cases to satisfy a defined condition.
Purposive sampling
A broad selection approach based on relevance to the research question.
Criterion sampling
A purposive strategy requiring selected cases to satisfy one or more specified conditions central to the inquiry.
For example, a researcher may purposively select institutions with contrasting AI governance arrangements. Within each institution, the researcher may use criterion sampling to recruit teachers who participated in implementing those arrangements.
The two strategies can therefore operate at different levels of the same research design.
How Do You Develop a Defensible Sampling Criterion?
A defensible criterion should be derived from the phenomenon and the research question.
Suppose the research asks how instructors experience the discontinuation of AI-generated feedback.
The essential criterion might be that participants previously used AI-generated feedback in an actual teaching or assessment activity and subsequently stopped using it.
Additional criteria could specify the relevant academic period or institutional context.
However, the researcher should avoid adding conditions that are not necessary to the inquiry.
For example, requiring ten years of teaching experience may exclude relevant instructors without improving the study's ability to understand discontinuation.
The criterion should be sufficiently specific to establish relevance but not unnecessarily restrictive.
What Is the Difference Between Essential and Additional Criteria?
Essential criteria define whether a case belongs to the phenomenon being investigated.
Additional criteria may help researchers obtain particular contexts or variation among eligible cases.
| Criterion |
Example |
Role in selection |
| Essential experience |
Previously implemented AI-generated feedback and subsequently discontinued it |
Establishes eligibility for a discontinuation study |
| Temporal boundary |
Discontinuation occurred within the previous two academic years |
Defines the study period |
| Institutional context |
Employed at a participating higher education institution |
Defines the accessible setting |
| Variation characteristic |
Different disciplines or institutional sectors |
Supports analysis of contextual differences |
| Analytical contrast |
Different stated reasons for discontinuation |
May help examine contrasting experiences after eligibility is established |
Researchers should distinguish conditions necessary to define the phenomenon from characteristics included to support additional analytical comparisons.
How Specific Should the Criterion Be?
Criteria should be operationally clear enough that eligible and ineligible cases can be distinguished consistently.
For example, the criterion "experienced problems with AI" is ambiguous. Participants may interpret problems as technical failures, dissatisfaction, ethical concerns, or institutional restrictions.
A more specific criterion might require participants to have discontinued an AI feedback system after using it in at least one actual assessment activity.
However, excessive specificity can exclude relevant experiences.
For example, requiring discontinuation after exactly three weeks may have little methodological value unless that period is central to the research question.
Criteria should balance conceptual clarity with sufficient openness to capture meaningful variation within the phenomenon.
How Should Researchers Verify That Participants Meet the Criterion?
Verification depends on the type of criterion and the ethical conditions of the study.
Some criteria may be established through a brief screening questionnaire or participant self-report.
Others may require institutional records, public documentation, or confirmation of a professional role.
For example, a study of formal AI policy committee members might use authorized committee records to verify participation.
A study of personal experiences with AI-generated feedback may appropriately rely on participants' accounts, particularly when the experience is not formally documented.
Researchers should avoid collecting unnecessary personal records merely to establish eligibility.
The verification procedure should be proportionate to the criterion and consistent with informed consent and privacy requirements.
Can Criterion Sampling Be Combined with Maximum Variation Sampling?
Yes. Criterion sampling can establish who is eligible, while maximum variation sampling determines which eligible cases are selected to examine contextual differences.
For example, all participants may need to have discontinued AI-generated feedback, but the researcher may deliberately recruit across different academic disciplines and institutional sectors.
This approach ensures a shared defining experience while allowing examination of contrasting circumstances.
Alternatively, researchers may use homogeneous sampling within the eligible population to investigate a narrower subgroup.
The choice between maximum variation and homogeneous sampling should depend on whether the analysis requires contrasts or focused depth.
Criterion sampling and variation strategies are therefore complementary rather than mutually exclusive.
Can Criterion Sampling Be Used with Snowball Recruitment?
Yes. Referrals may help identify people who satisfy a relatively uncommon criterion.
For example, teachers who discontinued an AI platform may not appear in institutional lists of current users.
Existing participants may know colleagues who had similar experiences.
However, referred candidates should still be screened independently against the eligibility criteria.
Referral networks may also favor socially connected participants or people with similar experiences.
