Written and maintained by CASRAI Editorial Board
Last updated
The nominal group technique (NGT) is a structured, face-to-face group process for generating and prioritising ideas in a single session. Participants are physically together but work silently for the first stage — the group is “nominal” (a group in name only) at the point where ideas are produced, which is precisely the mechanism that stops the loudest or most senior person in the room from setting the agenda for everyone else.
It was formalised by André Delbecq and Andrew Van de Ven in 1971 and set out in full in their 1975 book with David Gustafson, which is also where the technique’s long-standing pairing with the Delphi method comes from — the two were published as companion procedures for the same problem, and the choice between them is still the first real decision you make.
This page is a runnable session script: what happens in each stage, how long to allow, what the facilitator must and must not do, how the voting actually works, and — importantly — which parts of the “standard” procedure are genuinely settled and which vary so much in published practice that you need to justify your own choice rather than cite a rule.
The four stages, and what makes each one work
Every description of NGT since 1971 shares the same four-stage spine. The stages are not interchangeable and the order is load-bearing: silent generation before any talking is what protects independent judgement, and structured round-robin reporting before any evaluation is what stops early ideas being killed on arrival.
| Stage | What happens | What it protects against |
|---|---|---|
| 1. Silent generation | Participants write their own responses to the question privately, with no discussion. | Anchoring on whoever speaks first; production blocking, where waiting for a turn to talk stops you thinking. |
| 2. Round-robin | The facilitator takes one item from each participant in turn, recording it verbatim on a visible list. Repeat until everyone passes. | Unequal airtime. Everyone contributes the same number of times regardless of status or confidence. |
| 3. Clarification | Each listed item is discussed only for meaning — what does it mean, is it a duplicate, should two items merge. | Debate about merit leaking into the stage meant for comprehension; items dropped because they were poorly phrased rather than unimportant. |
| 4. Voting and ranking | Participants privately select and rank a subset of items. Scores are aggregated and displayed. | Consensus by verbal attrition. The output is an arithmetic aggregate of private judgements, not whatever the room could be talked into. |
Many published studies add a fifth step — a brief discussion of the vote, sometimes followed by a second vote. That is a legitimate and common extension, but it is an extension: report it as one rather than presenting it as part of the classical procedure.
A worked 90-minute session script
The timings below are a practical allocation for a single group of roughly seven people answering one question. They are an example, not a standard — see the section on what genuinely varies for why no source can honestly give you canonical minute counts. Use this as a starting template and adjust for group size, because stage 2 is the only stage whose length scales directly with headcount.
0:00–0:10 — Framing the question
State the single question, written up where everyone can see it, and leave it visible for the whole session. Take clarifying questions about scope only. Do not take answers yet — if someone starts answering, park it and tell them to write it down.
The question is the single highest-leverage thing in the session. It must be answerable in a short written phrase, must not contain two questions joined by “and”, and must not embed the answer. “What prevents researchers in this faculty from depositing data?” works. “How can we improve data deposit compliance and training?” does not — it is two questions, and it presumes the answer is compliance and training.
0:10–0:20 — Silent generation
Participants write privately. No talking, no comparing notes, no phones. The facilitator writes their own list too, or sits still — what they must not do is walk around reading over shoulders, which reintroduces exactly the observation effect the stage exists to remove.
Ten minutes is a common allocation; some scripts use five, others fifteen. Watch the room rather than the clock: when most people have stopped writing, give a one-minute warning.
0:20–0:45 — Round-robin
Go round the table taking one item per person per turn. Record it on a flipchart or shared screen in the participant’s own words, numbered. Anyone with nothing new says “pass” — and passing on one round does not exclude them from the next. Continue until everyone passes consecutively.
Two rules do most of the work here. First, record verbatim: the moment a facilitator paraphrases, they have started editing the data. Second, no discussion at all — not even agreement. “Yes, that’s a good one” is evaluation, and it tells the next speaker which kinds of item are welcome.
Budget roughly three minutes per participant. A group of seven typically produces 25–40 items; if you are getting many more, the question was too broad.
0:45–1:10 — Clarification
Walk the numbered list top to bottom. For each item ask only: is it clear what this means, and is it distinct from anything else on the list? The originator explains; the group may ask questions.
Merge duplicates only with the explicit agreement of both originators, and record the merge. This is where NGT quietly loses data if run carelessly — a facilitator collapsing “no time” and “competing priorities” into one item because they look similar has made a substantive analytic decision on the group’s behalf. If in doubt, leave both.
