Directed qualitative content analysis with your codebook

Scope: interview and focus group data analysed with a theory-based codebook. All example data are invented.

Short answer: directed content analysis is an approach to qualitative content analysis in which categories derived from an existing theory are applied to the data, and text that does not fit is analysed in order to validate or extend the theory (Hsieh & Shannon, 2005). It is often called deductive content analysis. The steps are:

1. State the theory, the questions and your expectations.
2. Build the codebook, or assess the one you were given.
3. Define units and segment the material.
4. Read the whole material and highlight relevant text.
5. Pilot the codebook and write decision rules.
6. Code all the material.
7. Analyse the residual, overlaps and empty categories.
8. Compare by category and by case.
9. Interpret against the theory and report.

One decision comes before coding: whether your codebook stays fixed or may grow with new categories from the data.

Three approaches to qualitative content analysis

Hsieh and Shannon (2005) distinguish three approaches by where the categories come from:

Their directed approach shares a theory-led starting point with Mayring's deductive category application, in which each category has an explicit definition, examples and coding rules collected in a coding agenda (Mayring, 2000, 2014). The two traditions differ in how they handle uncoded material and the development of new categories. Assarroudi et al. (2018) set out directed content analysis as a 16-step method; the nine steps here cover the same ground in the order an applied project meets them.

Fixed or growing codebook: decide before you code

Within deductive content analysis, Elo and Kyngäs (2008) describe two kinds of categorisation matrix. With a structured matrix, only aspects that fit the matrix are taken from the data. With an unconstrained matrix, new categories can be created within its bounds, following inductive principles. Hsieh and Shannon (2005) expect uncoded text to be analysed later to see whether it represents a new category or a subcategory, so their directed approach sits closer to the unconstrained variant.

In practice it helps to name one of four codebook modes in advance:

Do not mix modes quietly. If you change or add codes during coding, you are no longer applying the original codebook; the change needs a date, a reason and recoding of material already coded. Choose a fixed codebook for comparability with studies that used the same instrument, and a growing one to extend the theory.

Steps, with an invented example

Invented example. All names and data are invented for illustration. Rovina Regional Hospital pilots twelve-hour shifts and a shared team of nurses who move between wards. Two focus groups: FG-A, eight ward nurses (N1–N8), and FG-B, six ward managers (W1–W6). The codebook comes from the Job Demands–Resources model (JD-R). In its original form, job demands are linked to exhaustion and a lack of job resources to disengagement (Demerouti et al., 2001); later work adds a motivational process through engagement and the proposition that resources buffer the impact of demands (Bakker & Demerouti, 2007).

[A-13] N3: On the noticeboard. One day it was just up there. Nobody asked us,
           and then they were surprised that we swap shifts among ourselves.
[A-14] N5: It suits me, actually. I have a small child, and twelve-hour shifts
           mean more days off. If only we knew more than a week ahead.
[A-16] N3: […] Someone from the shared team comes in, doesn't know the
           patients, and I have to show her everything anyway. So I'm doing
           two jobs.
[A-17] N6: Mhm. That's exactly it.
[B-32] W4: It's gone down on the spreadsheet. When they swap a shift between
           themselves, nobody records it as overtime.

In B-32, "it" is the overtime figure that W2 mentioned in the turn before.

1. State the theory, the questions and your expectations

Turn the theory into questions and expectations: "Which demands and resources do participants link to the new rota? Do they see a loss of resources?" Choose the variant (fixed or growing codebook) and whether you code manifest content only or latent content too. Write down what you expect to find; the memo is your later check against confirmation bias.

2. Build the codebook, or assess the one you were given

Derive main categories and subcategories from the theory, and add a residual category for relevant text that fits nowhere.

