How consensus thresholds work in small Delphi panels
A 75% threshold sounds straightforward. With seven valid ratings, it requires six supporting votes—85.7%. Convert percentages into whole votes before launch so the study team knows exactly what its rules require.
A percentage becomes a whole number of people
For a rule requiring at least a specified percentage, multiply the valid-rating count by that percentage. Round up if the result is not already a whole number.
Seven valid ratings, at least 75% support
7 × 0.75 = 5.25, so the minimum is 6 supporting votes. Five of seven is 71.4%; six of seven is 85.7%. There is no possible result between those two percentages.
This is the arithmetic consequence of applying the rule. The protocol threshold remains 75%; the smallest attainable passing result happens to be 85.7% for this denominator.
Check the practical effect of your threshold
The examples below assume one equally weighted vote per person and an inclusive boundary: support must be greater than or equal to the specified percentage. They show the support requirement only; additional disagreement or subgroup rules must also be met.
| Valid ratings | At least 75% | At least 80% |
|---|---|---|
| 6 | 5 of 6 (83.3%) | 5 of 6 (83.3%) |
| 7 | 6 of 7 (85.7%) | 6 of 7 (85.7%) |
| 8 | 6 of 8 (75.0%) | 7 of 8 (87.5%) |
| 9 | 7 of 9 (77.8%) | 8 of 9 (88.9%) |
| 10 | 8 of 10 (80.0%) | 8 of 10 (80.0%) |
| 12 | 9 of 12 (75.0%) | 10 of 12 (83.3%) |
| 15 | 12 of 15 (80.0%) | 12 of 15 (80.0%) |
| 20 | 15 of 20 (75.0%) | 16 of 20 (80.0%) |
Different thresholds sometimes require the same votes. They can separate again when participation changes, so choose the threshold for its methodological rationale and check several plausible item counts.
Use the denominator specified in the protocol
An invited panel of 20 does not guarantee 20 valid ratings for every item. A participant may skip an item, select “not my expertise,” or leave the study. Specify whether the rule uses valid item ratings, all eligible panelists, or another justified denominator.
For an illustrative valid-rating rule, five supporting votes out of seven fail 75%. If one non-supporting panelist does not rate the item in the next round, the same five supporting votes out of six pass at 83.3%. No additional person endorsed the statement. Report the participation change alongside the classification.
Track explicit non-ratings separately from missing answers. Define any minimum valid count and any stakeholder-group requirements before launch. If no valid ratings remain, report the item as not evaluable under that rule; the percentage is undefined.
Write the boundary exactly—and classify before rounding
“At least 75%” includes six of eight. “More than 75%” requires seven of eight. For a strict greater-than rule, take the whole-number part of the threshold multiplied by the denominator, then add one.
Apply the rule to the underlying counts or full-precision result. Round only the displayed percentage. If the protocol also limits low ratings or requires agreement within each stakeholder group, evaluate each condition separately.
A method label does not specify the calculation
An October 2026 chronic-insomnia study used a modified RAND/UCLA process with seven experts. It defined acceptance as at least 75% of ratings in the 7–9 band and explicitly explained that six of seven votes were needed. Read the study’s rating procedure.
Describe the actual rule when citing this example. A percentage-in-band rule is a specified adaptation; it should not be presented as interchangeable with the median and disagreement calculations used in other RAM designs. Consult the RAND/UCLA manual and report which approach your protocol adopts.
Report counts, percentages, and the full decision rule
A useful item record includes the round, wording version, valid count, supporting count, exact comparison, displayed percentage, non-ratings, missing answers, and final classification. Include disagreement and subgroup results when required.
Illustrative reporting sentence
“Six of seven valid ratings were in the 7–9 band (85.7%), meeting the prespecified requirement of at least 75%. No explicit non-ratings or missing item responses were recorded.”
In Surveylet, configure the agreed scale and consensus settings, then verify the relevant outputs against worked examples before inviting panelists. Include cases at the exact boundary and cases with non-ratings. The tables in this article are planning examples; confirm any custom vote-count display or report separately.
Sources and scope
- Comprehensive management of chronic insomnia in Spain: a multidisciplinary modified RAND/UCLA consensus with recommendations on daridorexant. Frontiers in Psychiatry; October 13, 2026.
- Fitch K, et al. The RAND/UCLA Appropriateness Method User’s Manual. RAND; 2001.
The vote table and reporting example are Calibrum’s original illustrations. They explain threshold arithmetic and do not prescribe a universal panel size or consensus definition.
Make the decision rule clear before launch
Bring your protocol, likely panel counts, and reporting requirements. We can help configure the Surveylet study and check how the rules behave.