Nothing is deleted at this stage for being unimportant. Unimportant items are handled by the vote, which is what the vote is for.
1:10–1:25 — Voting and ranking
Each participant privately selects a fixed number of items and ranks them. The most common design is to pick the top five and assign 5 points to the most important down to 1 for the fifth, but selecting eight or ten is also common on longer lists. Give the exact instruction in writing — ambiguity about whether 1 means best or worst will silently corrupt the aggregate.
Collect the cards, aggregate, and display the result. Report two numbers per item, not one: the total score and the number of participants who voted for it at all. An item with 15 points from one enthusiast is a fundamentally different finding from one with 15 points spread across six people, and a total-score-only table hides that difference completely.
1:25–1:30 — Feed back and close
Show the ranked list to the group before they leave and confirm it is a fair record of the session. Say explicitly what happens to the output next. If you intend to run a discussion-and-revote round, this is where it goes — and it will add 20–30 minutes.
Facilitation rules that actually carry the method
NGT is unusually easy to run badly while appearing to run correctly, because the failure modes all look like helpfulness. The following are the rules whose violation changes the result:
- Never evaluate during stages 1–3. The facilitator’s approval is a stronger signal than any participant’s, and a single “good point” reshapes the remaining contributions.
- Record verbatim, in the participant’s own words. Paraphrasing is analysis, and analysis before the vote contaminates the vote.
- One item per turn, no exceptions. Allowing a participant to read their whole list because “it’s quicker” hands the agenda to whoever is fastest.
- Vote privately, always. A show of hands converts NGT into a public preference cascade and discards its main advantage over an ordinary meeting.
- Do not let the convener facilitate. If the person running the session is also the person whose programme is being evaluated, participants will read their reactions, whatever the script says. This is the same reason social desirability bias is a live threat in any face-to-face method.
- Decide the tie-breaking rule before you vote, not after you see a tie you dislike.
What genuinely varies — and what to say in your methods section
The four-stage structure is settled. Almost every specific number attached to it is not, and published NGT studies vary far more widely than the how-to literature implies. Two scoping reviews are the useful evidence here.
A scoping review of NGT used for survey item elicitation in health research (Journal of Clinical Epidemiology, 2021; PMID 34400255) found studies using between 1 and 41 nominal groups, with between 2 and 30 participants per group. It identified 30 distinct methodological decision points across the process and recommended that investigators document the reason for each choice in their protocol rather than treat any of them as a default.
A separate extended scoping review of virtual nominal groups (Lee SH, ten Cate O, Gottlieb M, et al. PLoS One. 2024;19(6):e0302437, doi:10.1371/journal.pone.0302437) screened 2,589 citations to 32 included articles and reported group sizes of 2–20 and session durations from 30 to 240 minutes, most clustering around 90–120 minutes.
| Parameter | Commonly cited guidance | What published studies actually do |
|---|---|---|
| Participants per group | A small group, usually given as about five to nine, with roughly ten as a practical ceiling | 2–30 per group (health-research survey elicitation review); 2–20 (virtual NGT review) |
| Number of groups | Not fixed; more groups where the population is heterogeneous | 1–41 groups per study |
| Session length | Around 90 minutes for one question | 30–240 minutes, clustering at 90–120 |
| Items each participant ranks | Typically the top five, sometimes eight or ten | Varies with list length; frequently unreported |
| Consensus threshold | Often none — NGT’s output is a ranking, not a consensus verdict | Only 16% (5/32) of virtual NGT studies pre-specified one; those used 50–75% agreement |
The practical consequence: you cannot defend a design choice here by citing convention, because there isn’t one narrow enough to cite. Pre-specify your group size, session length, voting rule and aggregation method in the protocol, state them in the methods section, and give your reason. That is what both reviews ask for, and it is what a reviewer will look for.
Note the last row especially. NGT does not inherently produce “consensus” in the sense a Delphi study means it. It produces a rank-ordered list with scores. If you need a defensible agreement threshold, you have to impose one deliberately and say so — and if you do, you have effectively borrowed a Delphi convention, so say that too.
