ACADEMIC WRITING GUIDE

How to Conduct and Report a Delphi Study in a Dissertation (With Worked Examples)

By Acadelyra Editorial TeamPublished

A dissertation researcher can ask experienced professionals what a good policy should contain and still receive incompatible answers. One practitioner prioritises fast decisions; another worries about false alarms; a third points out that a technically sound warning may never reach residents who need it. A single questionnaire records those differences. A Delphi study gives the panel a structured opportunity to reconsider them after receiving carefully prepared feedback.

A Delphi study is an iterative process for eliciting and examining informed judgments from a deliberately selected panel. Its defining work happens between rounds: responses are analysed, the range of views is returned to participants, and individuals can retain or revise their judgments. The researcher must explain both how agreement was assessed and what the remaining disagreement means.

This guide develops a new fictional study of community flood-warning readiness. All professionals, recruitment events, comments and ratings are synthetic teaching materials. They are not evidence about actual municipalities or a validated emergency-management checklist. The participant-level ratings are printed below so that the numerical example can be reproduced.

What makes a study a Delphi?

The early work of Dalkey and Helmer described repeated questionnaires interspersed with controlled feedback. Contemporary Delphi designs vary, but a defensible dissertation should identify the panel, the iterative process, the feedback participants received, and the extent to which individual responses were anonymous to other panellists.

The familiar sequence is: define the decision problem; recruit relevant expertise; elicit or present candidate statements; analyse responses; return feedback; invite independent reconsideration; and report the resulting agreement and dissent. Simply sending a questionnaire to experts does not demonstrate that this process occurred.

Hasson, Keeney and McKenna (2000) provide foundational guidance on preparation, conduct and difficulties. Their later 2025 discussion shows that definitions, panel selection and consensus criteria remain contested. Describe what you actually did instead of relying on the name of a variant to supply missing methodological detail.

Decide whether expert judgment answers your question

Delphi is useful when the intended product requires informed judgment: identifying priorities, specifying proposed competencies, evaluating candidate criteria, forecasting developments, or clarifying conditions under which experts support a recommendation. It can organise knowledge where published evidence is incomplete or where evidence alone cannot settle competing priorities.

A suitable dissertation question might ask which proposed indicators a defined panel considers essential for assessing readiness in a specified setting. An unsuitable substitution would ask whether a warning system reduces deaths and then treat panel agreement as the effect estimate. That question needs evidence about actual outcomes and an appropriate empirical design.

Before choosing Delphi, state the decision, the intended users and the kind of uncertainty expert deliberation can address. The broader guide to choosing a research methodology helps establish that fit; the distinctive task here is designing the panel's repeated judgments.

Choosing among Delphi and other ways of collecting expert perspectives
MethodMain contributionWhat to consider for this dissertation
DelphiIndependent judgments revisited after structured feedbackUse when the process of reconsideration and the pattern of agreement or dissent matter.
Expert interviewsDetailed individual accounts, reasoning and contextUse when understanding experiences and explanations is the main task; interviews can also inform Delphi item development.
One-time surveyResponses to a fixed instrument at one pointUse when you need a distribution of answers without a feedback and re-rating process; population claims require an appropriate sampling design.
Focus groupInteraction through which participants discuss an issueUse when discussion itself is valuable; a visible group discussion has different anonymity and status dynamics.

These methods can complement each other. Calling a focus group a Delphi round does not explain how feedback was controlled or individual ratings protected.

Choose and describe the Delphi variation

A traditional or classical Delphi commonly starts with open questions. Researchers convert the resulting material into statements for subsequent assessment. A modified Delphi may begin with statements drawn from literature, prior interviews or an existing framework. This can reduce initial burden but can also constrain which issues enter the discussion. Leave room to challenge omissions.

An e-Delphi describes electronic administration; it can use an open or structured opening. A real-time Delphi allows reconsideration as feedback becomes available rather than separating every exchange into scheduled rounds. A policy Delphi may emphasise the arguments behind contrasting positions. The RAND/UCLA Appropriateness Method is a specific structured approach to judging appropriateness; do not borrow its name or decision rules for an unrelated rating exercise without implementing and justifying them.

Hsu and Sandford (2007) explain open and structured beginnings. The labels do not replace a protocol. State the source of initial items, mode of contact, feedback format, planned rounds, revision process and decision rules. The fictional study below uses an open first round followed by scheduled online ratings.