D  Job demands
   D1 Time pressure and workload
   D2 Emotional demands
   D3 Unpredictability of work and rota
R  Job resources
   R1 Autonomy and influence over how work is organised
   R2 Support from colleagues
   R3 Support from supervisors
   R4 Information and feedback
E  Outcomes
   E1 Exhaustion
   E2 Engagement
X  Residual: relevant to the question, not classifiable

For each category, write the coding agenda: definition, anchor example and coding rule (Mayring, 2000). A useful codebook entry also says when to use the code and when not to, with a borderline example (MacQueen et al., 1998):

Check that subcategories within one dimension are mutually exclusive, or that a rule says when two may apply together, and that the residual category makes the frame exhaustive (Schreier, 2012). A definition that only repeats the name is not a definition. If the codebook comes from a client or the literature, assess it for missing definitions, overlaps and unclear rules, and agree with its owner whether it may be changed.

3. Define units and segment the material

Decide the unit of analysis (participant or group; here, participants are rows and the group is the level of comparison), the coding unit (the smallest piece of text that carries one meaning, usually part of a turn; Graneheim & Lundman, 2004) and the context unit (what a coder may read to understand it, such as the moderator's question and the previous turns). Moderator turns are context, not coded data, unless the research question includes moderation or interaction as an object of analysis. Uncertainty about unit boundaries is a major source of disagreement between coders (Campbell et al., 2013).

4. Read the whole material and highlight first

Read every transcript before the codebook cuts it into pieces. Hsieh and Shannon (2005) describe two strategies: highlight all text that appears to relate to the phenomenon and then code it, or code immediately with the predetermined codes. Highlighting first captures more; highlighted text that later gets no category is a candidate for the residual.

5. Pilot the codebook and write decision rules

Code a deliberately varied part of the material, including places where you expect borderline cases, and from each focus group. Keep a decision log. Two borderline cases from the example:

• A-13, "Nobody asked us." A lack of resource R1, or a demand? Rule: a lack of a resource is coded under the resource with negative polarity; a demand is only what requires effort.
• A-16, "I have to show her everything anyway." N3 is describing the colleague from the shared team. Support from colleagues (R2) or workload (D1)? Rule: code what the segment describes. This segment describes extra work, so it is D1, not R2.

Revise definitions and rules, and recode the pilot part. If the codebook may not change, refine only the decision rules and record what they cannot resolve as limits.

6. Code all the material

Apply the codebook to every coding unit: read its context, say what it means, then compare it with the definitions. Do not stretch a category to make text fit; that is what the residual is for. A new borderline case gets a new rule and a recheck of earlier decisions. In focus groups, a short agreeing turn like A-17 ("Mhm. That's exactly it.") expresses N6's alignment with N3 but is not a separate substantive account. Code it with the same category as the turn it agrees with, flagged as agreement, and do not count it as a new mention.

7. Analyse the residual, overlaps and empty categories

In directed content analysis this is where the theory is refined or extended (Hsieh & Shannon, 2005). In the example:

• A-13, "we swap shifts among ourselves": nurses re-arranging the rota themselves. The initial codebook did not capture this activity; job crafting (Tims & Bakker, 2010) is one possible interpretation to investigate, and later JD-R work incorporates that concept. With a growing codebook, it becomes a new subcategory, marked as derived from the data. With a fixed codebook, it stays in the residual with the finding "this codebook does not capture workers' own activity".
• A-14, "I have a small child, and twelve-hour shifts mean more days off": work–family relations, also outside this version of the model. Same decision.
• B-32, "It's gone down on the spreadsheet": how overtime is measured. Relevant to the hospital, not to JD-R. It stays in the residual as outside the theory's scope, but it is recorded.

For multiple-coded units, ask whether it is a real co-occurrence or an overlap between definitions; an overlap sends you back to the codebook. For each empty category, ask why: the phenomenon did not appear, the guide did not ask, or the category is badly defined. An empty category is information, not a gap to hide.

8. Compare by category and by case

Summarise each category, with supporting and non-supporting evidence for the theory's expectations (Hsieh & Shannon, 2005). Build a category-by-case matrix with short summaries and references, not just counts, and show empty cells. Look for co-occurrences (does D3 unpredictability appear with E1 exhaustion, and for whom?). Compare groups: in the full invented data, nurses talk mostly about lost resources (R1, R4) and managers about reduced demands (D1). Is that a difference of perspective or of information?

