Use this to choose and justify a method, and to make designs rigorous and defensible. AEC research is methodologically pluralistic (engineering experiments to social-science surveys to design science) — match method to question, then enforce that method's validity standards.
1. From question to design
- Formulate the question with a framework:
- PICO/PICOC (intervention/effect): Population, Intervention, Comparison, Outcome (, Context).
- PCC (scoping): Population, Concept, Context.
- SPIDER (qualitative): Sample, Phenomenon of Interest, Design, Evaluation, Research type.
- FINER test (is the question Feasible, Interesting, Novel, Ethical, Relevant?).
- Classify the aim: exploratory / descriptive / explanatory / predictive / design-and-evaluate.
- Pick the paradigm honestly: positivist (measure/test), interpretivist (understand/meaning), pragmatist (mixed, problem-driven), design science (build & evaluate an artifact).
2. Method families & when to use them (AEC-flavored)
Quantitative
- Survey / questionnaire — perceptions, adoption, practices at scale. Design: validated/established scales where possible, pilot test, Likert design, avoid double-barreled/leading items. Report sampling frame, response rate, non-response bias check. Common analysis: descriptive, factor analysis (EFA/CFA), reliability (Cronbach's α / composite reliability), regression, SEM/PLS-SEM (popular in AEC for adoption & performance models — report measurement + structural model, convergent/discriminant validity, fit indices; mind sample-size rules).
- Experiment / quasi-experiment — causal claims (e.g., VR safety training vs control; material tests). Control variables, randomization where possible, define IV/DV, power analysis for sample size, report effect sizes + CIs (not just p-values).
- Field measurement / monitoring study — instrument a building/site (energy, comfort, productivity, SHM). Define metrics, sensor calibration, measurement uncertainty, data QA/cleaning, period & seasonality.
- Secondary data / archival — cost/schedule/permit/energy datasets; document provenance & cleaning.
Qualitative
- Case study (Yin / Eisenhardt) — the AEC workhorse for projects/organizations. Single vs multiple case; embedded units; define the case & boundaries; triangulate sources (interviews, documents, observation, site data); use a case-study protocol & database for reliability; build a chain of evidence.
- Interviews (semi-structured) — sampling (purposive/theoretical), saturation, interview guide, recording/transcription, thematic analysis (Braun & Clarke) or grounded theory (open/axial/selective coding); report coding scheme & coder agreement; use NVivo/ATLAS.ti/Dedoose.
- Ethnography / observation — practice studies on site.
- Document/content analysis — contracts, specs, policies, standards.
Mixed methods (very common & valued in AEC)
- Designs: convergent (quant + qual in parallel, compare), explanatory sequential (quant → qual to explain), exploratory sequential (qual → build instrument → quant). State the design, the integration point, and why mixing adds value (don't just bolt interviews onto a survey).
Design Science Research (DSR) — central to BIM/tool/method papers
- Cycle (Hevner; Peffers DSRM): identify problem → define objectives → design & develop artifact (framework, tool, algorithm, ontology, model) → demonstrate → evaluate (case, experiment, expert panel, simulation) → communicate. The contribution is the artifact plus rigorous evaluation — demand the evaluation; a tool with only a toy demo is a common weak AEC paper.
Delphi method — expert consensus (e.g., ranking risks, KPIs, future scenarios)
- Multi-round anonymous expert survey with controlled feedback to convergence. Report: expert panel selection & size, rounds, consensus criterion (e.g., IQR, % agreement, Kendall's W), attrition.
Simulation & modeling studies (energy, structural, DES, ABM)
- Building performance simulation — declare tool & version, weather data, model assumptions, calibration against measured data + uncertainty/sensitivity analysis (Morris/Sobol); beware the performance gap; report inputs for reproducibility.
- Discrete-event simulation (DES) — construction operations/logistics; verify & validate the model, warm-up, replications, confidence intervals.
- Agent-based modeling (ABM) — occupant/worker behavior, markets; report ODD protocol (Overview, Design concepts, Details), calibration & validation, sensitivity analysis.
- Optimization studies — define objectives/constraints/decision variables, algorithm + parameters, convergence evidence, baseline comparison, Pareto analysis for multi-objective.
Multi-Criteria Decision Analysis (MCDM)
- AHP/ANP, TOPSIS, PROMETHEE, fuzzy variants for selection/ranking (materials, suppliers, delivery methods). Report criteria derivation, weighting method, consistency (AHP CR < 0.1), and sensitivity analysis.
3. Sampling
- Probability (random, stratified, cluster) vs non-probability (purposive, snowball, convenience — acknowledge limits). State the sampling frame, size justification (power analysis for quant; saturation for qual; rules-of-thumb for SEM e.g. ratio to indicators), and inclusion/exclusion.
4. Validity, reliability & rigor
- Quantitative: internal validity (confounds), external validity (generalization — AEC's chronic weakness; one building/one project rarely generalizes — say so), construct validity (measure what you claim), reliability (consistency); control for common-method bias in single-source surveys.
- Qualitative (Lincoln & Guba trustworthiness): credibility (triangulation, member checking), transferability (thick description), dependability (audit trail), confirmability (reflexivity).
- Simulation: verification (model built right) vs validation (right model — vs empirical data); uncertainty & sensitivity analysis are mandatory for credibility.
- Reproducibility: share data, code, model files where ethics permit (see
research-ethics-integrity).
5. Ethics gate (always check)
Human subjects (interviews/surveys/experiments/wearables) → informed consent, IRB/ethics approval, data protection (GDPR), anonymization. Site data → confidentiality/NDA. Declare conflicts and funding.
6. Method-selection heuristic
- "How widespread / what perceptions?" → survey.
- "Why/how in context?" → case study / interviews.
- "Does X cause/improve Y?" → experiment (or strong quasi-experiment).
- "Build a new tool/method and show it works" → design science + evaluation.
- "What will/should the field prioritize?" → Delphi.
- "How does the system behave under conditions we can't field-test?" → simulation (validated).
- "Rank options under multiple criteria" → MCDM.
- "Both numbers and meaning" → mixed methods (state the design & integration). Always pair the choice with its specific validity checklist above.