13 Research methods, statistics, critical review and evidence-based practice13.7 Qualitative research
13.7.4.10 Content analysis and constant comparison
3 min read
Written by Resident Doctor Committee · Published
Definitions
Content analysis organises and interprets text systematically.
Constant comparison sets incidents, codes and categories against one another as analysis develops; it is central to grounded theory and can inform other approaches.
Preparing and reading the data
Transcription detail should match the method: pauses, emphasis and overlap may matter in interactional analysis, while a readable transcript may suit applied work.
Check transcripts against recordings, and keep field notes that separate observed events from first interpretations.
Read accounts in context before breaking them into codes; familiarisation and coding may run iteratively.
Coding for meaning
A meaning unit may be a phrase or a longer passage, taken with enough context to interpret it, and may carry several codes.
Shared meaning need not share words: ignoring phone calls and leaving letters unopened may express one concern.
Manifest coding stays close to what was said, whereas latent analysis interprets the meaning underneath.
Feature | Qualitative content analysis | Quantitative content analysis |
|---|---|---|
Categories | Inductive, deductive or combined | Defined consistently enough for systematic counting |
Output | Meanings, patterns, variation and context | Frequencies and relationships across texts |
Typical question | What accounts mean in context | How media depictions of schizophrenia changed over time |
A count such as "18 of 24 participants mentioned stigma" describes the sample, not population prevalence.
Not mentioning something may reflect the questions asked or what someone chose to disclose, rather than absence of the experience.
Where codes come from
Inductive coding grows out of the data; deductive coding applies an existing framework. Neither is free of assumptions, and many studies combine them.
Approach | Gains | Costs |
|---|---|---|
Framework (deductive) | Organised comparison, links to existing theory | May miss material outside the initial categories |
Inductive | Scope for unexpected meanings | Still shaped by researchers; may miss useful existing concepts |
Framework analysis (Ritchie and Spencer, applied policy research) runs through familiarisation, a thematic framework, indexing, charting, then mapping and interpretation.
Its matrix puts cases in rows and categories in columns; an empty cell prompts a question rather than showing absence.
Constant comparison at three levels
Level | Example: disengagement from a community team |
|---|---|
Incident with incident | Missed appointments reflect travel and childcare costs for one person, avoiding being seen for another |
Code with code | Unanswered calls and unopened letters may be about control over contact, not rejection of all help |
Category with category | Service-controlled contact set against fear of being labelled non-compliant suggests limiting contact brings more assertive follow-up |
That suggested process is an interpretation to investigate, including accounts that challenge it.
Memos and theoretical sampling
Memos record why a code matters, how it relates to others, what contradicts it and what data would clarify it; dates and source links support the audit trail.
Coding software alone does not do the analysis.
Theoretical sampling in grounded theory follows emerging questions, such as seeking people comfortable with depot appointments to test an explanation built on control.
Negative cases
Actively seek cases that do not fit: if some detained patients describe relief, explore their circumstances rather than excluding them as outliers.
Revise claims to reflect variation without imposing a new universal rule, since coercion and relief can coexist.
Confirming and challenging cases are not automatically worth more than each other.
Two coders
Kappa suits structured coding treated as measurement; interpretative approaches use discussion to explore different readings.
Reflexive thematic analysis does not treat independent consensus coding as a quality standard.
Disagreement may reveal an unclear definition, different contextual knowledge or a passage with several meanings.
Two coders or a high agreement coefficient do not establish rigour on their own.
Reporting
Quotations should support the interpretation, carry variation and preserve context; one account cannot support an unqualified claim about everyone.
Removing names does not guarantee anonymity, because a ward, occupation, date or unusual experience may identify someone.
Seek publication consent where identification remains possible, and never promise anonymity you cannot deliver.