9. Interpret against the theory and report

For each expectation from step 1, conclude: supported, not supported, or supported under a condition, with evidence on both sides. Treat new categories and the residual as a contribution, for example that workers actively shape their own resources. Consider method artefacts: a category may be strong because the topic guide asked about it. Name what the deductive lens hid.

In the report, keep three voices apart: what the participant said ("It's gone down on the spreadsheet"), where the codebook put it (residual, measuring overtime), and how you interpret it.

Counting, agreement and bias

Frequencies are information, not meaning. Hsieh and Shannon (2005) allow rank-order comparisons of code frequencies, but frequency is not importance (Sandelowski, 2001). Always say what you counted: mentions, speakers or groups. Counting mentions across speakers lets one talkative participant manufacture a frequency.

Agreement between coders depends on your design. Intercoder reliability can be appropriate when the study treats the coding frame as a reproducible shared instrument, for example to compare with other studies using the same codebook. It is not a universal requirement, and the rationale should follow from the analytic design (O'Connor & Joffe, 2020). If you use it, report agreement by category, analyse disagreements, and check first that coders agree on unit boundaries (Campbell et al., 2013); Krippendorff's alpha is a common coefficient (Krippendorff, 2018). Negotiated agreement means that two coders code the same material and then discuss and reconcile their differences (Campbell et al., 2013). If only one person coded, a colleague's critical review of the decisions is peer debriefing, not agreement. High agreement is not validity: two coders can agree on the wrong category.

Bias is the main risk. Researchers working from a theory may be more likely to find supporting than non-supporting evidence, and emphasis on the theory can hide context (Hsieh & Shannon, 2005). The expectations memo, highlighting before coding and the residual are your safeguards.

Directed content analysis vs thematic analysis

Directed content analysis applies categories defined in advance and asks what the data mean for a theory; thematic analysis builds patterns of shared meaning, and in its reflexive form has no codebook (Braun et al., 2019). If your "codebook" is only a list of expected topics without definitions, you are closer to template analysis, where a priori themes may be changed or dropped. See Inductive vs deductive coding and How to choose a qualitative analysis method.

When this is not the right approach

• There is little or no theory about the phenomenon. Use conventional content analysis or thematic analysis.
• You want to know how people frame the phenomenon themselves, and imposed categories would hide it. Consider reflexive thematic analysis.
• Your "codebook" is a list of topics without definitions. Use template analysis with a priori themes.
• Your question is causal (did the intervention cause the change?). Content analysis of interviews cannot show that.
• You need to study interaction itself. A codebook from theory rarely contains interaction; see How to analyse focus group data.

Quality checklist

• The theory, written expectations and variant (fixed or growing codebook) were set before coding.
• Every category has a definition, an anchor example and a coding rule; there is a residual category.
• Units are defined; every codebook change has a date and a reason.
• The residual, overlaps and empty categories are analysed; new categories are marked as derived from the data.
• Counts say what was counted; non-supporting evidence is reported.

How the app supports this

The Deductive qualitative content analysis in SurveysAI Qualitative Analysis Workflow follows these steps with a codebook you supply in "Your codebook or list of themes". In "How to use your codebook" you choose:

• "Only my codebook": your categories and definitions are applied unchanged (unless you approve a proposed definition change after the pilot, which the report lists), and text no category covers is reported as a residual rather than forced into a category. This corresponds to the closed, with a residual mode above. The report names it deductive qualitative content analysis with a locked codebook (Mayring); methodologically, it is a locked-codebook variant informed by Mayring's deductive category application.
• "My codebook plus new categories from the data": your definitions still do not change, but new categories marked as derived from the data may be added beside your categories or under them, and are applied to all segments. This is closest to the unconstrained matrix. The report names it as directed content analysis following Hsieh & Shannon.

In both modes the analysis pilots the codebook to derive decision rules, codes every respondent segment, checks the residual and builds a category-by-group matrix. The analysis does not produce intercoder agreement: one AI model codes the data. An AI critic reads key outputs of the analysis, but it is the same AI model that wrote them, not a human and not a second coder, so it is not independent validation. If your design needs agreement statistics, have a human coder code a sample independently and compare. Quotations are checked word for word against the transcripts by program code. The app produces draft analytic outputs; you remain responsible for reviewing, revising and owning the interpretation. If you have only a list of themes without definitions, the "Template analysis for applied briefs" analysis is the better fit.

