4.2.3 - Content analysis

What content analysis is

Content analysis makes the features of communication material available for systematic comparison.

Content analysis lets psychologists investigate messages and representations by coding what people have produced. The material can already exist, such as newspaper reports or films, or be collected for the study, such as requested diaries or interview transcripts. Collecting an interview is a self-report technique; subsequently coding its transcript can be content analysis. These descriptions refer to different stages and can both be correct.

The word systematically matters. A researcher should not simply read a few examples and write impressions. They need a clear research question, a defined sample of material, a coding system, and a consistent way to record what they find.

Coding has its own AQA data-handling lesson later in Research methods. Here, you only need enough coding knowledge to understand content analysis as a method: the researcher turns messy real-world material into organised evidence.

The core idea. Content analysis is often used to investigate representations, themes, stereotypes, language use or emotional tone. For example, a psychologist could analyse how often mental illness is shown as dangerous in films, or whether male and female athletes are described differently in sports reporting.

Recognising the method. If the researcher systematically codes communications such as diaries or adverts, it is content analysis. Directly recording classroom behaviour is observational data collection, which can also use recordings. “Live versus recorded” alone is therefore not a sufficient distinction: identify what is being coded and why.

Interpreting the evidence. Content analysis can have high ecological validity because the material often already exists in the real world. However, the researcher still makes choices about what to include and how to interpret it, so the method is not automatically objective.

Planning a content analysis

A strong content analysis begins with a focused question. Broad questions such as "Is social media negative?" are too vague. Better questions identify the material, the feature being analysed and the population of content.

After the question, the researcher must decide on a sample of material. This means selecting which articles, episodes, posts, adverts or documents will be analysed. Sampling matters because biased selection can create biased findings. If a researcher only chooses the most sensational headlines, they may exaggerate how negative the media usually is.

A practical plan usually includes:

1. Aim or question: what pattern is being investigated?
2. Sampling rule: which items are included and why?
3. Unit of analysis: what exactly will be coded, such as a word, sentence, image, scene, article or whole post?
4. Categories: what features will be recorded?
5. Coding instructions: how coders decide which category applies.
6. Recording and summary: how the data will be counted, described or interpreted.

The unit of analysis is easy to overlook. A study counting individual words may answer a different question from a study coding whole scenes. For example, counting the word "dangerous" in film reviews is narrower than coding whether the overall review links schizophrenia with violence.

Worked example. Dr Singh studies portrayals of teenagers by sampling every online front-page story mentioning teenagers from four named news sites during April. The unit is one article. She records two separate dimensions: portrayal of behaviour (responsible, irresponsible, mixed or no relevant portrayal) and whether the article describes a teenager as a victim (yes/no). Keeping these dimensions separate avoids forcing an article about a responsible victim into only one inappropriate category. This sampling frame is replicable but does not represent all news sites or all months.

Classifying content

In content analysis, the researcher turns raw communication into evidence by classifying parts of the material into categories. This is enough coding knowledge for this lesson: content must be organised systematically so patterns can be counted, compared or interpreted. A later Research methods lesson looks at coding in more depth.

Some categories record directly visible features, such as the number of times a character speaks aggressively. Other categories involve more interpretation, such as whether a scene presents a character as threatening or sympathetic. The more interpretive the category, the more carefully it needs to be defined.

Content analysis can produce quantitative data, qualitative data, or both. A study might count the number of aggressive acts in children's cartoons, giving frequencies and percentages. It might also describe repeated themes, such as aggression being rewarded or presented as humorous.

Good categories should be clear, relevant to the research question and specific enough that another researcher could understand how the material was classified. Operationalising a category means stating what evidence counts. For example, “verbal threat” could mean an explicit statement of intent to harm another character, excluding insults without a threat. If each unit is assigned one category, categories must not overlap and must cover the relevant possibilities, including mixed or unclear material. If multiple codes are allowed, state that rule instead of treating double-coding as an error.

Two coders should independently code the same pilot material using the same instructions, then compare their records. Agreement indicates consistency, not necessarily validity: both could apply the same inappropriate categories. Disagreements can reveal unclear boundaries. Clarify the definitions, train coders with examples, and check agreement again on fresh material before analysing the full sample. Comparing independent records measures inter-observer reliability. Revising unclear rules, training with examples and retesting are ways to improve it. For a simple categorical frame, one transparent check is the percentage of units assigned the same code; for suitable numerical coding totals, researchers may compare the paired totals using a correlation. The chosen measure must suit the coding records. High agreement is consistency, not proof that the categories capture the intended meaning. Simply adding a second coder or forcing agreement does not repair reliability. If categories are vague, the findings may reflect the researcher's personal interpretation rather than a real pattern in the material.

Why this matters. Content analysis can look scientific because it uses categories and counts. But if the categories are vague, such as "negative" or "stereotypical" without clear rules, the numbers may give a false sense of objectivity.