Summative Content Analysis: from word counts to meaning

Stage one — manifest quantification. Identify the target terms (including synonyms, inflections, and euphemisms decided in advance or discovered iteratively), count occurrences, and map their distribution: by document type, speaker role, time period, section. The claims here are descriptive and checkable: who says this word, where, how often.

Stage two — latent interpretation. Examine occurrences in their contexts — the keyword-in-context discipline — to characterise usage: senses, functions (assertion, hedge, irony, quotation), collocations, and the situations that trigger substitution or silence. The claims here are interpretive: what the pattern of use means.

The link between stages is the method’s spine. Every interpretive claim should be traceable to counted, locatable instances; every count should eventually be read. Studies that decouple them fail in opposite directions — numerology or impressionism.

The approach is the third of Hsieh and Shannon’s Three approaches to qualitative content analysis: where conventional analysis derives categories from data and directed analysis from theory, summative analysis starts from manifest terms and proceeds to their latent use.

Its intellectual kin are corpus linguistics (concordances, collocation) and the manifest/latent vocabulary of the content-analysis tradition; what the summative approach adds is a qualitative research frame — purposive corpora, contextual interpretation, and trustworthiness standards rather than purely statistical ones.

It is also the approach most often performed unknowingly: any study that searches transcripts for mentions of X and then discusses “how X came up” is doing informal summative analysis, usually without the discipline either stage deserves.

• Terminology as tracer: how “burnout” migrated from private talk to board papers; whether “consent” appears in training materials and how it is glossed.
• Comparative usage: the same term across outlets, professions, or periods — where distribution differences are themselves findings.
• Euphemism and avoidance: what replaces the direct term, and in which contexts the direct term becomes sayable.
• Policy–practice language gaps: terms mandated in documents versus their presence and sense in recorded practice.
• Poor fits: phenomena not tied to identifiable surface forms (experiences, reasoning), and any question where paraphrase variety would make term lists arbitrary — those need conventional or thematic analysis.

A safeguarding study examined how physical restraint was documented across 12 care homes: 3,400 incident reports, care plans, and policy documents. The term set began with “restraint/restrain” and grew, via stage-two reading, to include “guided”, “redirected”, “supported to the floor”, and “comfort hold” — each addition logged.

Stage one showed the headline distribution: “restraint” was frequent in policies, rare in incident reports — while the euphemism set ran the other way, and one home accounted for a third of all “comfort hold” usage. Stage two read every incident-report occurrence: “restraint” appeared almost exclusively in negations (“no restraint was used”), while the euphemisms carried the descriptive load for physically identical acts, as revealed by the surrounding narrative detail.

The joint finding — the mandated term functioned as a category to be avoided rather than applied, with home-level euphemism dialects — was checkable at every step: term set published, counts by home tabled, and each claimed sense evidenced with quoted contexts. The regulator’s response targeted documentation training, and the follow-up audit reused the same instrument, which is the reproducibility this design buys.

• Stopping at stage one. Frequency tables presented as findings — the concordance fallacy.
• Unstated term sets. Counts of an instrument nobody can inspect; synonym choices silently drive results.
• Ignoring negation and quotation. “No restraint was used” counts as an occurrence only if the analysis knows what it is doing with polarity and voice.
• Interpreting unread counts. Assigning meaning to distributions whose instances were never examined in context.
• Corpus convenience. Counting whatever text was easy to get, then generalising to the practice the texts unevenly record.
• Treating usage change as attitude change. Language shifts under mandates and fashions; the inference from words to beliefs needs argument, not assertion.

The method is chained to surface forms: phenomena expressed without stable vocabulary escape it, and speakers’ paraphrase creativity is a permanent leak in any term set. Iterative expansion narrows the leak; nothing seals it.

Its corpora are records, with all the selection and production biases records carry — what gets written, by whom, under what incentives, is upstream of every count.

And the design’s comparative power depends on comparable units: counting across documents of wildly different lengths, genres, and authorship conventions requires normalisation choices that are themselves contestable and must be reported.

Conventional qualitative content analysis derives categories from the data when no term set can carry the question.

Directed content analysis codes against theory-derived categories rather than surface terms.

Critical discourse analysis takes the interpretation of usage into ideology and power with a declared standpoint.

Document and textual analysis treats the records themselves — their production and function — as the object.

Both stages are tool-shaped. Stage one — term-set extraction with context, across thousands of documents, tabulated by source and period — is mechanical and should be automated; Evidano’s frequency and content analysis does the counting with every instance linked to its passage. Stage two’s occurrence coding (sense, function, polarity) is a small codebook applied at scale, the same auditable-coding workflow, leaving the analyst a complete keyword-in-context evidence base instead of a sampling hope.

The interpretation of the joint pattern — what the distribution of use means — is stage two’s human remainder, and the write-up’s worth stands or falls on it.