Qualitative Analysis/ Inductive vs Deductive Coding
Qualitative Analysis, Coding

Inductive vs deductive coding in qualitative research

One of the most fundamental decisions in qualitative coding is whether your codes should come from the data or from a framework you bring to the data. This guide explains what inductive and deductive coding mean in practice, when to use each, and how most applied research combines both.

9 min read Thematic Coding Beginner to intermediate

Before you code a single passage of text, you face a choice that will shape everything that follows: are you going to let the data tell you what matters, or are you going to bring a framework and test it against the data? This is the difference between inductive and deductive coding, and it is not a technical distinction. It reflects a deeper question about the purpose of the research and what kind of knowledge you are trying to produce.

This guide explains each approach, when to use it, what it looks like in practice, and how the two are combined in most applied qualitative research.

01What inductive coding means

Inductive coding means allowing the codes to emerge from the data itself. You read the transcripts without a predetermined set of labels and attach codes to passages based on what you observe in the text. The codes are generated from the data, not imported from outside it.

In practice, inductive coding looks like this: you read a passage in which a participant describes a difficult experience at a health clinic, and you label it "negative clinic experience." You did not arrive at the interview with that code. The data prompted it. As you read more transcripts, you might find that some passages are specifically about waiting times and others about staff attitude. You refine your initial code into two: "waiting time frustration" and "staff manner." The codes become more precise as you read more data.

The assumption behind inductive coding is that you do not yet know, or should not assume, what the significant dimensions of the phenomenon are. You want the participants' own categories and concerns to drive the analysis.

Inductive does not mean theory-free. Every researcher brings assumptions, disciplinary knowledge, and interpretive frameworks to the data. Inductive coding means those frameworks do not predetermine the code structure. It does not mean reading without any conceptual lens at all. Staying close to the data is not the same as having no prior knowledge.

02When to use inductive coding

Inductive coding is most appropriate when:

03What deductive coding means

Deductive coding means you bring a set of codes to the data in advance and apply them systematically. The codes come from a theoretical model, a programme logic, an evaluation framework, a previous study, or the research questions themselves. You read the transcripts looking for evidence of each predefined code.

In practice, deductive coding looks like this: an evaluation of a nutrition programme has a theory of change with four constructs: knowledge, attitudes, social norms, and behaviour. The evaluator creates four codes before reading any transcript. When a participant says "I knew I should be giving my child more vegetables but I never thought it was important enough to change," the evaluator codes this passage as "attitude" and "behaviour." The structure was in place before the data was read.

The assumption behind deductive coding is that the relevant dimensions are already known, either from theory, from the programme design, or from the evaluation questions that need to be answered. The task is to assess the data against those dimensions, not to discover new ones.

Deductive coding is not less rigorous than inductive coding. The two serve different purposes. Deductive coding is more appropriate when the questions are predefined, when comparability across cases matters, or when the findings need to be framed within an established theoretical or programmatic structure. Neither approach is superior: the choice depends on what the research is for.

04When to use deductive coding

Deductive coding is most appropriate when:

05The mixed approach

Most applied qualitative research uses a combination of both. The starting point is a set of deductive codes derived from the research questions, the interview guide, or a relevant theoretical framework. These codes provide initial structure and ensure the analysis directly addresses the questions the study was designed to answer.

As coding progresses, new codes are added inductively when the data contains content that does not fit any existing code and that seems analytically significant. These inductive additions capture what the deductive framework did not anticipate.

The result is a hybrid codebook: a structured core of deductive codes, supplemented by a set of inductive codes that emerged during the analysis. The deductive codes ensure coverage of the research questions; the inductive codes ensure the analysis is not blind to what the data offers beyond those questions.

Pure inductive
All codes from data
Appropriate for exploratory research, grounded theory, and studies where the researcher's prior knowledge should not shape the emerging structure. Time-intensive. Best suited to academic research with sufficient resources.
Pure deductive
All codes from framework
Appropriate when the research questions are fully specified and the goal is to assess evidence for each construct. Risk of missing unexpected findings. Most common in rapid evaluation work.
Mixed (most common)
Framework plus inductive additions
Starts with a deductive structure, adds codes inductively as the analysis proceeds. Balances coverage of predefined questions with openness to unexpected findings. The standard approach in applied social research and programme evaluation.
Template analysis
Framework revised iteratively
A variant in which the initial template (deductive structure) is revised across multiple passes through the data, with codes added, removed, and reorganised until the template accurately reflects the data. More structured than pure inductive, more flexible than pure deductive.

06What this looks like in practice

Consider a study of barriers to women's participation in a rural savings programme. The evaluation team has four research questions: about awareness of the programme, attitudes toward financial participation, social norms around women and money, and practical barriers to attendance.

They create four deductive codes before reading any transcript, one for each research question. They begin coding the first transcript and immediately encounter a passage about a woman's husband preventing her from attending meetings. There is no code for spousal control. They add it inductively.

Further into the same transcript, they find a passage about a woman who said she felt embarrassed because she could not read the savings records. They add a code for literacy barriers. Neither of these was in the original research questions, but both are analytically significant.

By the end of the coding process, the codebook has four deductive codes and six inductive ones. The deductive codes ensure the analysis answers the four commissioned questions. The inductive codes ensure the analysis captures what the data contained beyond those questions, including findings that may be more actionable for the programme than the questions the evaluators started with.

