What is the Delphi method?

The Delphi method is a structured way to collect and reconsider expert judgment. Panelists answer in private. The research team returns an explicit summary of the group's answers, not each person's response. Panelists may then answer again. The study states its stopping rule before collection starts. That rule may be a maximum number of rounds, a stability rule the team assesses by hand, or both.

Agreement is one possible result. It is not guaranteed, and it is not the only legitimate result. A study can also identify persistent disagreement or uncertainty. The method is used when a question needs structured expert judgment. It is not a single survey, and it is not an unstructured meeting in which people argue in the same room.

Modified Delphi is an umbrella for designs that change some part of that procedure, such as starting from a prepared item set or revising items between rounds. It is not one protocol. Modified e-Delphi describes how Delphi Studio runs that family of designs on the web.

Delphi Studio is software for recording the process. It is not the method. A study's Setup records the protocol choices before data collection. Each round opens only after the principal investigator approves its content, rules, consent form, and schedule. What launch freezes, and what a recorded amendment can still change, is in Launch and amendments. This page explains the method. It is not the procedure for creating a study. To create a draft, use Start a new study.

Origins and purpose

Delphi is used when no single expert has the whole answer, and when a direct exchange could let one voice dominate. Experts provide private ratings and comments. They receive structured feedback that does not name who said what. They may rate again. The study stops when its pre-stated rule says to stop. The software does not decide that the panel has agreed.

You design the rules, choose what feedback to share, interpret disagreement, and decide what the result means. The software applies the study's rules. It does not replace that judgment.

The iterative Delphi process

A typical modified Delphi study in health research follows these stages. The round count and the item source are design choices, not a universal standard.

Preparation (pre-Round 1). A review, a prior study, a framework, interviews, a steering group, or an exploratory round can produce the first items. Some designs instead ask the panel to propose the items in an open first round. Panelists are identified with eligibility criteria that fit the question. In Delphi Studio, statements go into the Item bank, typed in or imported from a CSV file with a preview, and eligibility and panel targets are set on Eligibility and Size & mix.
Round 1. Panelists independently rate each item, often on an ordinal scale for importance, agreement, or appropriateness, and they may comment. An explicit "unable to rate" option is part of a careful design. In this software that option can be offered. Ratings and comments stay private to each panelist. Peers do not see who answered.
Analysis and controlled feedback. The team summarizes the round. Common summaries are the median, the interquartile range, and the share of ratings in pre-stated zones, plus a reading of the comments. Panelists then see only the feedback the design allows, including their own earlier answer when the design includes it. Delphi Studio can show the group median, the middle half of the ratings (the first quartile to the third), the three zone shares, a Status label, the person's own earlier rating, and comments the team chose to share. That block does not print the label IQR, and it does not draw a histogram. A reworded statement withholds the group figures, the earlier rating, and the comments.
Later rounds. Panelists read the feedback and may rate again. The team may revise an item. Revision is a researcher decision. An uncertain result does not by itself require a new wording. In Delphi Studio a statement keeps its permanent ID through every revision, and a revision cites the results it responds to. See Modified e-Delphi.
Stopping and classification. The study stops under the rule it stated in advance: a round limit, a stability judgment the team makes by hand, a target share of items that meet the study's own definition, or a combination. Each item is then classified. Consensus in, consensus out, and no consensus are different results. Delphi Studio does not measure a change in the median or the interquartile range between rounds. Stability can be configured and is not computed.

After collection closes, Delphi Studio computes each statement's results from the submitted answers with the study's confirmed rules and stores them as a sealed results version. See Consensus analytics. Group feedback for the next round is built from those results under the study's feedback settings. You remain responsible for reviewing comments, revising items, recording an override with a reason, and deciding when to stop. The stopping recommendation in the software is advice only. It does not move a round.