Define expertise and recruit a defensible panel

Expertise is relative to the question. Seniority, publications or a prestigious job title may be relevant, but none automatically establishes knowledge of the specific decision. Define eligibility in observable terms and explain how it will be checked. Depending on the purpose, relevant expertise may include technical knowledge, implementation experience or lived experience.

Recruit purposively from an explicit set of perspectives. Record where candidates were identified, who nominated them, how eligibility was assessed and whether any organisations dominate. Professional networks are practical recruitment routes, but relying entirely on a supervisor's contacts can narrow the range of assumptions represented. Declare conflicts, such as commercial interests in an option under consideration.

There is no universal minimum Delphi panel size. Justification should connect the scope of the judgment, diversity of relevant expertise, access, workload and expected retention. Skulmoski, Hartman and Krahn (2007) discuss panel composition and varied graduate research designs. A small, specialised panel and a broad stakeholder exercise need different arguments. Neither is justified by copying a number from another dissertation.

Use the sampling-method and sample-size guide for the broader sampling argument. A purposive Delphi panel is not a probability sample: a precise percentage describes its responses, not the proportion of an entire profession holding that view.

Prepare the first round and the statement revision trail

An open first round should ask participants for material that can become assessable statements: proposed criteria, reasons, qualifications and overlooked situations. Explain the setting and the decision without supplying a preferred answer. A rating-based opening instead needs a documented item source and an opportunity to identify missing or misleading statements.

Before launch, check interpretation, burden and technical operation with suitable people outside the main panel where feasible. Establish whether their suggestions will change the instrument and whether their data enter the analysis. Keep general question-writing guidance brief and use the existing questionnaire-design guide for detailed wording principles.

For qualitative responses, identify the unit being analysed and the method used to group suggestions. Preserve links between source comments, categories, draft statements and revisions. A suggestion made once can be consequential; frequency alone should not decide whether it survives. Explain who reviewed ambiguous wording and how competing interpretations were handled.

Keep permanent item IDs. Log mergers, splits, deletions and reasons. Substantive rewriting creates a different proposition: a rating increase cannot then be interpreted solely as experts changing their minds about an unchanged statement. If you use structured coding, the qualitative codebook guide supports that supporting task without turning Delphi analysis into a separate coding tutorial.

Plan feedback, anonymity and ethical safeguards

Controlled feedback should make reconsideration possible without instructing experts to conform. A useful item report includes its wording, valid response count, distribution, central summary, spread and a balanced summary of reasons. Return each panellist's own previous rating privately. Include objections and qualifications alongside supportive comments, and explain how comments were selected or paraphrased.

Avoid messages such as “most experts agree, so please reconsider your low score.” Instead ask whether the summary changes the participant's assessment and invite reasons for retaining a different position. Pilot feedback as carefully as the questions: a selective summary can change the decision being made.

Anonymity usually concerns responses being concealed from other panellists. Researchers may need identities for recruitment, reminders and linking responses across rounds. Experts may also recognise colleagues from a role, municipality or unusual incident. Promise only the protection the design can deliver.

Obtain the institutional ethics review and permissions required for the real project before recruitment. Explain voluntary participation, repeated contact, withdrawal arrangements, feedback use and quotation permissions. Separate contact details from coded responses, restrict access and set a retention plan. In small professional communities, aggregate or suppress identifying role combinations and redact recognisable operational incidents. Pseudonyms alone do not prevent deductive identification. The ethical-considerations guide provides the wider dissertation framework.

Prespecify consensus, stability and stopping separately

Consensus concerns the pattern of judgments at a given stage. Stability concerns how responses change across stages. Stopping concerns when the process ends. They answer different questions: a panel can repeatedly disagree, or show broad support that is still changing.

Choose a definition aligned with the decision. Percentage agreement needs a specified response region and denominator. A median summarises position; an interquartile range (IQR) summarises the middle portion of the distribution. Neither establishes that every participant agrees. A polarized panel can have a favourable median, so retain the distribution and substantive objections.

Diamond and colleagues (2014) found substantial variation and incomplete reporting of consensus definitions. A commonly observed threshold is descriptive evidence of practice, not a universally correct cutoff. Explain why your own combination of measures is appropriate before viewing the results.

