Quantitative content analysis: units, coding and counts

Quantitative content analysis turns a defined set of texts, recordings or images into coded observations. It can show how frequently a topic appears, which sources are quoted or whether a feature is more common in one group of documents than another. The numbers describe the material you selected and the coding choices you made. They do not reveal the motives of the people who produced the material or the effects it had on audiences unless the study collects evidence for those claims separately.

Suppose a community organization wants to know whether two local newsletters explain why planned bus routes are changing. One newsletter might mention a route change often but rarely explain the reason. The other might publish fewer notices yet explain most of them. The study needs both a way to count relevant items and a definition of what counts as an explanation.

Define the unit of analysis and the material it represents

Choose the unit of analysis before opening the first issue. It could be the issue, the article, an individual quotation or a sentence. If the question is how many notices give a reason, the notice makes sense as the counting unit. If it is whose voices receive the most space, a quotation or quoted speaker may be better. Decide which surrounding material coders may consult. A notice can be the counted unit even though the whole issue is available as context for deciding which route a shortened name refers to.

Define the population in terms that another reader could reconstruct: the two newsletters, the calendar period, the edition or archive, and the kinds of item included. Record missing issues and revisions. Treating a revised notice as an independent second notice can inflate the frequency of a topic if one publication updates its pages more often. If an archive has a gap, mark it as unavailable instead of counting the missing issue as a zero.

A small study can code all eligible notices. If that is impractical, explain how you chose a sample and what it represents. A comparison based on issues from a single busy month may be useful, but it says little about quieter months. The guide to sampling methods explains the assumptions behind common designs. Include exclusion rules and a log of borderline items so the sample can be checked later.

Build a coding frame that handles borderline cases

A coding frame translates a research question into categories with decision rules. For the newsletter study, a field called “reason supplied” could have three values: yes, no, or unclear. A reason is a stated cause or purpose for changing the service. “The road will close for resurfacing, so buses will divert” qualifies. “A diversion starts Monday” does not. If the item points readers to an explanation elsewhere, decide in advance whether the study counts explanations inside the notice only or also counts linked material.

Distinguish mutually exclusive categories from features that can co-occur. A single item may quote both a transport official and a rider, so source types can be separate yes/no fields. By contrast, the primary explanation code should normally have one value per unit. Define how to code content that is absent, unclear, or unavailable, and distinguish all three from a skipped data entry. Give coders examples for common and disputed cases, especially where the distinction changes the interpretation.

Pilot quantitative content analysis before counting the full sample

Ask two coders to apply the draft rules independently to the same small, varied set. Compare decisions field by field. If one coder counts implied reasons and the other requires an explicit causal statement, the disagreement is a finding about the instructions. Rewrite the rule and pilot it on a fresh set. Save a version of the codebook and record whether earlier records were recoded under the changed rule.

Report intercoder reliability before discussion resolves disputed cases. Percent agreement is understandable, but a high value can hide poor detection of rare categories. If nearly all notices omit rider quotations, two coders can agree most of the time by seldom choosing “yes.” Examine the disagreements, the distribution of the codes and, where appropriate, a chance-adjusted statistic suited to the data and coding design. A reliable code still needs to measure the intended concept; agreement about headline length cannot tell you whether a notice was clear.

Interpret counts with the right denominator

Imagine newsletter A gives a reason in 12 of 30 eligible notices and newsletter B in 9 of 15. A publishes more notices with reasons in absolute numbers. B supplies reasons in a larger share: 60 percent versus 40 percent. Report the count and denominator along with the percentage so a reader can see both facts. If a few issues contribute most of the notices, mention that clustering rather than implying that all entries supply independent evidence.

Describe the scope of any comparison. Publication formats may differ, or one outlet may cover a period with more planned work and fewer emergency changes. These are plausible explanations for the difference, not results from the coding exercise. Use short anonymized or illustrative excerpts where they help readers see how the codes work. Save the coding rules, sample record and aggregated results so another reader can understand what the percentages mean without needing access to every original document.

Show how unclear codes affect the comparison. If four of newsletter A’s 30 notices were unclear and the 12 definite yeses were unchanged, the reported share would be 40 percent across all notices or about 46 percent across the 26 classifiable notices. Neither denominator is self-explanatory. Report which one answers the study question and show the unclear count separately so readers can see how much rests on the coding rule. This matters when unclear notices cluster by outlet.

Quantitative content analysis can be paired with an interpretive study that asks how a text creates meaning. Thematic analysis can help with that part of the inquiry. A sound numerical analysis of content begins with transparent decisions about the corpus and ends with a claim no larger than that evidence supports.

Ready when you are

Start your order with the essentials

Enter the topic, length, and deadline. We will carry these details into the full order form.

Secure checkout Upload instructions on the order form Support available when you need it