Demographic information: definition and examples

Demographic data examples: 25 characteristics

The fields that appear on real forms, grouped. The table below adds an example answer and the analysis each one pays for.

• About the person: age band, year of birth, gender, language preference, highest education completed, race or ethnicity, religion, sexual orientation, functional difficulty, veteran status, political affiliation.
• Where they are: country of residence, region, state or province, postal code.
• Work: employment status, job role or level, department or team, tenure, industry, organisation size.
• Household: household income band, household size, children or dependants, relationship status, housing tenure.

Organisation size and industry are in the list because they are routinely called demographics. They are firmographics: they describe the company, not the respondent. Keep them in a separate block so nobody cross-tabs a personal field against a company field and reports it as a finding about people.

Age brackets for a survey: which groupings and options to offer

Two rules do most of the work. Bands must not overlap, so 18 to 24 and 25 to 34 rather than 18-25 and 25-34, which leaves a 25-year-old picking at random; the same discipline that makes you make answer options mutually exclusive anywhere else applies here. And bands must match whatever you plan to compare against. If the comparison is a national statistics office, take their bands; five-year and ten-year bands are the common shapes and re-cutting yours later is not possible.

That is the argument for a birth year. An age band cannot be split afterwards, so a seven-band answer locks your analysis for the life of the dataset. A four-digit year can be collapsed into any banding you invent in six months, which is why the third question below exists.

Be clear about what a birth year gives you: a birth year yields the age a person reaches in that calendar year, not their age today. Somebody born in 1987 is 38 or 39 on the date this page was updated, and the year alone cannot say which. If the analysis needs age at the time of response, ask for completed age in years ("How old are you?", whole years). Ask for a full date of birth only when the purpose requires it, such as age verification or matching to a record, and the privacy section explains why.

Demographic information and demographic data: the definition

The word carries two jobs. Demography is the statistical study of a population: births, deaths, migration, the shape of its age structure. Demographic information on a questionnaire is the narrower thing: the fields that describe or group the person answering, as distinct from the questions that measure what the survey is about.

The test is what an answer is about. A demographic item is about the respondent and would read the same on any survey they opened this month. A satisfaction item is about the encounter, and that is the thing you are actually measuring. Religion, gender and political affiliation pass the test even though they are identity and belief rather than record: the form is classifying the person, not measuring the survey's subject. So the definition does not require a field to be fixed for life, only that it describes the respondent.

Demographic data, demographic details, demographic background: one meaning

Those three phrases, plus demographic characteristics, demographic descriptors and demographic profile, all name the same set of fields. The wording changes with the form rather than the content, so treat them as interchangeable and stop looking for a distinction.

One neighbouring term is genuinely different. Firmographic information describes an organisation rather than a person (industry, headcount, revenue band) and it gets filed under demographics on business surveys constantly.

Demographic question examples with answer options

Twenty-seven questions in six blocks: the three age questions above and the twenty-four below. Every option list is written out. Nothing here says "adapt to your region" and leaves you to find the categories yourself.

Core demographic questions

Gender first, because the option set is where most forms go wrong. Woman and Man are the noun forms and they measure gender. Male and Female are the adjective forms and they belong to a question about sex. Four of the five gender questions in our library use the adjective pair under a question that asks about gender, which is a mismatch worth fixing.¹

Race and ethnicity

This is the field where "adapt it to your region" does the most damage, because published standards exist and they are specific. In the United States the governing one is Statistical Policy Directive No. 15, revised by the Office of Management and Budget on 28 March 2024. It replaced the two-question 1997 format with a single combined question, and it added Middle Eastern or North African as a seventh minimum category.³

The directive is explicit that a combined question is required for both self-response and proxy collection, and its own instruction to the respondent reads "Select all that apply". Order is not prescribed; alphabetical is what federal agencies use. Federal agencies must comply no later than 28 March 2029. SPD 15 binds those agencies rather than private companies, but it defines the categories every federal statistic will be published in, so matching it is what makes your numbers comparable to anything.

Our own equity template already ships the SPD 15 version, which is the fastest way to see it working in a live form: the DEI survey template carries all seven categories, an opt-out, and a free-text follow-up for anyone who would rather describe themselves in their own words. It is a United States category list, so use it outside the US only after reading the note on France below. The follow-up matters more than it looks. A fixed list of seven cannot be exhaustive, and SPD 15 recommends write-in detail wherever possible.

Work, role and household

These are the fields that pay for themselves in employee research, and none of the first six is sensitive. Household income is the exception in this block, and the refusal numbers further down explain why it sits apart from everything else. The tenure bands are half-open on purpose: somebody at exactly 12 months fits "6 to under 12 months" or "1 to under 3 years", never both, and the same holds at every other boundary.

Disability

A yes or no question about disability measures identity, not experience, and a large share of people whose daily life is affected will answer no to it. The instrument that avoids the problem is the Washington Group Short Set on Functioning: six questions about difficulty in six domains, designed for censuses, and short enough to sit inside an ordinary survey.⁶

The Washington Group counts a person as having a disability when at least one of the six answers is "A lot of difficulty" or "Cannot do at all". Publish that rule alongside your number, because four different cut-offs are in circulation and they produce prevalence estimates that differ by several multiples.

