Content Analysis

Krippendorff (2004, p. 18) defined content analysis as ’a research technique for making replicable and valid inferences from texts (or other meaningful matter) to the contexts of their use’. The goal is to link the results to their context or to the environment in which they were produced (Bengtsson, 2016, p. 9).

Тhe purpose of content analysis is to organize and elicit meaning from the data collected, and draw realistic conclusions from it. In a qualitative content analysis, data are presented in words and themes, which makes it possible to draw some interpretation of the results. The researcher must choose whether the analysis should be of a broad surface structure (a manifest analysis) or of a deep structure (a latent analysis). In a manifest analysis, the researcher describes what the informants actually say, stays very close to the text, uses the words themselves, and describes the visible and obvious in the text. In contrast, a latent analysis is extended to an interpretive level, in which the researcher seeks to find the underlying meaning of the text: what the text is talking about (Berg, 2001; Catanzaro, 1988).          

Content analysis comprises four main stages: decontextualisation, recontextualisation, categorisation, and compilation (Figure 2). However, each stage has to be performed several times in order to maintain the quality and trustworthiness of the analysis. It is the researcher’s responsibility to maintain the quality of the process by assuring validity and reliability throughout the entire study, as the results must be as rigorous and trustworthy as possible. In a qualitative study, validity means that the results truthfully reflect the phenomena studied, and reliability requires that the same results be obtained if the study were replicated (Morse & Richards, 2002).

The researchers must familiarize themselves with the data, and have to read through the transcribed text to obtain the sense of the whole, i.e. to find out what it is about, before it can be broken down into smaller meaning units. A meaning unit is the smallest unit that contains some of the insights the researcher needs, and it is the constellation of sentences or paragraphs containing aspects related to each other, and answering the question set out in the aim (Catanzaro, 1988). Each identified meaning unit is labeled with a code, which should be understood in relation to the context. This procedure is known as the ‘open coding process’ in the literature (Berg, 2001). During the analysis, codes facilitate the identification of concepts around which the data can be assembled into blocks and patterns (Catanzaro, 1988). The researcher should use a coding list, including the explanations of the codes, to minimize the cognitive change during the process of analysis in order to secure reliability (Catanzaro, 1988). Codes can be generated inductively or deductively, depending on the study design. If the study has a deductive reasoning design, the researcher has to create the coding list before starting the analysis. Otherwise, the list can be created in the course of the process (Catanzaro, 1988). Codes created inductively may change as the study progresses, as more data become available. The interpretations of the meaning units that seemed clear at the beginning may become obscured during the process. Therefore, the coding process should be performed repeatedly, starting on different pages of the text each time to increase the stability and reliability (Downe-Wambolt, 1992). However, it is much easier to obtain high reliability with code lists generated deductively rather than inductively (Catanzaro, 1988). There are also computer programmes which can be of help. Though their use is not imperative, they may facilitate the process. Although these programmes do not analyse the data, they do speed up the process by locating codes, and grouping data together in categories. Nevertheless, it is up to the researcher to decide what constitutes the themes, and what conclusions can be drawn from the results.

After the meaning units have been identified, the researcher has to check whether all the aspects of the content have been covered with regard to the aim (Burnard, 1991). The original text is re-read alongside the final list of meaning units. Coloured pencils are useful to distinguish each meaning unit in the original transcript. After this process has been performed, some unmarked text nearly always remains. The researcher must then consider whether or not the unmarked text should be included. If the unmarked text gives some answers to the research question, it should be included in the analysis (Burnard, 1995). When the researcher is deeply involved with the data, everything seems to be of importance. Nevertheless, the process of distancing is necessary, and the researcher must allow themselves to let go of the unimportant information that does not correspond to the aim of the study.

Before the researcher can begin to create categories, extended meaning units have to be condensed, which means that the number of words is reduced without losing the content of the unit (Graneheim & Lundman, 2004). The depth of the meaning units determines the level at which the analysis can be performed. This process of condensation is often needed when data are based on interviews, and when the latent content analysis is to be carried out. To extract the sense of the data, the coded material can be divided into domains – broad groups based on different focuses of the study. Graneheim and Lundman (2004) prefer the concept content area, since, in their view, this elucidates a specific, explicit area. For example, the material can be divided on the basis of the questions used when the data were collected or on theoretical assumptions from the literature (Bengtsson, 2016, p. 12).

