Glossary

Terms as they appear in Delphi Studio, plus a few from the consensus literature. Bold marks a label you can see on screen. A scientific term that is not a control is in plain type. This is a word list. A formula, a freeze list, or a procedure lives on the linked page, which is the authority if the two ever differ.

A–C

Term	Definition
Advance, Revise, Hold	Dispositions for a round that has another round after it. Advance: rated again as written. Revise: reworded in the Item bank, then rated again. Hold: not rated again, because the statement is settled in or out, or retired. "Hold means settled: not rated again." These three words are not the label on the last planned round. An uncertain or polarized statement is not Revise. While another round remains, the rules send it on as written (Advance), and the plan tile reads Rated again. On the last planned round the Decision stage groups that outcome under Final: no consensus. The shorter label No consensus is not that tile. It is not the only last-round result. See Interpreting results.
Agreement	A share of numeric ratings in a zone. It is not Kendall's W, and it is not a RAND/UCLA class by itself. See Consensus analytics.
Amendment	A recorded change after launch, with a written reason. What an amendment can change, and what it cannot, is in Launch and amendments. This entry does not add another list.
Anonymity	In this product, separation of identity from responses. Peers do not see peers. It is not a promise that every download omits a name or an email address. See Identity and pseudonyms.
Answer removal	A request, made when leaving the study, to have the answers already given removed. Leaving alone keeps answers. In a round not yet finalized the answers are deleted at once and the results are worked out again. For a finalized round the principal investigator approves or declines, and an approved removal creates a new results version. The earlier one stays on record.
Appropriateness	In a RAND/UCLA study, a class from the panel median and a disagreement test. It is not the same word as Retain. In this product a median of exactly 6.5 is uncertain. Whether that boundary should change is open. See RAND/UCLA.
Area	A heading that groups statements in the Item bank, on the panelist's page, and in results. An area may have one level of sub-areas.
Attrition	Panelists lost across rounds. Size & mix takes a planned attrition rate and sets the enrollment target to cover it. Response rates per round appear on the round header and in the participation report.
Codebook	The set of codes the coders apply to comments. The principal investigator and the Adjudicator manage it.
Coding	Comment coding by two coders working blind to each other, with an adjudicator who settles disagreements. Pooled kappa is reported against the study's minimum and never stops a round or changes a result. See Coding and adjudication.
Comment approval	How comments are cleared before they reach controlled feedback: Review every comment, or Approve when nothing identifying is found (the default).
Comment sharing	What panelists are told happens to their comments, recorded per study and changeable per round: verbatim, de-identified, edited, summarized, or withheld. With no recorded choice, no comments are shared.
Concordance	How similarly people ordered the statements in one round. That is what Kendall's W measures. It is not item-level consensus, and it is not stability.
Consensus	The rule the study stated in advance for what counts as settled, or the authors' conclusion that the rule was met. The software applies the study's rule. It does not discover that consensus occurred, and reaching it is not guaranteed. It is not Kendall's W, not stability, and not appropriateness.
Controlled feedback	What panelists see about the group before they rate again. The live block can show the group median, the middle half of the ratings, the three zone shares, the person's own earlier rating, and comments the team chose to share. The introduction defines the word Status. That sentence is not a Status row, and the block does not show the classification. It 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. Feedback shows the previous round's current verified results. It is not a frozen copy.
CREDES, ACCORD	Two reporting guidelines. ACCORD is the broader guideline for consensus methods in biomedicine. It guides the report. It does not prescribe the design, and a record in this software does not by itself make a study compliant. CREDES (Guidance on Conducting and REporting DElphi Studies, Jünger et al., Palliat Med 2017) is Delphi-specific guidance developed in palliative care. A study can record which guideline it follows. See the ACCORD and CREDES crosswalk.

