Inductive vs deductive coding: what is the difference, and which should you use?

This guide explains each approach, shows one interview excerpt coded both ways and gives a table to help you choose. If you have not coded before, start with our step-by-step guide to how to code qualitative data. It is part of our series on AI for qualitative research and research by field.

What is inductive coding?

Inductive coding is coding in which the codes are generated from the data itself, without a predefined list. You read a transcript, label what seems relevant to your question, and let the codebook grow as you go. It is sometimes called data-driven or bottom-up coding.

Inductive coding does not mean coding without any perspective. Your research question, discipline and experience still shape what you notice. Braun and Clarke make this point about thematic analysis: an inductive orientation is driven by the data's content, but the researcher is never a blank slate [1][2].

Inductive coding works well when:

• Little research exists on the topic or population.
• You want to capture participants' own language and categories.
• Existing theories may not fit the setting.

What is deductive coding?

Deductive coding is coding in which you apply a set of codes defined before the analysis, usually drawn from a theory, framework, interview guide or earlier study. These are called a priori codes. It is also called theory-driven or top-down coding.

Deductive coding works well when:

• Your research question is built on a named framework (for example, a behavior change model or an implementation framework).
• You want to compare your findings with earlier studies that used the same codes.
• Several coders must work consistently, and you plan to report agreement between them.

The risk is that you only see what the framework tells you to look for. Anything outside the codebook can disappear from the analysis.

The hybrid approach

Fereday and Muir-Cochrane described a hybrid approach in 2006: they built a codebook from a theoretical framework before coding, applied it to the data, and added new inductive codes where the data showed something the codebook did not capture [3].

In practice the hybrid works like this:

1. Draft a priori codes from your framework, each with a definition and include and exclude rules.
2. Code two or three transcripts with that codebook.
3. Flag passages that matter for your question but fit no code.
4. Write new inductive codes for them and add them to the codebook, marked as data-derived.
5. Continue coding, revisiting earlier transcripts when you add codes.
6. In your report, say which codes came from theory and which from the data.

A side-by-side example (invented)

Example (invented for illustration): a researcher studies why adults stop attending a free community exercise program. One participant, P04, says:

"At first I loved it, the instructor knew my name. But then they moved it to 6pm and I'm still on the bus at 6. I tried the Saturday one but it was full of fit young people and I felt like I was holding everyone back. I didn't want to be the one at the back puffing. So I just stopped going and didn't tell anyone."

Coded deductively

Suppose the study is framed by the COM-B model, in which behavior depends on capability, opportunity and motivation [4]. The a priori codebook has codes for each component and subcomponent.

Coded inductively

Now the same excerpt with no codebook, labelling what stands out:

What the comparison shows

The deductive coding sorts the reasons neatly into the model and makes the study easy to compare with other COM-B studies. The inductive coding catches two things the framework does not name: being recognized as a reason to come, and the "quiet exit", which may matter for how the program follows up dropouts. A hybrid would keep the COM-B codes and add "Quiet exit" as a new data-derived code.

Inductive vs deductive coding compared

How to choose

Ask three questions:

1. Does my research question name a theory or framework? If yes, deductive or hybrid. If no, inductive.
2. Do I need to compare with earlier studies using the same categories? If yes, deductive codes make that possible.
3. How much do I expect the data to surprise me? If a lot, leave room for inductive codes.

Then write the choice into your methods section. Reviewers do not object to either approach; they object to an approach that is not stated, or to a study that claims to be inductive while its codes match the interview guide one to one.

Common mistakes

• Calling an analysis inductive when the codes mirror the interview questions. That is a deductive structure; say so.
• Forcing data into a framework. If a passage fits no a priori code, it may be the most interesting one. Add a code.
• A priori codes without definitions. A framework label such as "social opportunity" still needs include and exclude rules for your data.
• Not marking code origin in a hybrid. Readers need to know which codes came from theory.
• Reporting inter-rater agreement for a reflexive inductive analysis. It contradicts the method; use an audit trail instead.

Doing it in AI for Research

The qualitative workbook in Workbooks (the /workbooks route) supports all three approaches. For deductive or hybrid coding, enter your a priori codes in the codebook first, each with a definition, "include when" and "exclude when". For inductive coding, start with an empty codebook and add codes as you select text in an uploaded transcript. Each coded excerpt keeps the exact quote, and you can flag negative cases and see code frequency by source.

Optional AI code suggestions cost 3 credits. Every suggestion quotes the passage it rests on, so you can check it and accept or reject it; in a deductive study, you can compare suggestions with your framework codes to find passages the codebook misses. The themes are yours. See qualitative coding and pricing for details.

FAQ

What is the difference between inductive and deductive coding?

Inductive coding builds codes from the data: you read the transcripts and label what you find without a list prepared in advance. Deductive coding starts from a list of codes drawn from a theory, framework or earlier study and applies it to the data. Inductive coding suits exploratory questions; deductive coding suits questions that test or extend an existing framework.

Can I use both inductive and deductive coding?

Yes. A hybrid approach, described by Fereday and Muir-Cochrane in 2006, starts with a codebook built from theory and then adds new codes for anything important in the data that the codebook does not cover. Many qualitative studies work this way. Say so in your methods section and mark which codes were set in advance and which came from the data.

Is thematic analysis inductive or deductive?

It can be either. Braun and Clarke describe an orientation to the data that runs from inductive, where codes and themes come from the content, to deductive, where existing concepts shape the coding. Most analyses sit somewhere between the two. What matters is that you state your orientation and apply it consistently.

Which coding approach is better for a thesis?

Neither is better in general. Choose inductive coding when little is known about the topic or you want participants' own framings. Choose deductive coding when your research question is built on a named framework, such as a behavior change model, and you want to see how the data fit it. If your proposal names a framework but you expect surprises, a hybrid approach is often the most defensible choice.

What are a priori codes?

A priori codes are codes defined before you look at the data, usually from a theory, framework, interview guide or earlier study. They are the basis of deductive coding. Each one still needs a definition and include and exclude rules in your codebook, so that you apply it consistently.

Learn more

• The full process from transcript to themes: how to code qualitative data.
• Reporting agreement between coders: Cohen's kappa and inter-rater reliability.
• Choosing a framework for your study: theoretical vs conceptual framework.
• Back to the series guide: AI for qualitative research.

Sources

1. Braun, V., & Clarke, V. (2006). Using thematic analysis in psychology. Qualitative Research in Psychology, 3(2), 77–101. https://doi.org/10.1191/1478088706qp063oa
2. Braun, V., & Clarke, V. (2022). Thematic Analysis: A Practical Guide. Sage.
3. Fereday, J., & Muir-Cochrane, E. (2006). Demonstrating rigor using thematic analysis: A hybrid approach of inductive and deductive coding and theme development. International Journal of Qualitative Methods, 5(1), 80–92. https://doi.org/10.1177/160940690600500107
4. Michie, S., van Stralen, M. M., & West, R. (2011). The behaviour change wheel: A new method for characterising and designing behaviour change interventions. Implementation Science, 6, 42. https://doi.org/10.1186/1748-5908-6-42