03 · What You Need to Know
Understanding Analytical, Sampling, Coding, and Context Units
What Does Unit of Analysis Mean?
The unit of analysis identifies what the study ultimately investigates and makes claims about.
For example, a study asking how individual university policies represent student responsibility may treat each policy as an analytical unit.
A study asking how universities differ in their governance arrangements may instead treat each university as the unit of analysis, even when several policies are collected from each institution.
In a study of social media communication, the analytical unit might be an individual post, discussion thread, account, or organization.
The appropriate choice follows the research question and the level at which findings will be interpreted.
Researchers should avoid identifying the unit of analysis solely from the file format or number of documents collected.
Why Is the Unit of Analysis Important?
The unit of analysis influences how evidence is organized, counted, compared, and interpreted.
For example, 20 references to AI disclosure within one university's policies do not necessarily represent 20 independent institutional commitments.
They may be repeated expressions of one institutional requirement.
If the researcher intends to compare universities, the relevant question may be whether each institution has an explicit disclosure requirement, rather than how many times the word appears.
Failure to distinguish these levels can inflate apparent sample sizes, distort descriptive statistics, and produce misleading interpretations.
Unit definition is therefore a substantive methodological decision, not merely a technical coding preference.
What Is a Sampling Unit?
A sampling unit is the entity selected through the study's sampling procedure.
For example, researchers may randomly select university policies from a defined document register.
Alternatively, they may purposively select universities and collect all relevant policies from those institutions.
In the first arrangement, documents are sampling units. In the second, institutions may be sampling units while documents are collected within selected cases.
Sampling units and analytical units can coincide, but they do not have to.
The distinction matters because the sampling procedure determines the population to which findings may reasonably apply.
What Is a Coding or Recording Unit?
A coding unit, sometimes called a recording unit, is the segment of material to which an analytical category or value is assigned.
Depending on the design, it might be a word, sentence, paragraph, policy provision, complete document, or interaction episode.
For example, researchers may code individual policy provisions according to whether they assign responsibility to students, instructors, or institutions.
Those provisions are coding units, even if the final comparison concerns universities.
Krippendorff's methodological treatment of content analysis emphasizes the importance of defining units and the procedures through which content becomes analyzable evidence.
The unit should be sufficiently clear for the intended measurement or interpretation.
What Is a Context Unit?
A context unit identifies the surrounding material needed to interpret a coding unit appropriately.
For example, a sentence stating that students are responsible for verifying AI-generated information may appear within a section describing shared responsibilities among students, teachers, and institutions.
Coding the sentence in isolation could obscure that broader arrangement.
The researcher might code the sentence while consulting the entire policy section or document as its context unit.
Context units are especially important when the meaning of a statement depends on preceding qualifications, definitions, or exceptions.
Researchers should establish how much surrounding material coders may consult and apply those rules consistently.
How Do the Different Units Compare?
| Type of unit |
What it determines |
Illustrative example |
| Unit of analysis |
Entity about which conclusions are made |
University |
| Sampling unit |
Entity selected through sampling |
University or policy document |
| Coding or recording unit |
Segment receiving an analytical classification |
Policy provision |
| Context unit |
Surrounding material used for interpretation |
Policy section or complete document |
| Unit of collection |
Material obtained and stored for analysis |
Downloaded policy file |
These distinctions are useful, but terminology can vary across methodological traditions. Researchers should define their terms explicitly rather than assume every reader uses the same classification.
Can a Sentence Be the Unit of Analysis?
Yes, when the research question and intended conclusions concern sentences or sentence-level features.
For example, a linguistic study might investigate the grammatical construction of responsibility statements across policy documents.
Each sentence could then be an analytical unit.
However, if the research concerns institutional governance, sentences are more likely to function as coding units within larger analytical cases.
The researcher should avoid confusing the smallest segment being examined with the entity to which the findings ultimately refer.
Can a Whole Document Be the Unit of Analysis?
Yes. Document-level analysis is appropriate when the research concerns characteristics of individual documents.
For example, a researcher may classify each policy according to whether it contains an explicit disclosure requirement.
Each policy receives one classification for that variable.
However, documents may differ substantially in length, scope, and institutional authority.
Researchers should consider whether those differences affect comparability.
Document-level analysis is not necessarily equivalent to institution-level analysis, particularly when organizations issue multiple related documents.
When Should the Institution Be the Unit of Analysis?
The institution is an appropriate analytical unit when the research question concerns organizational characteristics or comparisons.
For example, a study may ask what proportion of universities formally require disclosure of generative AI assistance.
If several policies from one university contain relevant provisions, the researcher must establish how those documents collectively determine the institution's classification.
A university might be classified as having an explicit disclosure requirement if at least one current, authoritative policy establishes that obligation.
However, this is an illustrative decision rule, not a universal standard.
The rule must account for authority, scope, and potentially conflicting provisions.
How Do You Handle Multiple Documents from One Institution?
Documents originating from the same institution may be related rather than independent.
For example, a university might issue a policy, an implementation memorandum, and a student-facing FAQ describing the same requirement.
