The Guernsey Business Census · Methodology

How this census counts

Every figure on the census is meant to be checkable by a stranger with an afternoon. That means naming the population it is counted over, saying how much of that population the figure actually knows about, and refusing to print the ones that know too little. This page is those rules.

Figures on this page are live — as of 25 August 2026

Three populations, always named

#populations

There is no such thing as “the number of businesses” without a base. Every figure on the census is counted over one of three, and says which.

Everything active
3,017
Every organisation with a live trading or public presence in the Bailiwick that we have verified: businesses, charities, clubs, churches, public bodies. A company registration on its own does not qualify; a market stall with customers does.
With a primary sector
2,744
Those we have classified into one of the sixteen sectors. Sector counts add up to this number, not to the one above — the difference is organisations still awaiting classification, and the census says so beside the chart.
With the attribute on record
varies
A distribution of founding years describes the organisations that publish a founding year, not the island. That base is printed under every chart, and it is what the coverage figures below measure.

One consequence worth knowing before you quote anything: an organisation tagged in two sectors is counted in both wherever the census says “works in”. Those charts add up to more than the island does, on purpose, and they never say “is in”. The cross-sector chart is the one built on that basis.

What a number is allowed to do

#tiers

Coverage is the share of the population that has the attribute on record. It decides what a figure may be used for — and the renderer enforces it, so a thin number cannot be promoted into a headline by choosing a different chart.

TierTestMay be
CitableCoverage at or above 60% and at least 30 organisations — or the base is complete by construction, or the figure is reconciled against a public registerA headline, a share card, a line in a press kit
IndicativeCoverage at or above 30% and at least 15 organisationsA chart inside its own section, wearing a visible “indicative” badge. Never a headline
SuppressedAnything below thatNothing. The section prints the gap instead — the blank is the point

“Complete by construction” means every organisation necessarily has an answer: it is on one island or another, it either publishes a website or it does not. Those figures need no coverage test because there is nothing to be missing.

What the census knows today

#coverage

Live coverage per attribute, from the same aggregates the charts read. Follow a row to the section it governs.

On recordOrganisationsCoverageTier
A primary sector2,744 of 3,01791%Citable
A parish (Guernsey only)2,654 of 3,01788%Citable
A founding year1,510 of 3,01750%Indicative
A website of its own3,017 of 3,017completeCitable
A social link1,940 of 3,01764%Citable
Capabilities resolved to the graph2,519 of 3,01783%Citable
A weekly opening-hours grid1,328 of 3,01744%Indicative
A legal form1,104 of 3,01737%Indicative

Every count is a floor

#floors

The census is a verified subset of the island, and it grows every week. So a count here is the number we can stand behind, not the number that exists. The copy is written to carry that: organisations are “counted”, “on record”, “verified” — never “Guernsey has”. Superlatives are qualified the same way: the oldest on record, the only one we have found.

If you are writing a sentence from a figure here, that qualifier should survive the edit. “3,017 organisations counted” is supportable; “Guernsey has 3,017 businesses” is not, and the difference is the whole point of the page.

Ratios between two counts from the same population are fine and appear throughout. “One in N” is used only where the ratio genuinely rounds to a whole number; otherwise the census prints a percentage rather than flattering the arithmetic.

How complete the count is, and how we know

#recall

The coverage panel states the estimated true trading population as a range — 5,0006,000 — because it is an estimate and a point estimate would pretend otherwise. It comes from capture–recapture against independent listings (Google Business Profile and OpenStreetMap), measured on 13 August 2026.

The measurement only works while the two enumerations are independent, which is a real constraint rather than a formality: once a source has been used to add organisations to the census, that source can no longer be used to measure how much the census is missing. Each re-measurement therefore uses a pair that has stayed independent, and it is re-run rather than assumed. The bar on the coverage meter takes the conservative end of the range, so it can never overstate what has been counted.

The bias sits next to the chart, not in a footnote

#bias

Four families of figure on the census have a known lean. Each carries its sentence beside the chart, where the number is.

  • Founding years

    Counts organisations still trading that publish a founding year; older firms are likelier to do so. Two biases at once — survivorship, because the ones that closed are not here, and reporting, because a founding date is a thing established firms advertise.

  • Anything web-derived

    Verified from published evidence; businesses without a web presence are under-represented. Sole traders and Facebook-only businesses are the ones this misses, and the miss is not evenly spread across trades.

  • Capabilities

    Capabilities as published by each organisation and resolved to the graph. A service nobody has written down anywhere is not in the graph, and one written down in a way that did not resolve to a known node is not either.

  • Register comparisons

    Register rows not yet matched to a trading presence are counted as unmatched, not as non-trading. An unmatched row is a job we have not done, not a business that does not exist.

Where the facts come from

#provenance

Facts on the census are one of three kinds, and only two of them are published here.

Graph-verified
the default
Our own claims, each linked to the published source it came from and re-checked over time.
Register-anchored
cited
Counts reconciled against a public register or directory, cited to that register. These carry a reconciliation badge rather than a coverage percentage.
Third-party observed
not published
Facts observed on someone else's platform — ratings, listing status, map tags. These are used to measure how much the census is missing; they are not published as census figures.

People are aggregate-only on the census. It will tell you how many people hold roles at two or more organisations; it will not rank named individuals on a statistical page.

Rounding, medians and per-head figures

#precision

Counts are exact. Shares are whole percentages. Estimates are ranges, never a point. A median is only printed over a large enough sample; below that the census says “typically” and gives the range, or prints nothing. Per-head figures cite the resident-population source beside the figure, because a density is two numbers and only one of them is ours.

Corrections, and reusing this

#corrections

Coverage is partial and stated openly; a closure can lag reality by weeks. If something here is wrong, that is worth more to us than a polite silence — tell us and corrections ship continuously. Press enquiries: press@island.gg.

The figures are licensed CC BY 4.0 — credit “Island Graph” with a link. Both machine formats are on the data page, and every figure carries its denominator there too.

← Back to the census