Why researchers use the Delphi method

Strength	Explanation	What Delphi Studio does
Limits dominance in the exchange	Private answers reduce the chance that one voice steers a live discussion. This is not a promise of absolute anonymity.	Peers do not see peer names or individual answers. After enrollment, staff see a pseudonym on ordinary screens. Revealing a name needs a written reason and is audited. Some downloads can still contain a name and an email address. See Identity and pseudonyms.
Geographic and temporal reach	Experts can answer at different times and places.	Web pages reached by a personal link, email invitations and reminders, and round-by-round participation counts. Launch does not send the invitation.
Rules stated before collection	The consensus rule and the stopping rule are declared before data collection. A particular percentage, spread, panel size, or round count is not a universal standard.	Setup asks you to confirm the scale, zones, and consensus rules before launch. What that launch freezes is in Launch and amendments.
A record that can be checked	Inputs, decisions, and outputs can be documented. The record does not by itself make a manuscript complete.	Sealed results versions, permanent statement IDs, a decision log, a study audit, and CSV exports. Investigators still write the rationale and check every manuscript sentence.
Room for a large item set	The method can cover many statements, such as guidelines, curricula, or quality indicators.	An item bank grouped by area, with recorded sources, and comment coding.

Classical versus modified Delphi

The supported contrast is how the items begin. Classical designs often open with a round that gathers the panel's own items. Modified designs often start from a prepared set and a planned number of rounds. Modified Delphi remains an umbrella. Round counts, a literature review, and "tighter governance" do not define it.

Aspect	Classical Delphi	Modified Delphi	In Delphi Studio
How items begin	Often an open first round that gathers items from the panel.	Often a prepared item set. The label is not one protocol.	Classical Delphi pairs with an open first round. Modified Delphi can rate a prepared bank. Choosing the method does not change the scale.
Rounds	The study states its own plan. A particular count is not a standard.	The same. A planned number of rounds is common, not required.	"Rate existing items", "Generate items, then rate", and "Define and refine, then rate" start from 3 rounds. "Build a custom workflow" starts from one rating round. The count is a seed. What can change after launch is in Launch and amendments.
Feedback	An explicit group summary, sometimes with comments. Direct exchange is avoided.	The same idea, with whatever changes the protocol states. Some modified designs add a meeting.	The group median, the middle half of the ratings, the three zone shares, the Status label, the person's own earlier rating, and comments the team chose to share. Each can be switched off. The panelist view does not draw a histogram and does not print the label IQR. A reword withholds the group figures, the earlier rating, and the comments.
Identity	Often anonymity among panelists, which the report should define.	The same duty: say to whom anonymity applies. It is not a promise that no file contains a name.	Peers never see peer identities or individual responses. Staff see pseudonyms after enrollment. See Identity and pseudonyms.
Item revision	Items may be reworded between rounds.	Revision is tracked when the protocol says so.	A statement keeps its permanent ID, shown as "S12 v2". Each revision cites the round results it responds to.

The Method step marks Modified Delphi as the recommended preset: independent responding, controlled feedback, and tracked item revision. That label is not a scientific requirement to use the preset, and it is not a complete protocol. State what this study modified, and why. What launch freezes is in Launch and amendments.

What the Delphi method is — and is not

The Delphi method is:

A structured procedure for expert judgment on statements, scenarios, competencies, or indicators. Agreement is not guaranteed.
Iterative by design. One survey is not a Delphi study. Convergence is not assumed at the end.
Reportable when the paper states the panel, the scale, the study's own consensus rule, attrition, the feedback, and how the team judged stopping. Guidance such as CREDES does not replace that account. CREDES was developed for palliative-care Delphi studies.

The Delphi method is not:

A substitute for a systematic review or a meta-analysis. A Delphi study often follows one.
A simple vote. The study's pre-specified rule defines what counts as settled. Plurality is not that rule, and no single percentage is a universal standard.
Defined by the absence of discussion. Classic Delphi avoids direct exchange between panelists. Some modified designs add a meeting. This application does not provide a meeting or a live chat.