Counting these as three independent institutional policies could exaggerate the apparent prevalence of the requirement.
Researchers should preserve document-level records while linking them to the relevant institutional case.
Where institution-level conclusions are intended, an explicit aggregation procedure is necessary.
The procedure for defining a document corpus should identify these relationships before analysis begins.
What Is Unitizing in Content Analysis?
Unitizing is the process of identifying the boundaries of meaningful segments within content.
For example, a researcher may divide policy text into provisions addressing separate obligations.
One provision might begin with a statement of responsibility and end before the next distinct requirement.
Different researchers may disagree about where one meaningful segment ends and another begins.
These disagreements matter because they affect the number and content of coded units.
Researchers should establish segmentation rules and, where appropriate, evaluate the consistency of unit identification separately from agreement about category assignment.
Should Coding Units Be Mutually Exclusive?
Not necessarily.
Some quantitative coding schemes require every unit to receive one category within a specified dimension.
Other approaches permit multiple categories because a passage may address several concepts simultaneously.
For example, a policy provision may assign responsibility to students while also requiring instructor oversight.
Forcing the provision into only one category could misrepresent the content.
Researchers should distinguish mutually exclusive categories from overlapping codes and establish appropriate recording rules.
The choice depends on the conceptual framework and intended analysis.
How Does Unit Size Affect the Findings?
Smaller coding units may capture detailed distinctions but can separate statements from their context.
Larger units preserve more context but may contain several distinct ideas that are difficult to classify using one value.
For example, coding entire policy sections may obscure differences among disclosure, verification, and accountability provisions.
Coding isolated words may miss qualifications that determine what those provisions mean.
Researchers should choose unit boundaries that preserve the information necessary for the research question.
There is no universally optimal unit size.
What Happens When the Unit of Analysis Is Defined Incorrectly?
Incorrect unit definition can lead to conclusions that do not follow from the evidence.
For example, a researcher might identify 300 statements about AI disclosure across 40 policy documents and conclude that disclosure is emphasized by 300 institutions.
This would confuse statements with organizations.
More subtle errors occur when researchers treat repeated passages from the same institution as statistically independent observations.
Such treatment can underestimate uncertainty when observations are correlated.
The analytical design should reflect the structure of the data and the actual level of inference.
What Is Pseudoreplication in Content Analysis?
Pseudoreplication occurs when nonindependent observations are treated as though they provide independent replication for a particular inference.
For example, 100 coded passages from five universities do not necessarily provide 100 independent observations about institutional governance.
Passages within the same university may reflect shared policy authorship, governance structures, or repeated requirements.
If the study makes statistical comparisons across institutions, the analysis should account for this dependence.
Possible approaches include aggregation to the institutional level or statistical models that accommodate clustered observations, depending on the research question and design.
Increasing the number of coded passages does not automatically increase the number of independent institutional cases.
How Do Units Affect Frequency Counts?
Frequency has meaning only when the counted entity is specified.
For example, a finding that 60% of policy documents mention verification differs from a finding that 60% of universities have an explicit verification requirement.
The first uses documents as the denominator. The second uses institutions.
Similarly, counting how many times a concept appears measures textual occurrence rather than necessarily measuring its prevalence across institutions.
Researchers should report numerators and denominators clearly and avoid using the word frequency without specifying what is being counted.
How Do Units Affect Qualitative Content Analysis?
Qualitative content analysis also requires clear decisions about analytical units, although segmentation may be more flexible in some traditions.
For example, a researcher examining how AI responsibility is constructed may analyze passages within their broader policy context.
Individual statements may be coded, while themes are developed across documents and institutional cases.
The analysis should preserve relationships among statements rather than treating every coded excerpt as a separate, self-contained account.
Researchers should explain how material was segmented and how interpretations moved from excerpts to broader findings.
How Do Units Affect Mixed-Methods Content Analysis?
Mixed-methods studies may analyze content at different levels across their quantitative and qualitative components.
For example, a quantitative component may classify universities according to policy provisions, while a qualitative component examines selected passages in depth.
These levels can be integrated, but the relationship must be explicit.
The decision about quantitative, qualitative, and mixed-methods content analysis should therefore include consideration of unit alignment.
Researchers should avoid merging results from different analytical levels without explaining how they relate.
Can the Unit of Analysis Change During a Study?
It can, but changes have methodological consequences.
For example, preliminary analysis may reveal that institutional comparisons are more meaningful than comparisons among individual policy documents.
The researcher may revise the analytical unit accordingly.
However, the revised design may require new aggregation rules, additional document retrieval, or different statistical procedures.
Changes should be justified and documented rather than introduced solely because they produce more favorable findings.
The final report should accurately describe the units actually used.
How Should the Unit of Analysis Be Reported?
The methods section should identify the sampling units, coding units, context units, and analytical units when these differ.
Researchers should explain how units were defined, how overlapping or ambiguous segments were handled, and how lower-level observations were aggregated.
For quantitative studies, relevant denominators and statistical dependence should be addressed.
For qualitative studies, the relationship between coded excerpts and broader interpretations should be made transparent.
Readers should be able to determine exactly what the reported findings describe.