07Common mistakes

Treating the approach as fixed once decided. Starting with a deductive framework does not mean you cannot add codes. Starting inductively does not mean you cannot use prior theory to interpret what you find. The approach should be responsive to what the data demands.

Creating too many codes too quickly. Inductive coding in particular can generate a very large number of codes if the analyst codes every distinct topic mentioned in the data. Codes that apply to only one passage in one transcript are usually not analytically useful at the theme level. Keep codes that appear across multiple participants or that are analytically significant even if rare.

Conflating codes and themes. A code is a label for a passage. A theme is a pattern across passages and participants. Inductive coding produces codes; themes are constructed through a further analytical step of reviewing, grouping, and interpreting the codes. Presenting codes directly as findings, without this synthesis step, produces a catalogue rather than an analysis.

Leaving deductive codes unchallenged. If a deductive code has very few or no passages assigned to it after coding all the data, that is a finding. It may mean the construct is not relevant to this population, that the interview guide did not explore it adequately, or that participants discussed it in terms your code did not capture. Investigate before concluding the theme is simply absent.

Not documenting the choice. Whether you are coding inductively, deductively, or with a mixed approach, state this in your methods section. Reviewers and readers need to know whether the themes emerged from the data or were imported from a framework, because the two make different epistemological claims. A methods section that says only "themes were identified from the data" when a deductive framework was used is misleading.

08How to do it in AnalyZ Solutions

Both approaches are supported within the same coding workflow. The difference is in how you set up the codebook before you begin.

Deductive: build the codebook first

Open Manage Codes in Thematic Coding and create your codes before reading any transcript. Derive the code names and definitions from your research questions, evaluation framework, or theoretical model. Add a description to each code explaining what it captures and, where helpful, what it does not include. Then move to Assign Codes and work through each transcript applying the existing codes.

Inductive: build the codebook during coding

Open Assign Codes directly and begin reading the first transcript. When you encounter a passage that seems significant and no existing code fits, create a new code from within the assignment view without leaving the transcript. Add it to the passage immediately. Continue through the transcripts, adding codes as the data demands. After coding all transcripts, open Manage Codes and review the codebook: merge codes that are capturing the same idea, rename codes whose labels no longer feel precise, and remove codes that appear only once and carry no analytical weight.

Mixed: start with a framework and extend inductively

Create your deductive codes in Manage Codes, then move to Assign Codes and begin coding. Whenever you encounter content that no existing code captures, create a new code in the same session. Keep the inductive codes visually distinct by assigning them a different colour from the deductive codes. After all transcripts are coded, review the inductive codes in Manage Codes to assess which are analytically substantive and which can be merged or discarded.

Frequently Asked Questions
Does the choice of approach affect what I can claim in my findings?
Yes, and this matters for how you write up the research. Findings from inductive coding can claim that the themes emerged from participants' own accounts, which is an epistemological claim about how the categories were generated. Findings from deductive coding can claim that evidence was systematically assessed against a predefined framework, which is a different kind of claim. Mixed approaches need to be transparent about which findings come from which source. Reviewers of academic work in particular will scrutinise the match between the stated approach and the claims made about findings.
Can I switch from deductive to inductive mid-study?
You can adjust your approach as the analysis develops, and in practice most researchers do. What matters is that you document the shift: when you added inductive codes to a deductive framework, what prompted the addition, and how the inductive codes relate to the deductive ones. An undocumented shift in approach creates ambiguity about the status of different codes in the final analysis. A documented shift is a methodological decision, not an inconsistency.
How do I know if a code is inductive or deductive in a mixed codebook?
Track the provenance of each code as you build the codebook. A simple convention: note in the code description whether it was defined before coding began (deductive), added during coding (inductive), or derived from a deductive code through splitting or refinement (hybrid). This record is useful both for your own analytical transparency and for the methods section of the final report. In AnalyZ Solutions, colour-coding deductive and inductive codes differently makes the distinction visible at a glance during coding.
What if my deductive codes do not fit the data at all?
This is an important finding in itself. If participants systematically fail to speak to the constructs in your framework, there are several possible explanations: the framework may not be appropriate for this population or context, the interview guide may not have explored the constructs adequately, or the phenomenon operates differently from how the framework predicts. Before concluding the framework simply does not apply, read through the transcripts to see whether participants are discussing the same underlying concern in different language. If so, a code revision rather than a framework rejection may be the right response.
Is grounded theory the same as inductive coding?
Grounded theory uses inductive coding, but inductive coding is not the same as grounded theory. Grounded theory is a full methodology with specific procedures: open coding, axial coding, selective coding, constant comparison, theoretical sampling, and saturation. Inductive coding, as discussed in this guide, refers simply to allowing codes to emerge from the data rather than importing them from a framework. You can code inductively without following grounded theory procedures. Grounded theory is one application of inductive principles, not the only one.
How many codes is too many?
There is no fixed upper limit, but a codebook with more than 40 to 50 codes is usually a sign that the analyst has not yet moved from coding to analysis. Large codebooks often contain many codes that are subtle variations of the same underlying idea. Before concluding that you have 60 distinct codes, review the codebook for codes that could be merged without losing analytical distinction. The goal is a codebook that is comprehensive enough to capture the full range of content in the data and parsimonious enough that each code represents a genuinely distinct analytical category.

Start coding your qualitative data

Upload your transcripts and build your codebook in AnalyZ Solutions. Free to start, free to analyze.

Get started
Related Guides