What Is the Delphi Method? A Step-by-Step Researcher's Guide
The Delphi method is a structured, multi-round survey technique used to build consensus among experts on questions where no single dataset exists. It's widely used in HRM future-skills research, health-care guideline development, education-policy planning, and curriculum design.
Why use Delphi?
Sometimes the question you're answering doesn't have a data warehouse behind it. Examples:
- Which HR competencies will be most important in five years?
- What should a master's-level data-science curriculum include?
- Which clinical guideline should apply when randomised trial evidence is unavailable?
For these, the best available data is structured expert judgement. Delphi makes that judgement transparent, repeatable, and consensus-based — rather than relying on a single committee meeting.
The four pillars of a Delphi study
- Anonymity. Experts respond independently; their identities are not shared between rounds. This neutralises personality, hierarchy, and social-pressure effects.
- Iteration. Experts respond in two or more rounds. After each round, they see aggregated feedback before the next round.
- Controlled feedback. Between rounds, the researcher shares summary statistics (median, interquartile range, sometimes verbatim minority comments) — not free-form discussion.
- Statistical aggregation. The final result is a numerical summary of expert opinion, with explicit thresholds for what counts as consensus.
How many rounds?
Three rounds is the modal choice. Two-round Delphi studies are common in time-pressured contexts. Four or more rounds risk fatigue and dropout.
| Round | Task |
|---|---|
| Round 1 | Generate items + initial rating. Often mixes open-ended generation ("what additional skills should be added?") with rating of a starter list. |
| Round 2 | Re-rate the items in light of aggregated Round 1 feedback. Items below a threshold are dropped. |
| Round 3 | Final re-rating of the trimmed list, often with prioritisation (top-5 or top-10 ranking). |
How many experts?
There's no statistical sample-size formula because Delphi is not about generalising to a population — it's about saturating expert perspectives. Practical norms:
- 10–15 experts — minimum for credibility; common in highly specialised domains.
- 20–30 experts — typical for management, HRM, education research.
- 30+ experts — useful when the topic spans multiple disciplines or regions.
Expect 20–30% attrition between rounds; recruit accordingly.
What counts as consensus?
This is where researchers disagree, and where you should pre-register your criterion. Common thresholds:
- Percentage agreement — e.g., ≥70% of experts rate an item as Important or Very Important.
- Interquartile range (IQR) — IQR ≤ 1.5 on a 7-point scale, indicating tight spread.
- Mean / median threshold — e.g., median ≥ 5.5 on a 7-point scale.
- Stability across rounds — no significant change in median rating between consecutive rounds.
Best practice: pre-register two of these criteria (a substantive and a stability criterion) and report both.
Fuzzy Delphi — the modified variant
Fuzzy Delphi was developed to capture uncertainty more honestly. Instead of asking each expert for a single rating, you ask for three values per item: most pessimistic, most likely, and most optimistic. These form a triangular fuzzy number, which is then aggregated mathematically across experts.
Fuzzy Delphi is increasingly common in HR competency frameworks, education policy, and strategic foresight research. The advantage: it reveals where experts are confident versus where they are merely guessing.
Common pitfalls
- Selecting like-minded experts. Diversity of perspectives matters more than depth in a single perspective. Recruit across sectors, regions, and career stages.
- Failing to define consensus before Round 1. Researchers who define consensus post-hoc are accused of cherry-picking.
- Excessive rounds. Each additional round increases attrition. Three is usually enough.
- Not closing the loop. Send experts a summary of the final results and credit them in the acknowledgements (with permission).
Run your Delphi in SmartForm
SmartForm includes three consensus-design templates — Delphi Round 1 (mixed open-generation + rating), Fuzzy Delphi (three-value triangular ratings), and Nominal Group (rate-and-rank). Each is editable, supports anonymous SGPC matching across rounds, and exports CSV and Word summaries.
Final thought
Delphi is the right method when expert judgement is the best available evidence and when you need that judgement to be transparent, anonymised, and repeatable. It's not the right method when you have real data — it's a complement to empirical research, not a substitute for it.