In the categorization process, themes and categories are identified. However, in the literature there is no consensus for which headings or concepts are to be used in content analysis. Sub-categories, which Burnard (1991) terms sub-headings, are the smallest units based on meaning units. In the manifest analysis, sometimes these are the same as the codes of the meaning units. Sub-categories can be sorted into broader categories. The concept sub-theme can be used in the latent analysis instead of the concept categories. Identified themes and categories should be internally homogeneous and externally heterogeneous, which means that no data should fall between two groups nor fit into more than one group (Krippendorff, 2004). A theme is an overall concept of the underlying meaning on an interpretative latent level, and it answers the question ‘How?’

All categories must be rooted in the data from which they arise. Moving meaning units back and forth between categories ensures the progressive development of the category outcome. Initially, several categories are often generated, but the number is later reduced (Burnard, 1991). How the researcher knows when the categorization is good enough depends on the aim of the study, and the categorization is finished when a reasonable explanation has been reached (Bengtsson, 2016, p. 12).

Once the categories are established, the analysis and the writing process begin. One difference between the various qualitative analysing methods is how the researcher relates to the analysing process itself, and adapts to the results. When performing qualitative content analysis, the investigator must consider the data collected from a neutral perspective, and consider their objectivity. However, the researcher has a choice between the manifest and latent level, and the depth of the analysis will depend on how the data are collected. In a manifest analysis, the researcher works this way gradually through each identified category, and in a latent analysis through the themes. In a manifest analysis, the researcher often uses the informants’ words, and they remain aware of the need to refer back to the original text. In this way, it is possible to stay closer to the original meanings and contexts (Burnard, 1991). In contrast, a latent analysis invites the researcher to immerse themselves to some extent in the data in order to identify hidden meanings in the text. For each category or theme, the researcher chooses appropriate meaning units presented in the running text as quotations. Regardless of the form of the analysis, the researcher can present a summary of themes, categories/sub-themes and sub-categories/sub-headings as a table to allow the reader to get a quick overview of the results. In addition, it is appropriate to present one example of the analysis process. There is also the possibility to add information by performing some quantification in which sub-categories and categories are counted. This is not normally done in other qualitative research methods. However, nearly everything can be counted in written messages – such as words, characters, paragraphs and concepts – depending on the focus of the study. By combining the quantification with a qualitative approach, the magnitude of the individual phenomena studied appears more clearly (Berg, 2001). However, the variables cannot be ranked, since not all informants have had the opportunity to discuss all the phenomena that the researcher finally counts.

Finally, the researcher must consider how the new findings correspond to the literature and whether or not the result is reasonable and logical (Burnard, 1991; Morse and Richards, 2002). To validate the outcome and to strengthen the validity of the study, the researcher can perform a respondent validation, a member check, which means that the researcher goes back to the informants and presents the results in order to achieve agreement (Burnard, 1991; Catanzaro, 1988). However, there is a time-delay between the data collection and analysis. This approach, therefore, constitutes a risk for various reasons, one of which might be the possible unreliability of the informants’ memory. Another risk is that informants have a tendency to deny less attractive aspects of their behavior. In addition, as the researcher often creates a deeper holistic understanding of the studied phenomenon, the informants may not recognise how the data is presented. Keeping this in mind, it is better for the researcher to obtain some confirmation on the content from the informants in connection with the data collection (Catanzaro, 1988). Another way to increase validity is for a colleague not involved in the study, or an inquiry auditor, to read the original text and results and then judge whether they are reasonable or not (Burnard, 1991; Catanzaro, 1988). However, it is obviously difficult for an independent person to familiarise themselves with another person’s coding (Bengtsson, 2016, p. 13).

Bengtsson, M. (2016). How to plan and perform a qualitative study using content analysis. Nursing Plus Open, 2,  8–14.

Berg, B. L. (2001). Qualitative research, message for the social sciences (4th ed). Allin and Bacon, Boston, 15–35.

Burnard, P. (1995). Learning human skills. An experiential and reflective guide for nurses (3rd ed.). Butterworth-Heineman, Oxford.

Catanzaro, M. (1988). Using qualitative analytical techniques. Nursing Research: Theory and Practice. 437–456.

Creswell, J. W. (2009). Research design qualitative, quantitative, and mixed methods approaches. Sage.

Downe-Wamboldt, B. (1992). Content analysis: Method, applications, and issues. Downe, 13, 313–321.

Graneheim, U. H. & Lundman, B. (2004). Qualitative content analysis in nursing research: Concepts, procedures and measures to achieve trustworthiness. Nurse Education Today, 24, 105–112.

Krippendorff, K. (2004). Content analysis an introduction to its methodology (2nd ed.). Sage.

Morse, J. M. &  Richards, L. (2002). Readme first for a user's guide to qualitative methods. Sage.