For ordinal ratings, medians, response-category counts and IQRs are often useful descriptive summaries. State the quantile algorithm because software conventions can yield different quartiles in small panels. If participants rank options, Kendall's coefficient of concordance may address agreement in their ordering; it does not directly answer whether an individual statement is supported. Kappa or other agreement statistics require an appropriate rating design and assumptions. Adding an impressive statistic does not repair a poorly defined decision rule.

A stability analysis should compare linked participants on unchanged items. Comparing only the overall percentages can confuse withdrawal with opinion change. Inferential tests of change also answer a different question from a prespecified practical stability criterion: a nonsignificant test does not by itself demonstrate equivalence.

Specify how many rounds are possible, which items return, what happens to unresolved items, and what happens if retention becomes inadequate. A defensible stopping rule can combine a practical round limit with explicit reporting of consensus and dissent. Do not continue indefinitely or lower the threshold after seeing an inconvenient result.

Worked example: community flood-warning readiness

Question, panel and scope

The fictional question is: Which proposed criteria should a municipal emergency-management panel regard as essential when reviewing community flood-warning readiness? The product is a provisional set of supported criteria, with unresolved proposals documented separately. It is not a ranked investment plan, an operational instruction or a test of warning effectiveness.

Eligibility requires a current or recent municipal emergency-management role, at least five years of relevant experience, and direct participation in a flood-warning exercise or activation during the preceding three years. Candidates also need experience in operations, warning communications or community liaison, willingness to complete the scheduled rounds, and disclosure of interests. These are choices for this fictional scope, not a general definition of an expert.

The researcher seeks fifteen panellists spanning the three practice areas and municipalities with different settlement patterns. That target balances coverage of these perspectives against the feasibility of reviewing detailed comments. It does not establish statistical representativeness or prove that fifteen is sufficient for every Delphi. This professional-only scope excludes direct resident participation, which limits the legitimacy of claims about residents' priorities or accessibility needs.

Synthetic recruitment and round-by-round participation
StageRecorded flowMeaning of the denominator
Candidate screening24 identified; 4 ineligible; 20 eligible invitedIneligible candidates were removed before invitation.
Consent16 consented; 4 declinedConsent yield = 80% of eligible invitees.
Round 1: open responses15 responded; 1 consenting person did not startInitial response = 75% of the 20 invitees; analysis IDs P01–P15.
Round 2: initial ratings14 of the 15 Round 1 respondents returnedRound-to-round participation = 93.3%; P15 became unavailable.
Round 3: feedback and re-rating12 of the 14 Round 2 respondents returnedRound-to-round participation = 85.7%; P13 and P14 became unavailable.
Retention from the starting panel12 of 15 Round 1 respondents completed Round 3Retention = 80%; cumulative panel attrition = 20%.

P01–P05 primarily work in operations, P06–P10 in warning communications and P11–P15 in community liaison. These broad fictional categories show why losing participants can alter the panel even when the overall retention figure looks acceptable. Round 3 contains only 2 of the original five community-liaison panellists. In a real report, publish role detail only when confidentiality permits it.

Round 1: elicit candidates and develop statements

The opening instrument asks:

  1. What evidence would make you regard a community as ready to receive and act on a flood warning? Explain why each proposed criterion matters.

  2. Which failure in the warning process would make a readiness claim unconvincing?

  3. Under what local conditions would an apparently useful criterion be inappropriate or difficult to assess?

  4. Which perspectives or situations might this professional panel overlook?

The researcher groups responses by decision ownership, communication resilience and alert timing. The following extracts are invented to demonstrate the analytic trail; no qualitative frequency or empirical completeness is claimed. For this deliberately small exercise, the resulting rating instrument contains exactly the three statements below.

Synthetic Round 1 suggestions and the resulting item history
ItemInvented suggestionStatement used unchanged in Rounds 2 and 3
A: responsibility“Everyone assumes someone else can authorise the alert.”A readiness review should require a documented role authorised to initiate a community flood warning.
B: fallback channel“The usual warning channel can fail when power or connectivity is lost.”A readiness review should require a tested fallback route for issuing community flood warnings when the primary channel fails.
C: fixed timing target“An alert needs a deadline, but terrain and forecast confidence change what is feasible.”A readiness review should require every neighbourhood to receive a warning within ten minutes of the municipal alert decision.

The first suggestion is narrowed to authorisation because staffing and training would be separate propositions. The second becomes a fallback-route criterion rather than a requirement for a particular technology. The timing statement preserves a contestable universal requirement so the panel can judge it. Ten minutes is an invented candidate target, with no claim that emergency-management evidence recommends it.

