A methodology chapter should do more than name a design and list tools. Its real job is to show a coherent chain of reasoning: why the question requires a particular kind of evidence, how that evidence will be collected, how it will be analysed and what the resulting claims can—and cannot—support.
When examiners challenge methodology, they often challenge gaps in that chain. A sample may not represent the population implied by the research question. A measurement may not capture the construct being discussed. An analysis may answer a different question from the one stated. The strongest methodology makes these connections visible before data collection begins.
1. Start with the decision the study must support
A broad topic such as “AI in education” is not yet a research question. Define the population, phenomenon, context and type of claim. Are you describing current practice, estimating an association, comparing interventions, explaining a process or designing an artefact?
Write the question in a form that indicates what evidence would count as an answer. Then list the main claim you expect the study to make and the strongest alternative explanation a reviewer might raise. This immediately exposes what the design must control, observe or acknowledge.
2. Match the design to the claim
| Research purpose | Possible design direction | Primary risk to address |
|---|---|---|
| Describe a population or practice | Survey, observational or descriptive study | Coverage, measurement and non-response bias |
| Understand experience or process | Interview, focus group, case study or ethnographic design | Reflexivity, sampling rationale and interpretive transparency |
| Estimate association | Correlational or longitudinal design | Confounding, directionality and model assumptions |
| Estimate an intervention effect | Experimental or quasi-experimental design | Assignment, comparability, attrition and implementation fidelity |
| Build and evaluate a system | Design science or engineering experiment | Baselines, evaluation realism, reproducibility and generalisation |
The design name is not a justification. Explain why its structure is appropriate for the claim and what it cannot establish. A transparent limitation is stronger than an unsupported causal implication.
3. Make sampling part of the argument
Sampling decisions determine the boundaries of inference. Identify the target population, accessible population, sampling frame, inclusion criteria and recruitment process. If probability sampling is not feasible, explain the practical constraint and how it affects transferability or generalisation.
For quantitative work, justify sample size using the intended analysis, expected uncertainty, power or precision—not a generic rule. For qualitative work, connect sample strategy and stopping logic to information needs, diversity of cases and analytic depth.
4. Define constructs before selecting instruments
Terms such as engagement, quality, trust and performance are constructs, not direct observations. Define what each means in the study, how it will be operationalised and what evidence supports the measurement.
- Use validated instruments when they fit the population and context.
- Document adaptation, translation, pilot testing and scoring decisions.
- Separate predictors, outcomes, confounders and descriptive variables.
- Record data provenance, missingness and quality checks.
5. Plan analysis before seeing the results
Map each research question to variables or qualitative material, preparation steps, analytic technique and interpretation rule. For statistical analysis, document assumptions, missing-data handling, sensitivity checks and effect-size reporting. For qualitative analysis, describe coding, reflexivity, comparison, negative cases and how interpretations will be supported by evidence.
6. Treat validity and ethics as design work
Quality is not a paragraph added at the end. Build it into recruitment, measurement, analysis and reporting. Depending on the design, address internal and external validity, credibility, dependability, triangulation, robustness, calibration, reproducibility or benchmark relevance.
Ethics should cover consent, privacy, data minimisation, security, risk, vulnerable groups, conflicts of interest and responsible use of AI or automated tools. Institutional approval requirements must be resolved before relevant data collection begins.
7. Write the methodology as an auditable sequence
A clear chapter commonly moves from research philosophy and design to setting, population, sampling, variables or constructs, instruments, procedure, analysis, quality safeguards, ethics and limitations. The exact order matters less than the reasoning between sections.
Finish by checking alignment. Every objective should have evidence. Every dataset should have a purpose. Every analysis should answer a stated question. Every claim should stay within the boundaries created by design and sampling.
Bring your research questions, proposed design and current methodology outline. We can identify alignment gaps and turn them into a practical revision plan.
Discuss your methodology