NGT or Delphi? The decision, on cost, anonymity and speed
These two techniques were published together and are still routinely treated as interchangeable ways of “getting expert consensus”. They are not. They differ on the one axis that matters most — whether participants ever meet — and everything else follows from that.
| Dimension | Nominal group technique | Delphi |
|---|---|---|
| Contact | Participants meet, in one session | Participants never meet; asynchronous rounds |
| Anonymity | Partial. Idea generation and voting are private, but round-robin attribution is public — everyone knows who said what | Full. Responses are anonymous throughout and fed back only as an aggregate |
| Speed | Hours. A ranked output exists at the end of the session | Weeks to months, driven by round turnaround |
| Cost driver | Getting people into one room at one time — travel, scheduling, venue | Administration across rounds, plus attrition management between them |
| Geography | Constrained, unless run virtually | Unconstrained by design |
| Panel size | Small — a single facilitated table | Can run to dozens or hundreds |
| Output | A ranked list with scores, plus the clarification discussion as qualitative data | Degree of agreement against a pre-set threshold |
| Main threat | Status and dominance effects that survive the structure | Attrition across rounds, and consensus that is really fatigue |
A workable decision rule:
- Choose NGT when you need a prioritised list quickly, the group is small enough to convene, and the discussion itself has value — the clarification stage generates usable qualitative material that a Delphi never produces.
- Choose Delphi when participants are geographically dispersed, when the panel needs to be large, or when status differences are severe enough that partial anonymity will not hold. A junior clinician in a room with their head of department is not protected by a private ballot if they had to read their idea aloud first.
- Consider both in sequence — NGT to generate and prioritise a candidate item set, Delphi to test agreement on it across a wider panel. This is a common and defensible design, and it plays to each technique’s actual strength.
On thresholds, the contrast is sharp. Delphi studies overwhelmingly define consensus in advance — a systematic review of consensus definitions in published Delphi studies found percent agreement the most common criterion, with a median threshold of 75% (Diamond IR, et al. J Clin Epidemiol. 2014; PMID 24581294). NGT studies mostly do not, because a ranking does not require one. Do not import the language of consensus into an NGT write-up without importing the threshold that makes it meaningful. Full round mechanics, panel sizing and stopping rules are covered on the Delphi method guide.
Running NGT virtually
Virtual delivery is now well documented rather than improvised. The PLoS One 2024 review found Zoom used in 12 of the 18 studies that named a platform (66.6%), with Microsoft Teams and GoTo at two each — though 14 of 32 studies (43.7%) did not report the platform at all. Roughly 25% moved idea generation to an asynchronous channel such as email or an e-survey before the live session.
Author perceptions in that review were mixed but not negative: of 16 respondents, 44% judged the virtual format better than in-person, 36% comparable, and 19% inferior. The advantages reported were geographic reach, no travel cost and easier scheduling; the recurring limitation was technical difficulty persisting despite pre-meeting checks, particularly for less technically confident participants.
Three adaptations matter if you run it online:
- Silent generation needs enforcing differently. In a room, silence is self-policing. On a call, use a private text box or a timed asynchronous submission window so nobody sees anyone else’s items early.
- Round-robin needs an explicit speaking order, stated up front. Video calls have no natural turn-taking cue, and without an order the fastest unmuter dominates — the precise failure NGT exists to prevent.
- Use a polling tool for the vote, not the chat. Chat-based voting is visible to everyone and is not a private ballot.
Report the platform, whether generation was synchronous or asynchronous, and how the ballot was kept private. The 43.7% non-reporting rate above is a reporting gap you can trivially avoid being part of.
Reporting an NGT study
There is no NGT-specific reporting guideline equivalent to those catalogued on the EQUATOR Network for trials or reviews, so the reporting burden falls on the methods section. At minimum, state:
- The exact question put to the group, verbatim.
- How participants were selected and why those people — this is purposive sampling, and the sampling rationale is part of the method, not an afterthought.
- Number of groups, participants per group, and the composition of each.
- Session length, and time allowed for each stage.
- Whether it ran in person or virtually, and on what platform.
- The voting rule: how many items each person ranked, the point scale, and the direction of the scale.
- The aggregation method, and whether you report vote counts alongside totals.
- Any consensus threshold, if you set one — and that you set it in advance.
- What was merged during clarification, and on whose agreement.
- The full ranked item list, not just the top few. Truncating the list at the point where it stops supporting your argument is a real and common problem.
If the clarification discussion is being analysed as qualitative data in its own right, say how — thematic analysis and content analysis are the usual routes, and each carries its own reporting expectations. Where the ranked output feeds a subsequent instrument, the link to questionnaire design should be explicit.
Frequently asked questions
What is the nominal group technique in simple terms?
A structured meeting for generating and prioritising ideas, in which participants first write answers privately, then share them one at a time in strict rotation, then clarify them as a group, then rank them by private ballot. The ranked list is the output. It is called “nominal” because for the crucial idea-generation stage the group is a group in name only — people are present but not interacting.
