Summary

  • APNIC reports an average governance score of 77% among respondents able to assess eight member-only statements, while excluding “don't know” responses from those positive proportions.
  • The same report separately records a 20% average “don't know” score; regionally, the paired results range from 82%/12% in Least Developed Economies to 62%/35% in Oceania.
  • The survey release's overall 95%-confidence, within-three-points statement should not be assumed to apply unchanged to Q27's 731-member base or its smaller regional slices.

APNIC's 2026 Member and Stakeholder Survey gives its governance chapter a reassuring headline: an average score of 77%. On the same page, it publishes another number that is easier to lose in a presentation: an average “don't know” score of 20%.

These figures are not rivals. They answer different questions.

The 77% describes the average share choosing one of the two positive answers across eight governance statements, after people who selected “don't know” have been removed from the relevant positive proportion. The 20% describes the average frequency with which respondents said they could not assess those statements. One is confidence among assessors. The other is the measured boundary of assessment.

That distinction is the story. A registry may be highly regarded by people who know enough to form a view while remaining difficult to evaluate for a substantial share of its members. Combining those conditions into one unqualified approval number would erase the very uncertainty that APNIC's report has chosen to disclose.

One survey result, two denominators

The governance questions were not asked of all 1,385 valid respondents celebrated in APNIC's 10 September release. The questionnaire routes them to APNIC Members or Account Holders. The detailed report gives Q27 a base of 731 APNIC members.

Nor does every percentage on the governance page use all 731 answers in the same way. The report displays the full response distribution, including “don't know”, for each statement. It then calculates its summary proportions among respondents able to provide an assessment. The denominator for the positive score therefore becomes the people who expressed a substantive view on the relevant item.

This is a defensible analytical choice. Treating “don't know” as disapproval would invent an opinion. Treating it as neutral would also invent an opinion. Excluding it from an assessor-only score preserves the meaning of the answer.

But exclusion narrows the claim. “Seventy-seven per cent of assessors rated the measures positively on average” is not the same proposition as “77% of members endorse APNIC governance”. The first is supported by the stated method. The second would silently move respondents between ledgers.

APNIC itself keeps the distinction visible. It says the eight measures were grouped into Trust, Information and Engagement, Governance Processes, and Reporting and Meetings. It reports both an overall governance confidence score and an average “don't know” score for each respondent group. The useful reading is therefore paired, not singular: 77% positive among assessors, with a 20% average uncertainty rate in the member response distribution.

“Don't know” is an evidence state

The item rows show why the unknown ledger matters.

For the statement that issues raised with APNIC are handled fairly and objectively, 68% of the full displayed distribution agrees or strongly agrees and 15% answers “don't know”. Among people with enough knowledge to rate the claim, the report presents the Trust dimension at 80%.

For the statement that APNIC's financial reporting is transparent and sufficient to understand its financial position and performance, the appendix shows 38% agree, 12% strongly agree and 30% “don't know”. For whether APNIC Member Meetings provide sufficient information about activities, it shows 44% agree, 15% strongly agree and 23% “don't know”.

Those are not signs of a failed survey. They are signs that governance claims have different observability. A member can use registry services, receive an invoice or attend training without reading EC minutes or financial statements closely enough to assess their sufficiency. The survey's separate unknown category records that limit instead of forcing a synthetic judgement.

The report offers cautious possible explanations: limited interest, lower awareness, or no felt need to engage more deeply. It does not establish which explanation applies to any respondent, and it does not turn an unknown answer into criticism. That boundary should survive any later Executive Council response.

Regional comparisons need their bases

The paired regional results are striking. South Asia records an 82% governance score and a 14% average “don't know” score. South East Asia records 81% and 13%. East Asia records 76% and 26%. Oceania records 62% and 35%.

Those pairs are more informative than the positive scores alone. Oceania's lower assessor score travels with the highest average unknown rate. East Asia's overall positive score sits near the total, but its unknown rate is six points above the total. The data describe differences; they do not explain them.

The appendix also shows why the base belongs beside every comparison. Q27 has 92 East Asia responses, 159 Oceania responses, 204 South East Asia responses and 247 South Asia responses. The Least Developed Economies and Other-economy columns are overlapping economic classifications, not extra regions that can be added to those geographic counts.

The report adds a further caveat: Australia and New Zealand account for about 70% of the Oceania sample. “Oceania” therefore does not carry an evenly distributed Pacific voice. APNIC states this plainly. A later slide, budget note or strategy paper should preserve it just as plainly whenever it relies on the 62%/35% pair.

The three-point headline does not travel by itself

APNIC's release says 1,385 valid responses provide 95% confidence that results are within three percentage points of the presented figures. The public report and appendix then expose several smaller effective bases. Initial classification tables use 1,384. Service-rating tables fall to roughly 1,072. Governance-document readership uses 744 members, and Q27 uses 731. Regional Q27 slices fall as low as 92.

The public package reviewed for this article does not publish a sampling frame, invitation count, response rate, weighting design, design effect or question-by-question margin-of-error table. It also does not explain the one-response difference between the headline 1,385 and the appendix's classified base of 1,384. That is not evidence of error; it is an unanswered reconciliation point.

An elementary simple-random-sample calculation illustrates the scale issue without pretending to reconstruct APNIC's design. At maximum variance, a 95% margin is about 2.6 percentage points for 1,385 observations, about 3.6 for 731, about 7.8 for 159 and about 10.2 for 92. These are editorial illustrations, not APNIC's actual margins.

The Australian Bureau of Statistics makes the underlying rule clear: a complex survey needs enough units in each subgroup to support reliable subgroup estimates, and a larger sample does not cure non-response bias. A single aggregate confidence phrase is therefore not a portable warranty for every filtered question and regional cell.

Eleven languages widen reach, not authority

The release says the questionnaire was available in 11 languages and that 363 surveys, 26% of all responses, were completed in a language other than English. That is valuable reach evidence. It reduces one obvious barrier and records a sizeable demand for non-English participation.

It does not, by itself, establish linguistic representativeness. The report's appendix shows how the 363 non-English completions were distributed across ten language choices, but the reviewed package provides no denominator for how many eligible people preferred each language or never received, opened or completed the survey.

The correct conclusion is narrower and more useful: APNIC made multilingual completion possible, and more than one in four respondents used it. Future survey design can now ask whether that access changes who enters the scored base, who remains in the unknown ledger and which regional comparisons become more stable.

Preserve both sides of the result

The report deserves credit for not burying “don't know” inside a neutral category and for stating that it excludes those answers from assessor-only proportions. The next governance step is not to collapse the two scores. It is to make their relationship operational.

A compact denominator strip beside every headline result would do the work. It should state total valid responses, eligibility for the question, number answering, “don't know” count and rate, number included in the positive-score denominator, segment base, sampling or weighting note, and a margin of error only where the design supports one.

That strip is an editorial recommendation, not an announced APNIC commitment. Its purpose is modest: make it impossible for a planning document to carry 77% forward while dropping the 20% that defines the edge of the claim.

What the sources establish

The sources establish the release date, fieldwork dates, independent research firm, aggregate response and language counts, Q26 and Q27 bases, answer distributions, score construction, regional score pairs and the report's Oceania caveat. They do not establish why every base changes, who did not respond, a probability-sampling design, causal explanations for regional differences or a member mandate.

Sources