Whether a study that omits the RAND/UCLA meeting may still use that label is open. This page does not decide it. See RAND/UCLA.

Key methodological commitments for peer-reviewed Delphi studies

Reviewers expect the design to be explicit. Delphi Studio records many of the fields. You still supply the rationale and the manuscript context.

Panel composition. How experts were identified, which perspectives are represented, and the response rates. Delphi Studio records eligibility, Size & mix groups, profile fields, and per-round participation counts.
Rating scale. Anchors, the numeric range, and how "unable to rate" is handled. Delphi Studio offers a 1–9 scale (the default for a new study, and the scale RAND/UCLA uses), Likert 1–5, Likert 1–7, 0–10, and custom scales. "Unable to rate" is left out of the denominator, and its count is still shown.
Consensus definition. The study's own rule, stated before collection. It may combine a center (often the median), a spread (often the interquartile range), and a share of ratings in defined zones. Those pieces are not a universal formula. Rules are set on Rating & consensus. What launch freezes is in Launch and amendments. The rules in force are applied when each round's results are computed. See Consensus analytics.
Stopping criteria. A maximum number of rounds, a stability judgment, or a share of items that meet the study's rule. These are separate from the item rule. Delphi Studio records a stopping recommendation (by default, at least 90% of statements settled) as advice only. Stability between rounds can be configured and is not computed, so a protocol that relies on it must assess it by hand.
Feedback content. What panelists receive between rounds. The study's feedback settings are chosen in Setup. The comment-sharing choice is recorded at each round's decision. With no choice recorded, no comments are shown. See Participant feedback.
Attrition and missing data. Who dropped out, and whether non-responders were carried forward. The participation report gives per-round counts (in the round, submitted, withdrew, answers removed) and the response rate. Results add a note at or below the study's warning level, and again below its lower reporting level. Nothing is blocked by either note.
A checkable record. The protocol, the data, and the analytic decisions. What launch freezes is on the governed study record. Each round's approval is tied to its content, rules, and consent version. Each results version is sealed. The decision log and the study audit are on that same page, including what the audit page does when the chain does not verify. That page does not treat the audit as a guarantee that every action is blocked.

Method	Best suited for	Preset in Delphi Studio
Modified e-Delphi	Agreement on statements, competencies, quality indicators, or policy options	Modified Delphi (marked recommended on the Method step)
Classical Delphi	An open first round that gathers the panel's own items before rating	Classical Delphi
RAND/UCLA Appropriateness Method	Appropriateness of clinical actions under defined indications	RAND/UCLA Appropriateness
Custom	Designs the presets do not cover	Custom method

These are the choices on the Method step. Read Comparing Delphi types before you choose. The preset does not settle the open method choices on the RAND page.

Ensuring reporting quality and auditability

For biomedical consensus research, ACCORD is a reporting guideline. It guides what a paper reports. It does not prescribe how the exercise must be run, and the software does not make a study compliant. CREDES is Delphi-specific guidance developed in palliative care. See the ACCORD and CREDES crosswalk.

Reporting topic	What the record holds
Panel selection	Eligibility, Size & mix groups, profile fields, and recruitment and participation counts
Consensus process	The rules the study confirmed. What launch freezes, and the rules each results version used, are on the governed study record and in Consensus analytics.
Feedback	The study's feedback settings and each round's comment-sharing choice
Attrition	The participation report and its CSV
Reproducible analysis	The responses, consensus, and participation CSVs, and the study audit CSV. See Reporting.

These files can support a methods and results section. They are not the manuscript.

Origins and purpose​

The iterative Delphi process​

Why researchers use the Delphi method​

Classical versus modified Delphi​

What the Delphi method is — and is not​

Key methodological commitments for peer-reviewed Delphi studies​

Related methods available in Delphi Studio​

Ensuring reporting quality and auditability​

Further reading​