What are the four steps of the nominal group technique?
Silent generation, round-robin recording, clarification, and voting or ranking. Many studies add a fifth step in which the vote is discussed and sometimes repeated; that is a documented extension rather than part of the original procedure.
How many people should be in a nominal group?
The figure usually cited is a small group of around five to nine, with about ten as a working ceiling for a single facilitated table. Published practice is much wider — 2 to 30 participants per group in one health-research review — so state and justify your own number rather than citing a rule. If you need more voices than one table holds, run multiple groups or use a Delphi.
How long does a nominal group session take?
Around 90 minutes is a reasonable plan for one question with about seven participants. Reported sessions range from 30 to 240 minutes, with most clustering between 90 and 120. Round-robin is the stage that scales with group size; budget roughly three minutes per participant for it.
What is the difference between the nominal group technique and Delphi?
NGT participants meet face to face and finish in a single session, producing a ranked list within hours; anonymity is partial because round-robin contributions are attributed. Delphi participants never meet, respond anonymously across multiple rounds over weeks or months, and the output is degree of agreement against a pre-set threshold. Choose NGT for speed and small convenable groups, Delphi for dispersed or large panels and where full anonymity is essential.
Is the nominal group technique qualitative or quantitative?
Both, which is a large part of its appeal. Stages 1 to 3 generate qualitative data — the items themselves and the clarification discussion. Stage 4 produces quantitative rank and score data. It sits alongside the other approaches surveyed in qualitative research methods, but is unusual in delivering a numeric prioritisation from the same session.
How is NGT different from a focus group?
A focus group is a moderated open discussion designed to surface interaction, disagreement and shared meaning-making. NGT deliberately suppresses interaction during idea generation and ends in a vote. Use a focus group when the interaction is the data; use NGT when you need a prioritised list and want to limit dominance effects.
How do you analyse nominal group technique results?
Aggregate the ranks into a total score per item and report it alongside the number of participants who ranked that item at all. Present the full ordered list. Where the clarification discussion is part of the analysis, code it separately using an explicit qualitative method and report that method.
Does NGT produce consensus?
Not automatically. It produces a prioritisation. Only a minority of published NGT studies define a consensus threshold in advance — 16% in one review of virtual nominal groups, which used cutoffs of 50% to 75% agreement. If you want to claim consensus, define the threshold before the vote and report it; otherwise report the ranking as a ranking.
Can NGT be run online?
Yes, and it is now well documented. Keep silent generation genuinely private with a text box or an asynchronous submission window, state an explicit speaking order for round-robin, and use a polling tool rather than the chat for the ballot so the vote stays secret.
Sources
- Delbecq AL, Van de Ven AH. A Group Process Model for Problem Identification and Program Planning. Journal of Applied Behavioral Science. 1971;7(4). doi:10.1177/002188637100700404 — the original formulation.
- Delbecq AL, Van de Ven AH, Gustafson DH. Group Techniques for Program Planning: A Guide to Nominal Group and Delphi Processes. Scott, Foresman; 1975 — the full procedural treatment, and the source of the NGT/Delphi pairing.
- Methodological options of the nominal group technique for survey item elicitation in health research: a scoping review. Journal of Clinical Epidemiology. 2021. PMID 34400255 — 30 methodological decision points; group and participant number ranges.
- Lee SH, ten Cate O, Gottlieb M, et al. The use of virtual nominal groups in healthcare research: an extended scoping review. PLoS One. 2024;19(6):e0302437. doi:10.1371/journal.pone.0302437 — 32 included studies; durations, platforms, consensus-threshold reporting.
- Diamond IR, Grant RC, Feldman BM, et al. Defining consensus: a systematic review recommends methodologic criteria for reporting of Delphi studies. Journal of Clinical Epidemiology. 2014. PMID 24581294 — the 75% median agreement threshold used for the Delphi contrast.
Related CASRAI resources
- The Delphi Method: Building Expert Consensus in Rounds — the asynchronous, fully anonymous alternative, with round structure and stopping rules.
- Focus Groups in Research: Design, Moderation, and Analysis — when the interaction itself is the data.
- Qualitative Research Methods — where consensus methods sit among the major methodologies.
- Purposive Sampling — how to justify who is in the room.
- Thematic Analysis — for analysing the clarification discussion.
- Social Desirability Bias — the standing threat to any face-to-face method.
- Publishing and evidence synthesis — the wider cluster this guide belongs to.