Between the rating rounds, the researcher clarifies the instructions but makes no substantive changes to these three statements. Comments requesting different timing conditions are preserved as dissent. If C were rewritten to apply only to selected neighbourhoods, it would need a new version and could not inherit the stability result for this wording.

Rating scale and protocol rules

Participants rate agreement that each statement should be included as an essential readiness criterion: 1 = strongly disagree; 2 = disagree; 3 = neither agree nor disagree; 4 = agree; 5 = strongly agree. “Unable to judge” is a separate nonnumeric response. It is not coded as the midpoint.

The fictional protocol is fixed before constructing the ratings:

  • An item qualifies for inclusion when at least 80% of valid ratings are 4 or 5, the median is at least 4, and the IQR is at most 1, with at least ten valid ratings. All conditions must hold.

  • The minimum of ten valid ratings is a local feasibility safeguard. Below it, an item is labelled insufficient response rather than classified for inclusion.

  • Quartiles use linear interpolation at zero-based position (n - 1) × p, corresponding to R's type 7 convention. IQR = Q3 − Q1.

  • Item stability is described as at least 80% identical ratings among participants with valid linked responses in both rating rounds. This deliberately simple local rule is not a validated stability standard; the full distribution remains necessary.

  • All items return after feedback, including those provisionally qualifying in Round 2. The study stops after the planned third round; unresolved items are reported, and there is no automatic additional round to obtain agreement.

  • Use available valid ratings for each item, display the item denominator, retain unable-to-judge counts, and do not impute withdrawals or carry ratings forward. No replacement experts enter after Round 1.

The inclusion rule defines consensus in favour. Failure to meet it does not prove consensus against an item. No separate exclusion-consensus rule is used here. The figures above are illustrative design choices, not recommendations that every dissertation should copy.

Participant-level synthetic ratings

In the table, U means unable to judge; W means withdrawn before that round. P15 supplies open comments only and has no rating-round data. Each row represents the same fictional participant across rounds.

Complete synthetic rating matrix for the two rating rounds
ParticipantA R2A R3B R2B R3C R2C R3
P01444411
P02554411
P03443422
P04554422
P05444422
P06553334
P07444444
P08443444
P09554445
P10444455
P11554455
P1244335U
P134W4W5W
P143W3WUW

Rounds 2 and 3: results and controlled feedback

After Round 2, each participant receives the category counts, median, quartiles and valid denominator for each item, their own previous answer, and a balanced synthesis of comments. For B, the synthetic comments distinguish a tested fallback route from a requirement to purchase duplicate equipment. For C, they contrast a clear service target with concerns about uneven connectivity and the danger of encouraging premature alerts. These are illustrative arguments, not tested factual findings.

A feedback message for C could read: “The panel includes both support for a common timing target and objections to applying one target in every neighbourhood. The objections concern feasibility and the conditions under which an alert decision is made. Please re-rate the unchanged statement; retaining your previous judgment is acceptable.” The actual feedback should also display the numerical distribution below. It should not imply that a low rating is an error.

Synthetic item results and inclusion decisions using the prespecified rule
Item and roundValid nRatings 4–5AgreementMedianQ1Q3IQRQualifies for inclusion?
A, Round 2141392.9%4451Yes, provisionally
A, Round 31212100%4451Yes, final
B, Round 214964.3%4341No
B, Round 3121083.3%4440Yes, final
C, Round 213753.8%4253No
C, Round 311654.5%424.52.5No; retain disagreement

Percentage agreement is 100 × count of ratings 4 or 5 / valid item ratings. For B in Round 3, agreement = 100 × 10 / 12 = 83.3%. The denominator includes the two midpoint responses. It excludes neither a valid disagreement nor a neutral response merely because that answer obstructs consensus.

For C in Round 3, the sorted valid ratings are 1, 1, 2, 2, 2, 4, 4, 4, 5, 5, 5. Its median is 4. For Q1, position = 2.5, between two values of 2, giving Q1 = 2. For Q3, position = 7.5, halfway between 4 and 5, giving Q3 = 4.5. Thus IQR = 2.5, while agreement = 100 × 6 / 11 = 54.5%. The favourable median conceals a divided panel: 5 valid ratings remain in categories 1–2.

