Reporting preferences, abstentions, and consensus in Delphi studies
The option receiving the most votes may be the preferred approach while falling short of the study’s consensus threshold. Show the support, denominator, and decision rule so readers can see exactly what was established.
Define what the question is intended to decide
A preference question identifies which option attracts the most support. A consensus rule asks whether a specified level and pattern of agreement has been achieved. A study can ask both questions and report different answers.
Specify whether panelists select one option, rank alternatives, or rate each independently. The results have different meanings. If respondents may select multiple options, percentages need not sum to 100%, and the most-selected choice is not necessarily an exclusive preference.
Use separate report fields for the leading option and whether the predefined consensus rule was met. If no consensus rule was specified for that question, say so. Also define how ties will be handled.
One result can have several legitimate denominators
Consider an original example with 30 eligible panelists. Fifteen choose A, five choose B, two choose C, two select “not my expertise,” and six provide no item response. There are 22 substantive choices and 24 recorded item responses.
| Denominator | Support for A | What it describes |
|---|---|---|
| 22 substantive choices | 15/22 = 68.2% | Preference among those choosing A, B, or C. |
| 24 recorded responses | 15/24 = 62.5% | Support among item responders, including explicit non-ratings. |
| 30 eligible panelists | 15/30 = 50.0% | Support across the eligible panel, including missing responses. |
If the prespecified rule requires at least 75% of substantive choices, A is the preferred option but does not achieve consensus: it needs at least 17 of 22 votes. The other two percentages describe coverage; they do not replace the agreed classification denominator.
Keep non-ratings distinct from disagreement and missingness
“Not my expertise,” “I do not know,” and “I abstain” may express different reasons for withholding a judgment. Use clear labels and explain their treatment in the instructions and analysis plan. A skipped question is another category.
A midpoint rating can represent a substantive judgment of uncertainty or moderate agreement. Do not automatically turn it into a non-rating. Equally, do not recode an explicit non-rating as zero or as the scale midpoint.
Report non-ratings by item and round. Where feasible and consistent with confidentiality, examine whether they cluster within a stakeholder group. If every response is excluded under the approved rule, report no evaluable result rather than zero support.
A recent study separates preferred approaches from fixed thresholds
A 2026 economic-evaluation study of CAR T therapies combined an evidence review, three Delphi rounds, and clinical sense-checking. Round 1 used a 70% rule to retain issues considered specific to CAR T appraisals. For preferred solutions, the authors specified no numerical consensus threshold and excluded “no opinion” or insufficient-expertise responses from calculations; one leading costing option received 61% support. Read the methods and results.
The useful reporting distinction is between selecting a leading approach and satisfying a predefined consensus rule. Describe the purpose and rule for each stage, rather than transferring one stage’s threshold to every later question.
Show the complete preference profile
Report every option’s count and percentage, not only the winner. State the response format, eligible panel size, valid-choice denominator, explicit non-ratings, missing answers, round, and exact decision rule.
Illustrative result
“Option A was preferred by 15 of 22 panelists providing a substantive choice (68.2%). It did not meet the prespecified 75% consensus threshold. Two panelists selected ‘not my expertise’ and six gave no item response.”
If the team recommends A anyway, describe that recommendation and its rationale separately from the panel’s classification. Preserve minority explanations and state any additional committee review or external evidence used.
Build the labels and exclusions into the study workflow
In Surveylet, use the question format and response labels appropriate to the protocol. Where an explicit non-rating is offered, check its treatment in the relevant calculation and export. Review a worked example containing substantive choices, a non-rating, and a missing answer before launch.
Keep an analysis record that distinguishes the preferred option, consensus status, and any later steering-group decision. Define any custom table, denominator reconciliation, or reporting labels when agreeing the analysis scope.
Complete Survey Care can help with questionnaire configuration. The Delphi Analysis & Final Report service can be scoped around the team’s approved reporting requirements.
Source and scope
Methodological Challenges and Preferred Approaches for the Economic Evaluation of CAR T Therapies: Evidence Review and Expert Consensus from NICE Appraisals. PharmacoEconomics; October 9, 2026.
The numerical example and workflow suggestions are Calibrum’s original illustrations. They demonstrate transparent labeling and denominator handling; they do not establish a universal consensus threshold or exclusion policy.
Make every result interpretable
Plan the response options, exclusions, and reporting labels together so the final report answers the question your study set out to ask.