Deductive vs Inductive Coding: Choosing and Combining

What Is the Difference?

Deductive coding starts from codes you set in advance, taken from the evaluation questions, a program's theory of change or an established framework. Inductive coding starts from the data: you read, notice what is there and name it. Hsieh and Shannon describe this contrast in content analysis as conventional (codes derived directly from the text) and directed (analysis starts with a theory or prior findings to guide initial codes).

Most evaluations combine them. Fereday and Muir-Cochrane describe a hybrid process that integrates data-driven codes with theory-driven ones and set out the stages in an exemplar. The question is not which is better but which source of codes fits each part of the question, and whether the analysis is honest about it.

Decision This Supports

This choice decides what the analysis can find. A frame built only from the program's theory of change can tell you whether people talk about the expected pathways. It is poor at revealing pathways nobody expected. An entirely open reading can surface the unexpected, but may not answer the commissioner's questions. The decision is where to be closed and where to stay open.

Before You Start

• Write the evaluation questions. See evaluation questions.
• List the codes you already expect and where they came from (a theory, a framework, earlier studies, or the team's own hunches).
• Decide whether your approach assumes a fixed codebook. Reflexive thematic analysis does not; see thematic analysis.

Comparing the Two

Thomas's general inductive approach was written for evaluators: it condenses raw text into a summary, links the findings to the evaluation objectives and develops a framework of the structure in the data. He also notes it is not as strong as some other strategies for theory development.

A Small Worked Example

This example is fictional.

An evaluator starts with two deductive codes from the program's theory of change: "Session Timing" and "Care Responsibilities." While reading, a passage describes the walk home after dark. It fits neither code. She does not force it. She adds an inductive code, "Travel Safety After Dark," writes a memo recording which excerpt prompted it and the date, and checks earlier transcripts for similar passages she skipped because the frame had no place for them. The codebook now has labels that show where each code came from: two deductive, one inductive. See the worked case.

Limits and Cautions

• A deductive frame can be treated as a hypothesis the data must confirm. Keep a place for "does not fit."
• "Emerging themes" is not a guarantee that no theory shaped the reading. Your assumptions are always present; see positionality and reflexivity.
• When coding is shared or assisted by software, code definitions carry the weight. See qualitative codebook.
• An AI system asked to code with a given codebook is doing deductive coding. Check its assignments against excerpts; see checking AI-assisted qualitative coding.

Frequently Asked Questions

Can I Start Deductive and Switch to Inductive?

Yes, if you record when and why. Changing codes midway means earlier transcripts may need to be revisited, and the codebook should show its versions.

Is Inductive Coding Free of Theory?

No. Researchers bring prior knowledge and assumptions. The aim is to let the data challenge them, and to say what they were.