Stability, withdrawal and the stopping decision

Linked-participant stability on unchanged statements
ItemParticipants with both valid ratingsIdentical ratingsPercentage identicalMeets the local stability rule?
A1212100%Yes
B121083.3%Yes
C11981.8%Yes, despite disagreement

The paired calculation for C excludes P12's unable-to-judge response and the two withdrawals. Stability therefore refers to those with valid linked ratings, not everyone recruited. Even this rule can permit consequential changes: a stable label should not replace inspection of the actual transitions.

A's overall agreement rose from 92.9% to 100%, yet none of the returning participants changed their A rating. P14's midpoint response disappeared through withdrawal. The apparent improvement must not be attributed to persuasion by feedback. B's matrix, by contrast, shows P03 and P08 moving from 3 to 4.

For C, one unable-to-judge response occurs in each rating round, from different people. These are item-level missing ratings, not additional panel withdrawals. No midpoint or previous response is substituted. As a conservative descriptive check, count everyone who participated in Round 3 in the denominator: C then has 6 / 12 = 50% positive ratings. This sensitivity calculation does not treat the missing response as observed disagreement.

The study stops at its planned cap. A and B qualify for the provisional criterion set; C remains unresolved and is reported with its arguments and distribution. No forced ranking between A and B is supported by these agreement ratings. Prioritising expenditure or selecting a limited number of actions would require a separate decision task.

Reproduce the calculations

The following standard-library Python code reproduces the final C summaries directly. Apply the same function to the other columns in the printed matrix. All table summaries and linked-participant calculations were also independently recomputed from the saved participant CSV using a separate JavaScript implementation.

from math import floor, ceil
from statistics import median

def quantile(values, p):
    ordered = sorted(values)
    h = (len(ordered) - 1) * p
    lo, hi = floor(h), ceil(h)
    return ordered[lo] + (h - lo) * (ordered[hi] - ordered[lo])

ratings = [1, 1, 2, 2, 2, 4, 4, 4, 5, 5, 5, None]
valid = [x for x in ratings if x is not None]
q1, q3 = quantile(valid, .25), quantile(valid, .75)
agreement = 100 * sum(x >= 4 for x in valid) / len(valid)
print(len(valid), median(valid), q1, q3, q3 - q1)
print(round(agreement, 1))

The output is 11 4 2.0 4.5 2.5, followed by 54.5. This verifies arithmetic under the specified convention. It does not validate the panel composition, thresholds, synthetic comments or emergency-management criteria.

Write the Delphi methods and results transparently

The methods should let a reader reconstruct the process: purpose and variant; eligibility and recruitment; panel composition; item sources and revisions; instrument and feedback versions; anonymity and ethics; round schedule and reminders; analysis and missing-response rules; and prespecified consensus, stability and stopping criteria. Identify deviations and who made them.

The results should account for invitations, participation and item denominators, then show final statements, distributions, decisions and unresolved views. Report items removed or rewritten as well as those retained. Appendices can preserve questionnaires, de-identified feedback templates and a statement revision log. Release participant-level data only when consent, ethics and disclosure risk allow it; the complete matrix here is publishable because it is fictional.

CREDES, developed by Jünger and colleagues (2017), was proposed for Delphi guidance development in palliative care. Its emphasis on explaining design decisions and reporting the process is useful beyond that setting, but using it for municipal research requires an explicit adaptation. The ACCORD explanation and elaboration by Logullo and colleagues (2024) provides further reporting guidance for consensus methods in health research. Neither checklist certifies that a panel's conclusion is correct.

Illustrative methodology paragraph

This fictional study used an open first round followed by two online rating rounds to examine municipal professionals' judgments about proposed flood-warning readiness criteria. Purposive eligibility required relevant municipal experience and recent involvement in an exercise or activation. Open responses informed three statements, whose substantive wording remained unchanged across rating rounds. Participants received their own previous responses, group distributions and summaries of supportive and dissenting reasoning before re-rating. Inclusion required at least 80% positive valid ratings, a median of at least 4, an IQR no greater than 1 and at least ten valid responses. Item-level missing ratings were excluded from the item denominator without imputation. Stability used linked valid responses and a prespecified 80% identical-rating criterion. The process ended after the planned third round, with unresolved proposals retained in the report.