D–I

Term	Definition
Decision log	The append-only, hash-chained record of each round's confirmed decision plan, including overrides and steering decisions.
Deferred for clarity review	Under the protocol rule set only: a statement that too many responders were unable to rate. It goes to clarity review and is not given another status by the classifier.
Dimension	A rating axis. A new study has one, Importance. A study can add more, up to five, and names one as primary. The primary dimension is the one classified.
Disagreement	In a modified Delphi study, the spread or polarization rule that study selected. In a RAND/UCLA study, the disagreement test that study selected. The two tests are not the same, and neither is Kendall's W.
Disagreement index (DI)	The product's stored spread measure on a 1–9 scale. It is not the RAND/UCLA disagreement test. The results table does not show it. A study can select it, instead of the interquartile range, as the spread test for a modified Delphi classification. The formula is on Consensus analytics.
Expert directory	The organization's shared list of experts, used to recruit panelists across studies. It keeps each expert's future-contact answer separately from any study consent.
Final results	After the last round, each statement gets one of five names: Final: retained, Final: excluded, Final: retired, Final: no consensus, or Final: deferred, still unclear. These are not the same three disposition words. The Decision-stage group for an unsettled statement that did not come from a clarity review is Final: no consensus. The shorter label No consensus is not that tile. It is not the only final result.
Framework originator	A staff-only yes or no on each panelist. Each round reports how many of its responders are originators, which must not be a majority. It is reported, not enforced.
Grace window	Optional minutes (0 to 1,440) after a round's deadline during which answers are still accepted. The deadline plus the grace window is the cutoff, after which saves are refused.
Group disagreement	A status: the statement would be retained, but a stakeholder group with enough panelists opposes it by more than the set limit. Applies only when the group check is on.
Interpercentile range	On a 1–9 scale, the width the IPRAS test compares with its limit: the 70th percentile minus the 30th percentile, using the same type 7 method as the quartiles. It is not the interquartile range, and it is not the disagreement index. The comparison stays on Consensus analytics.
Invitation	A message the study can queue for a person. A queue row is not proof of delivery, and it is not enrollment. Creating a draft does not send one. Launch does not send one.
IPRAS	The RAND/UCLA disagreement limit used when the study selects that test. Disagreement means the interpercentile range is greater than the limit. A range equal to the limit is not disagreement. IPRAS is not the disagreement index. The constants, and the open question about a median of 6.5, stay on RAND/UCLA and Consensus analytics.
IQR	Interquartile range: the type 7 third quartile minus the type 7 first quartile. Another quartile convention can change the number. It is the spread measure a new study starts with. A new study's seed asks for an IQR of at most 2 on the 1–9 scale. That seed is not a universal standard, and this entry does not say whether a study should adopt it. See Consensus analytics.
Item bank	The study's statements, grouped by area. Statements can be added by hand or imported from a CSV file with a preview. See Round design.

K–P

Term	Definition
Kendall's W	Kendall's coefficient of concordance. It measures how similarly the panel ordered the statements in one round. It is not an item-consensus test, and it is not stability. Delphi Studio computes it with the tie correction and shows it. No rule uses it. No p-value is reported. When W can be computed, it is between 0 and 1. When it cannot, W is undefined, not zero. The results page uses four exclusive labels for how the panel ordered the statements: 0.7 or more, strong agreement; 0.5 to under 0.7, moderate agreement; 0.3 to under 0.5, weak agreement; under 0.3, very weak agreement. Those are the words on the screen. They are not a validated scale, and they are not Landis and Koch's kappa benchmarks. This entry does not cite a paper as their source, and it does not say whether a manuscript should quote them. The description to follow is Kendall's W.
Launch	The move from draft to active, after the ready steps pass. It creates the planned rounds. It does not send an invitation, and it does not open a round. It does not freeze every study rule. See Launch and amendments and Review and launch.
Locked at launch	Not every configuration field becomes immutable at launch. A control the screen hides is not the same thing as a save the server refuses. What launch freezes, what stays editable, and what a recorded amendment can still change are in Launch and amendments.
Opening a round	Making a round accept answers after the principal investigator has approved it. The principal investigator or the coordinator may open it. Opening is not launch, and it is not an invitation.
Override	A decision that leads somewhere other than the rule outcome. It needs a reason of at least 20 characters that names no one, and it is written to the decision log.
Panelist status	Where a person stands on the panel, in order of precedence: Removed, Withdrawn, Application declined, Declined invitation, Enrolled, Needs review, Invited, Approved, Applied. Accepting an invitation is not consenting, and consenting is not enrolling.
Polarized	A status: neither retained nor excluded, with at least 30% of ratings in each extreme zone and at most 20% in the middle, from at least 6 ratings. Those figures are the new-study seeds, not a universal standard. This entry does not say whether a study should adopt them. Polarized is not Revise. Under the protocol rule set the same pattern is a flag instead of a status.
Pseudonym	A stable, study-scoped label for an enrolled panelist, such as ABC-0004. It is not derived from the name. Panelists do not see it. On ordinary study screens after enrollment, staff see the pseudonym instead of the name. Showing the name takes an audited reveal with a written reason. Some downloads can still contain a name and an email address. See Identity and pseudonyms.