The limitations of snowball sampling remain relevant when referrals are used to identify criterion-eligible cases.
Does Meeting a Criterion Guarantee Information Richness?
No. A participant may satisfy the eligibility criterion but have limited ability or opportunity to describe the phenomenon in the depth required by the study.
For example, two teachers may both have discontinued AI feedback. One may have extensive experience evaluating implementation problems, while the other may have used the system only briefly.
Both may be eligible, but their contributions to particular analytical questions may differ.
Researchers should distinguish eligibility from the broader assessment of information richness.
The process of selecting information-rich participants may require additional attention to experience, contextual relevance, and analytical contribution.
However, researchers should not exclude less articulate or less experienced participants automatically when their accounts may reveal important aspects of the phenomenon.
How Many Criterion-Eligible Participants Are Needed?
There is no universal sample size for criterion sampling.
The appropriate number depends on the study aim, specificity of the criterion, variation among eligible cases, analytical approach, and quality of evidence.
A narrowly focused study of one shared experience may require a different sample from a study examining the same experience across several institutional contexts.
Malterud, Siersma, and Guassora's information power framework provides one way to consider qualitative sample adequacy in relation to the study aim and sample specificity.
Some qualitative approaches may also consider saturation, but the meaning and relevance of saturation differ across methodological traditions.
Researchers should not assume that meeting a numerical recruitment target automatically establishes analytical adequacy.
Can Criterion Sampling Change During Data Collection?
Eligibility criteria should ordinarily be established before recruitment to ensure consistency and transparency.
However, qualitative inquiry may reveal that the original definition does not adequately capture the phenomenon.
For example, early interviews may suggest that teachers who temporarily suspended AI use have experiences relevant to a study initially restricted to permanent discontinuation.
The researcher may consider revising the criterion if doing so is consistent with the research question, methodological approach, and ethics approval.
Such changes should be documented and their implications for the sample and analysis explained.
Researchers should not revise criteria merely to include convenient participants or exclude cases that contradict preferred findings.
What Are the Risks of Overly Restrictive Criteria?
Overly restrictive criteria can exclude cases that would contribute important evidence.
For example, requiring participants to have used AI feedback for at least one full academic year may exclude teachers whose early discontinuation was caused by severe implementation difficulties.
If the study seeks to understand discontinuation generally, excluding early discontinuation cases could distort the phenomenon being examined.
Similarly, requiring participants to work in one institutional sector may be appropriate for a bounded study but inadequate for a question about cross-sector differences.
Researchers should evaluate whether each restriction is necessary to answer the question or merely reflects recruitment convenience.
What Are the Risks of Criteria That Are Too Broad?
Criteria that are too broad may produce a sample containing cases with substantially different relationships to the phenomenon.
For example, including anyone who has heard of generative AI would provide little assurance that participants can describe experiences of implementing AI-supported assessment.
Broad criteria can make it difficult to distinguish direct experience from general opinion.
Researchers should define the phenomenon carefully and establish criteria that identify cases capable of contributing relevant evidence.
However, the criteria should still permit variation that is important to the analysis.
Does Criterion Sampling Support Statistical Generalization?
Ordinary criterion sampling is non-probabilistic and does not establish known population inclusion probabilities.
Consequently, the frequency of a theme or characteristic among selected participants should not automatically be interpreted as its prevalence among all eligible population members.
For example, if 15 of 20 criterion-selected teachers report workload concerns, that observation does not establish that 75% of all teachers who discontinued AI feedback experienced those concerns.
The sample may support detailed interpretation of why workload concerns arise and how they influence discontinuation.
Broader theoretical or contextual interpretation requires appropriate analytical reasoning and evidence rather than assuming statistical representativeness.
How Should Criterion Sampling Be Reported?
The methodology should specify the defining criterion, its relationship to the research question, and the procedure used to verify eligibility.
Researchers should also describe recruitment sources, relevant exclusions, and any additional selection strategies used among eligible cases.
If criteria changed during data collection, the reasons and implications should be reported.
Sample adequacy should be justified according to the qualitative approach.
Reporting frameworks such as COREQ and the Standards for Reporting Qualitative Research can help researchers describe participant selection transparently.
The strongest justification explains why the selected cases belong to the phenomenon and how their inclusion supports the intended analysis.