Illustrative results paragraph

Of 20 eligible invitees, 15 contributed to the opening round, 14 returned for initial ratings and 12 completed re-rating. Retention from the opening panel was 80%. In the final round, A received 12 positive ratings from 12 valid responses and B received 10 from 12; both met the prespecified inclusion rule. C received 6 positive ratings from 11 valid responses (54.5%), with a median of 4 and an IQR of 2.5. Its linked ratings met the local stability rule while showing persistent disagreement. Withdrawal reduced community-liaison representation, and the professional panel did not directly represent residents. The supported criteria are provisional panel judgments; these synthetic results demonstrate neither operational feasibility nor warning effectiveness.

Interpret consensus without confusing it with correctness

Agreement describes judgments produced by this panel under this process. Experts can share an incorrect assumption, overlook evidence, defer to a selectively framed summary, or agree on a value that is inappropriate for another setting. Repeated feedback does not provide an independent test of the underlying factual claim.

A readiness criterion supported by the fictional panel would still need examination with communities and relevant technical evidence. Its practical application could require feasibility work, accessibility assessment and evaluation of actual warning performance. The Delphi identifies a reasoned proposal to examine; it cannot substitute for those studies.

State limitations in terms of the conclusions they restrict. A professional-only panel limits stakeholder coverage. Network recruitment can concentrate a particular practice culture. Attrition can remove precisely the perspectives that created disagreement. Fixed response options and edited feedback shape what participants can express. A chosen threshold can change which items qualify. A practical round cap can leave judgments unresolved. None of these limitations is repaired by reporting more decimal places.

Common mistakes that weaken Delphi dissertations

  • Naming a variant without describing it: supply the actual item-development, feedback and re-rating procedure.

  • Calling every invited professional an expert: show eligibility criteria and screening evidence relevant to the judgment.

  • Choosing thresholds after inspecting ratings: retain the prespecified rule and label exploratory alternatives honestly.

  • Using only a median: inspect category counts, spread and reasons for minority positions.

  • Treating withdrawal as convergence: distinguish all-responder summaries from linked-participant changes.

  • Combining substantially rewritten items across rounds: retain versions and analyse the changed proposition separately.

  • Dropping dissent from the final report: explain unresolved items and why the process stopped.

  • Claiming a validated checklist from consensus alone: identify which empirical and stakeholder checks remain outstanding.

Frequently asked questions

Is a Delphi study qualitative, quantitative or mixed methods?

It depends on the question, data and analytic design. An open round may produce qualitative material and later rounds numeric judgments. Explain how those components relate. The presence of comments beside ratings does not by itself establish a fully developed mixed-methods design.

Can I start with statements instead of open questions?

Yes, if an evidence-informed structured opening fits the purpose. Explain where statements came from, how omissions can be challenged and why a modified beginning is appropriate. Do not describe a literature-derived list as though participants generated it.

How many experts and rounds do I need?

There is no universally sufficient number. Justify expertise coverage, the judgment's scope, burden and retention. Plan enough opportunity for meaningful feedback and reconsideration, with an explicit stopping rule. A count copied from a previous study is an incomplete rationale.

Does an 80% threshold make my study rigorous?

No. It is one possible operational choice. Rigour also depends on the panel, item wording, feedback, missingness and transparent interpretation. In this example, 80% belongs to a declared teaching protocol; it is not a general methodological standard.

What if the panel never reaches consensus?

Report which items remain unresolved, their distributions and the reasoning behind disagreement. Explain the stopping decision. Persistent differences can reveal scope conditions, competing values or an inappropriate universal statement and can be a useful result.

Should I replace experts who withdraw?

Replacement changes who has received the accumulated feedback. Decide the policy in advance and distinguish new entrants in analysis and reporting. In the worked example no replacements enter, and the resulting loss of community-liaison perspectives is a limitation.

Do neutral and unable-to-judge answers count the same way?

No. A valid midpoint belongs in the denominator under the example's rule. Unable to judge is recorded separately and excluded from the valid-rating denominator. Display both counts and avoid silently turning uncertainty or nonresponse into agreement.

Can a Delphi validate my dissertation questionnaire?

Expert judgments may inform content development or assess perceived relevance. They do not establish every form of measurement validity, reliability or real-world performance. Describe exactly what the panel evaluated and which further investigations the instrument requires.

Support with a defensible Delphi dissertation

Acadelyra can help you review the fit between your question and Delphi design, clarify a protocol, check calculations and improve your reporting. Through Acadelyra Services, support should strengthen your own reasoning and writing. Recruitment, ethics approval, data collection and honest representation of findings remain the researcher's responsibility; expert responses and study events must never be fabricated for an actual dissertation.

References and methodological reading