Run it with your codebook: create an account with Deductive qualitative content analysis, upload your transcripts and codebook, and pick "Only my codebook" for comparability or "My codebook plus new categories from the data" to extend the theory. If the analysis is not preselected, choose Deductive qualitative content analysis in the start dialog.

References

• Assarroudi, A., Heshmati Nabavi, F., Armat, M. R., Ebadi, A., & Vaismoradi, M. (2018). Directed qualitative content analysis: The description and elaboration of its underpinning methods and data analysis process. Journal of Research in Nursing, 23(1), 42–55. https://doi.org/10.1177/1744987117741667
• Bakker, A. B., & Demerouti, E. (2007). The Job Demands–Resources model: State of the art. Journal of Managerial Psychology, 22(3), 309–328. https://doi.org/10.1108/02683940710733115
• Braun, V., Clarke, V., Hayfield, N., & Terry, G. (2019). Thematic analysis. In P. Liamputtong (Ed.), Handbook of research methods in health social sciences (pp. 843–860). Springer. https://doi.org/10.1007/978-981-10-5251-4_103
• Campbell, J. L., Quincy, C., Osserman, J., & Pedersen, O. K. (2013). Coding in-depth semistructured interviews: Problems of unitization and intercoder reliability and agreement. Sociological Methods & Research, 42(3), 294–320. https://doi.org/10.1177/0049124113500475
• Demerouti, E., Bakker, A. B., Nachreiner, F., & Schaufeli, W. B. (2001). The job demands–resources model of burnout. Journal of Applied Psychology, 86(3), 499–512. https://doi.org/10.1037/0021-9010.86.3.499
• Elo, S., & Kyngäs, H. (2008). The qualitative content analysis process. Journal of Advanced Nursing, 62(1), 107–115. https://doi.org/10.1111/j.1365-2648.2007.04569.x
• Graneheim, U. H., & Lundman, B. (2004). Qualitative content analysis in nursing research: Concepts, procedures and measures to achieve trustworthiness. Nurse Education Today, 24(2), 105–112. https://doi.org/10.1016/j.nedt.2003.10.001
• Hsieh, H.-F., & Shannon, S. E. (2005). Three approaches to qualitative content analysis. Qualitative Health Research, 15(9), 1277–1288. https://doi.org/10.1177/1049732305276687
• Krippendorff, K. (2018). Content analysis: An introduction to its methodology (4th ed.). SAGE.
• MacQueen, K. M., McLellan, E., Kay, K., & Milstein, B. (1998). Codebook development for team-based qualitative analysis. Cultural Anthropology Methods, 10(2), 31–36. https://doi.org/10.1177/1525822X980100020301
• Mayring, P. (2000). Qualitative content analysis. Forum Qualitative Sozialforschung / Forum: Qualitative Social Research, 1(2), Article 20. https://doi.org/10.17169/fqs-1.2.1089
• Mayring, P. (2014). Qualitative content analysis: Theoretical foundation, basic procedures and software solution. Klagenfurt. https://nbn-resolving.org/urn:nbn:de:0168-ssoar-395173
• O'Connor, C., & Joffe, H. (2020). Intercoder reliability in qualitative research: Debates and practical guidelines. International Journal of Qualitative Methods, 19. https://doi.org/10.1177/1609406919899220
• Sandelowski, M. (2001). Real qualitative researchers do not count: The use of numbers in qualitative research. Research in Nursing & Health, 24(3), 230–240. https://doi.org/10.1002/nur.1025
• Schreier, M. (2012). Qualitative content analysis in practice. SAGE.
• Tims, M., & Bakker, A. B. (2010). Job crafting: Towards a new model of individual job redesign. SA Journal of Industrial Psychology, 36(2), Article 841. https://doi.org/10.4102/sajip.v36i2.841