Religion, political affiliation and sexual orientation

Three fields with one thing in common: no global standard exists, so the honest move is to take a published national instrument and say which one you took. All three are special category data in Europe, which the next section covers.

For religion, take the England and Wales census question. It was put to the entire population of two countries in 2021, and it is the one item on that census form the law makes voluntary. Even so, 94.0 per cent of people answered it, worth remembering the next time somebody argues that a sensitive question will wreck a survey.⁵

For political affiliation a party list stops working the moment you cross a border. The American National Election Studies item is the standard in the United States and its wording has not changed in decades.¹³ Where a party list will not travel, the European Social Survey left-right placement does, because every political system has a left and a right even when the parties differ.¹⁴

The option sets above are the easy half. Wording the surrounding survey so the answers mean anything is a separate craft, covered in how to write survey questions.

Which demographic questions get asked in practice

What counts as basic demographic information

No standards body publishes a list. In practice the short set that earns its place is five fields: an age band, gender, country or region, education, and one field describing how the respondent is connected to you. That last one changes with the survey. For an employee survey it is tenure or role. For a customer survey it is how long they have been buying.

Everything past those five needs a reason you could say out loud. If you cannot name the cut you will run and the decision it will change, the field is costing you completion and buying nothing.

What our own shelf asks

Which of the twenty-five actually get asked is a different question, and we can answer it from our own shelf rather than from a hunch. Across 109 published templates and 1,422 questions, 41 questions are demographic, 2.9 per cent of everything we ask.¹ Here is what those 41 are about.

Tenure is asked more often than age and gender put together

Each of the 41 demographic questions is counted once, under the first topic its wording matches.

Tenure wins by a distance: 19 of the 41, more than age and gender combined. The reason is not hard to see. A person three weeks into a job and a person eleven years in are describing different companies, so the same answer means two different things. Tenure is also the one field nobody hesitates over.

The bottom of the chart is the more interesting half. Two questions about race or ethnicity in the whole library, and two about income. Across all 1,422 questions there is not one about religion or political affiliation, and exactly one about sexual orientation, inside the equity template. Those fields take up a lot of space in guidance, including on this page, and almost none in practice.

Which demographic questions are legally sensitive

Five of the fields in the bank above are regulated in ways the rest are not. Which rules bite depends on where the respondent is, not on where you are. Each subsection names the jurisdiction it describes and none of it transfers to another one.

European Union and the United Kingdom: Article 9 of the GDPR

Article 9 starts from a prohibition rather than a permission. In the EU, and in the UK under the retained UK GDPR, processing of the following is banned unless a listed exemption applies:⁷

• personal data revealing racial or ethnic origin
• political opinions
• religious or philosophical beliefs
• trade union membership
• genetic data
• biometric data, where it is processed for the purpose of uniquely identifying someone
• data concerning health
• data concerning a natural person's sex life or sexual orientation

Five of those eight are ordinary rows in a demographic block. The exemption a survey normally relies on is Article 9(2)(a): the respondent "has given explicit consent to the processing of those personal data for one or more specified purposes". Explicit is the operative word. A privacy policy link at the foot of the page is not explicit consent.

What explicit consent looks like in a form: a separate, unticked box immediately before the sensitive block, naming what the answers will be used for, with the questions skippable either way. The French regulator describes that shape, a clear positive act such as an unchecked checkbox at the start of the form or before the section collecting sensitive data.⁸ The box is the evidence of consent, not the whole of compliance: consent is one of ten Article 9(2) conditions, it has to be freely given (regulators treat consent from an employee to an employer with suspicion for that reason), and the ordinary duties of purpose, retention and security still apply to everything the box covers.

Two fields people expect to find on that list are absent. Age is not special category data. Neither is gender identity. They are ordinary personal data, which means the ordinary rules apply and nothing more.

California: a right to limit, not a consent gate

California takes the opposite structure and it is routinely described wrongly. Under the CCPA as amended, which applies to businesses meeting its thresholds that handle Californian residents' data, racial or ethnic origin, citizenship or immigration status, religious or philosophical beliefs and union membership are sensitive personal information; sexual orientation joins them once the information is collected and analysed.⁹

There is no opt-in requirement before you may collect any of it. What the statute gives a consumer is the right to direct a business to limit use of sensitive personal information to what an average consumer requesting the service would reasonably expect. That is surfaced through a "Limit the Use of My Sensitive Personal Information" link, not through a consent box on your survey.

United States employment: EEO-1, and a form about to disappear

A private US employer with 100 or more employees files an EEO-1 report each year, and that filing still uses the older two-question format: Hispanic or Latino asked first as a standalone item, then six race categories including Two or More Races for everyone who answered no. Self-identification is the preferred method and employers must offer it; where an employee declines, employment records or observer identification may be used.¹⁰

So a US employer today may need both sets. The EEO-1 categories for the filing, and the SPD 15 categories for anything meant to line up with federal statistics. That double life ends when the EEOC adopts the 2024 revision, which it has not yet done.