Q–Z

Term	Definition
Questionnaire round	A round of questions rather than ratings, usually Round 1. In a study with structural rules, its results go to the steering committee's structural gate decision.
Rating round	A round in which panelists rate statements on the study's scale.
Results version	The sealed, stored results of a round, computed after collection closes. A new version replaces the current one only with a reason, and the seal is checked on every read.
Retain, Exclude, Uncertain	The main modified-Delphi statuses. Retain: the study's retain conditions are all met, including enough ratings in the agree zone and a narrow enough spread. Exclude: the same pattern in the disagree zone. Uncertain: neither, and no other status applies. A narrow spread is not itself consensus. A RAND/UCLA study uses Appropriate, Inappropriate, and Uncertain instead. Those words are a different classification. Anchors on a new study do not say Retain or Exclude. Panelists are not shown a Status row or the classification label.
Round states	Planned, Building, Open, Reviewing, Decision, Closed. See Key concepts.
Size & mix	The Setup step that sets the panel's target and its groups. Each group is core or context. Context groups are reported but kept out of some pass-or-fail checks.
Smallest reportable group	The floor from the Analysis plan. Any result based on fewer people is withheld, with no figures, on every screen and in every file. The default floor is 5. This is separate from the question of which downloads can contain a name.
Source	A reference on a statement: a DOI, a PMID, a web address, or a free-text reference.
Statement	One thing the panel rates. Each has a permanent ID, such as S07, that is never reused and survives every revision. A revision shows as "S12 v2".
Stability	How much ratings change between rounds. It can be set on Rating & consensus. Nothing computes it. The screen says "Not computed yet". It is not Kendall's W.
Steering review	A result the rules send to the steering committee to decide, for example an exclude that a group check flags.
Stopping advice	An optional suggestion to wrap up once a share of statements is settled (90% by default). It is advice only and never moves a round.
Structural gate	In a study with structural rules, the steering committee's recorded decision after the questionnaire round: proceed, revise naming or boundaries, re-map the composition, or stop. The next round cannot open without it.
Subgroup comparison	Results by group. It can be set up. It is not computed.
Suggestions	New statements or rewordings proposed by panelists, when the round allows them. They never change the round in progress.
Thresholds table	One read-only table, on Review & launch and on each round's Review tab, listing every number the study uses and whether it is enforced, reported only, advice only, not applied yet, or off. See Thresholds.
Too few responses	The status when there are fewer numeric ratings than the study's minimum. The classifier returns this before the other classes. Nothing else is decided. A new study starts at 7. That minimum is a seed for one statement, not a universal panel size, and this status is not the same sentence as "counts as uncertain".
Unable to rate	An answer meaning the panelist cannot judge the statement. It is left out of every percentage, median, and interquartile range, and the count is shown beside the results.
Uncertainty	The situation in which the study's rule did not settle the statement. On a modified Delphi classification the status is Uncertain. It is not Too few responses, and it is not Revise. A narrow interquartile range does not by itself mean consensus.
Withdrawal	A panelist's decision to leave, effective at once. It stops their email and their links and ends their part in rounds still collecting or not yet open. Their answers in closed rounds are kept. Only the panelist can undo it, by returning through their withdrawal link.
Your round link	Each enrolled panelist's standing link to their rounds. It is revoked when they withdraw or are removed. A lost link is resent by email, never shown. Staff sign in with a separate emailed link.
Zone	One of three bands of the scale: Disagree, Uncertain, and Agree. On the 1–9 scale they are 1–3, 4–6, and 7–9. The zone named Uncertain is a band of the scale. It is not the same thing as the status Uncertain, though ratings in that band feed the status.

Related

• Key concepts: how these terms connect in a study
• Modified e-Delphi: the design and the new-study seeds
• What is the Delphi method?: the research process
• Consensus analytics: measures, including the disagreement index and Kendall's W
• Governed study record: what launch changes, and which downloads can contain a name