What is the difference between inductive and deductive thematic analysis?

The choice is usually decided by the question rather than by preference. If you are asking what is going on in a setting nobody has described, a predetermined frame would filter out exactly what you came to find. If you are asking whether an established model accounts for a new context, coding without that model wastes the theory that made the question worth asking.

The purely inductive position is harder to hold than it sounds. Nobody codes from nowhere. You have read the literature, you have expectations, and those shape what you notice. Grounded theory's advice to bracket the literature is a discipline for managing that influence rather than a claim that it can be removed, and pretending to a blank slate is less credible than acknowledging your starting point.

Deductive coding has the opposite risk. A frame tells you where to look, and everything outside it becomes invisible. The safeguard is an explicit category for data that does not fit, reviewed at the end rather than discarded during coding. That residual category is often where the interesting finding is.

One study ran both routes over the same material and reported what each produced. Bonner and colleagues took systematic reviews of parent-level barriers to childhood vaccination, coded 583 barrier descriptions inductively into 74 barriers across seven categories, then mapped the same material deductively onto the COM-B model and the Theoretical Domains Framework. Two authors coded independently and settled disagreements by discussion, and the first deductive mapping reached 89 percent agreement at domain level, though construct level needed further definitions before it stabilised (Bonner et al., 2021, p. 1).

The two outputs were not interchangeable. The inductive framework held concrete items such as false contraindications, language barriers, distance, cost, distrust, and fear of needles. The deductive framework guaranteed systematic attention to specified behavioural domains, but only 10 of the 14 domains were clearly represented in the data, Behavioural Regulation was absent, and Intentions and Goals appeared only under lenient readings. Some domains, Environmental Context and Resources in particular, were too broad to separate one barrier type from another (Bonner et al., 2021, p. 6). Theory found gaps the inductive review had missed, and the authors still judged the inductive categories more relevant to the implementation problem in front of them (Bonner et al., 2021, p. 7).

Whichever you choose, the reporting requirement is the same: a reader should be able to see where your categories came from. That is what separates an analysis from an assertion.

Can I combine both approaches?

Yes, and many studies do. A common design applies a deductive frame from existing theory, then codes inductively for anything the frame does not capture. Report it as a hybrid approach and be explicit about which themes came from which route, since a reader cannot otherwise tell.

Sequence matters and should be stated. Coding deductively first and then inductively produces a different result from the reverse, because the first pass shapes what you notice in the second.

Budget for the second pass before you commit to it. The deductive mapping and the discussion that resolved it took two authors with prior knowledge of the behavioural frameworks about two weeks, on material that had already been coded once (Bonner et al., 2021, pp. 2-3). Bonner and colleagues also warn that such frameworks can render internal motivation more sharply than practical or physical opportunity, which understates barriers like time, cost, distance, and service availability. If your research question turns on those, a deductive frame alone will flatten them.

The hybrid design is particularly suited to a doctoral study extending an existing model, since it lets you test the model and remain open to what it misses, which is usually the contribution.

How do I report which approach I used?

Name it in the methods, say where the codes came from, and state what happened when data did not fit. For deductive work, give the frame and its source. For inductive work, describe how codes were generated and refined. For hybrid work, say which themes originated where.

Include the coding frame itself, in an appendix if it is long. A reader who can see your categories can evaluate your analysis, and one who cannot has to take it on trust.

Say how the frame changed during analysis, because it almost certainly did. A frame revised in response to the data, with the revision reported, is stronger evidence of careful work than one presented as fixed from the start.

Separate the layers when you write up. Thomas and Harden distinguish descriptive themes, which stay close to the content of the primary material, from analytical themes, which go beyond it to answer the review question, and they treat the move between the two as the step most in need of explicit reporting (Thomas and Harden, 2008, pp. 8-9). If a reader cannot tell which of your higher-order categories came from the data and which came from your interpretation of it, the frame you used has stopped being checkable.

One sentence is worth writing whichever route you took: what you would have had to see in the data to reach a different conclusion. Answering that shows the analysis could have come out otherwise, which is what makes it an analysis.