Ask users for ethnicity

Optional, explained, self-identified, and never used to adjust a clinical threshold without a named clinical owner.

When to use

There are two legitimate reasons to ask, and they call for different handling.

Equity monitoring. Checking that your product works as well for everyone who uses it. This is a good reason, it is the more common one, and it does not require the answer to affect anything the individual reader sees.
A clinical purpose with a named owner. Some risk calculators and reference ranges have historically been adjusted by ethnicity. This is contested clinical ground, several such adjustments have been withdrawn in recent years, and it is not a decision a design system may take. If your product does this, a named clinician owns the decision and it is documented on your own thresholds page. See Reference ranges.

If you cannot name which of the two you are doing, you are not ready to ask.

When not to use

You have no analysis plan. Collecting equity data you never analyse is extraction, not equity. Decide what you will look at before you add the field.
You want it as a personalisation input. Ethnicity is not a preference signal. Using it to decide what content to show is a decision that will eventually be read out loud in a room you would rather it were not.
You mean nationality, language, or country of birth. Those are different questions with different answers, and one of them is usually what a product actually needs.
The reader is mid-task and worried. Never ask this on a results screen or inside a symptom flow. It belongs in a profile, at a moment of the reader's choosing.

How it works

Diagram source (mermaid)

flowchart TD
A["We want ethnicity data"] --> B{"Equity monitoring, or a clinical adjustment?"}
B -->|"neither can be named"| C["Do not ask"]
B -->|"equity"| D["Optional field in the profile, with an analysis plan"]
B -->|"clinical"| E{"Is a named clinician accountable for the adjustment?"}
E -->|"no"| F["Do not adjust; do not ask"]
E -->|"yes"| G["Ask, explain the adjustment, and let the reader see it"]
D --> H["Self-identified, multi-select, prefer not to say, free text"]
G --> H

Rules:

Always optional, always with "prefer not to say". In a health product this question has a history that makes refusal a reasonable and common response.
Self-identification only. Never infer it from a name, a postcode, a language setting or anything else. Inference here is both unreliable and offensive.
Multi-select, with free text. People have more than one ethnicity, and every fixed list excludes somebody. Offer a list because it makes analysis possible, and allow an answer outside it because accuracy matters more than tidy data.
Use a recognised classification for the reader's country, and say which one you are using. Categories differ substantially between national schemes. A list built for one country is confusing and sometimes insulting in another. Do not invent your own.
Never let the answer silently change a number. If ethnicity adjusts a threshold, the adjustment is visible to the reader, attributed, and explicable. A reader who changes their answer and watches their result change with no explanation has been shown something they cannot interpret.
Say what it is used for, in one sentence, at the field. And make it true.

Content

"Your ethnic group (optional). We ask so we can check that this product works as well for everyone who uses it. It does not change anything you see, and you can skip it."

Don’t

"Ethnicity *". The field is required, unexplained, mid-signup, with a five-item list and an "Other" radio.

Where an adjustment genuinely applies: "This calculation is adjusted for ethnicity. Here is what it uses and why," with a link to the clinical source and the name of the group that owns it.

Don’t

Applying an adjustment invisibly, so the same reading produces two different results depending on an answer given eighteen months ago in a settings screen.

Sensitive-question wording is owned by Asking sensitive questions.

Accessibility

No autofill token exists for ethnicity. Use autocomplete="off" and do not invent one.
A multi-select is a fieldset of checkboxes with a legend carrying the question, not a listbox. Readers need to see the options to answer this honestly.
"Prefer not to say" is an explicit option, mutually exclusive with the others, and its exclusivity is announced rather than silently enforced.
Free text is a labelled field, revealed by an option whose label says a text field will appear, so the change is not a surprise to a screen-reader user.
The explanation precedes the options and is part of the group's accessible description.
Long lists need a heading structure or grouping, not a single flat list of forty options. National classifications are often nested, and the nesting is meaningful.

National statistical agencies publish ethnicity classifications and the research behind their wording; use the one for the country your readers are in and cite it, rather than composing a list. We deliberately do not reproduce any classification here. They are country-specific, they change, and copying one into a design system is how an out-of-date list spreads.

The position that ethnicity-based clinical adjustment requires a named accountable clinician is ours and is opinion, though it is a conservative one: several such adjustments have been re-examined and withdrawn in recent years, and a design system is not the right layer to encode any of them. Our rule is procedural rather than clinical. We do not say whether an adjustment is right, only that a design system may not make it silently and that the reader must be able to see it.

What would change our mind: nothing on visibility or on optionality.

Opinion. No study is being claimed here. If you have evidence that contradicts it, that is the fastest way to change this page.

Updates to this page

Last read through against the system on 2026-09-20. Due for review every 6 months; expiry is reported by pnpm run check:freshness.