The disability form is changing sooner. Federal contractors in the United States, under Section 503 of the Rehabilitation Act, have had to use form CC-305 with only the employer-use section alterable. A Department of Labor final rule published on 21 August 2026 rescinds the self-identification requirement, and the department states that the form "will be discontinued, effective September 21, 2026".¹¹ The obligation goes; the measurement problem does not, and the Washington Group set answers it better than a yes or no ever did.

Review scheduled: 21 September 2026, the day the CC-305 discontinuation takes effect. This paragraph was written ten days before that date from the rule as published, and describes US federal contractors only.

Anyone under 13, in the United States

COPPA applies to an operator of a service directed to children, and to any operator with actual knowledge that it is collecting personal information from a child under 13. Consent has to be verifiable and it has to come first. The 2025 amendments to the rule, whose compliance date was 22 April 2026, widened personal information to include biometric identifiers and government-issued identifiers, and added a mixed-audience category.¹²

None of the above is legal advice, it was written from the primary texts by a survey methodologist rather than a lawyer, and the thresholds move. The point of the section is narrower: to tell you which rows in your demographic block are worth five minutes with a lawyer in the respondent's jurisdiction before the survey goes out.

How many demographic questions to ask, and where they go

Start with the number, because it is smaller than most guidance implies. Of 109 live templates on our own shelf, 26 carry a demographic question at all.¹

Most of what we publish asks nothing about the respondent

Eighty-three published templates ask nothing whatsoever about the respondent, and they are not deficient for it. A five-question checkout survey has no analysis that a gender field would improve. The default is zero, and every field you add has to argue its way in.

Where they sit is the second question, and our own shelf answers it consistently. The median demographic question appears 83 per cent of the way through its template. Split by type the pattern sharpens: the 22 questions that are not about tenure sit at a median of 89 per cent, with 19 of the 22 inside the final quarter. The 19 tenure questions sit at a median of 57 per cent, because nine of them open in the first fifth of their template, where a screener goes.¹

What the experiment actually found

One randomized experiment speaks to the first half of the question, within its limits. It ran through a mailed paper survey to 4,448 adults in six Midwest communities in 2019, drawn from the patients and members of one health system, in three arms: no demographic questions at all, demographic questions integrated at the end of the survey, or the same questions printed on a separate sheet.²

Response rates to the main survey, from the paper's Table 2: 34.2 per cent with no demographic questions (507 of 1,484 delivered), 33.0 per cent with the block integrated (489 of 1,480), and 33.1 per cent with it on a separate sheet (491 of 1,482). A chi-square test across the three arms gives p = 0.762: no statistically significant difference was detected. That is not the same as no effect. The observed gap between the no-demographics arm and the integrated arm is 1.1 points (34.16 against 33.04 per cent), and an experiment of this size cannot separate a gap that small from zero.

What it does rule out, for a mailed survey to this population, is a large cost: the folklore that a demographic block tanks response did not appear.

Separating the block did cost something measurable. Among people who responded at all, 99.0 per cent answered at least one demographic question when the block sat inside the survey, against 85.5 per cent when it arrived on its own sheet. Across the whole sample that was 32.7 per cent against 28.3 (p = 0.012). So a separate mailed sheet produced less demographic data. That is the whole of the finding, and its scope is the study's scope: paper, by post, adults, one health system's population. A separate web page, a second screen or a follow-up email was not tested, and nothing here claims otherwise.

One field does all the refusing

Item nonresponse among the 484 people who returned the demographic block from inside the mailed survey. In the arm where the same block arrived on a separate sheet: income 10.5 per cent, race 0.95, ethnicity 1.7, education 0.95.

One field does the refusing and it is income. Among the 484 people who filled in the block inside the survey, 16.5 per cent (80 of 484) declined the income question, against 1.9 per cent for race, 1.4 for ethnicity and 0.41 (2 of 484) for education. Income is not a sensitive-question problem in general, it is an income problem, and it deserves a design decision of its own: wider bands, an explicit reason for asking, or leaving it out.

The result runs the other way in the separate-sheet arm, where income refusal was 10.5 per cent. That looks like a win for separation until you count the people who never returned the sheet. The integrated arm produced more usable income answers overall (404 against 376) despite the higher refusal rate, which is the finding to carry: judge a design by how much data it yields, not by how clean the responding subset looks.

Sensitive items are also where wording quietly moves an answer, and the ways that happens are catalogued in response bias. Where the demographic block sits relative to the rest of the questionnaire follows the same rules as the rest of its order: screeners first, the topic in the middle, classification last.

Privacy, consent and re-identification

Two words get used as synonyms and are not. An anonymous survey is one where nobody, including you, can work out who gave an answer, which means you never collected anything that could identify them. A confidential survey is one where you can, but you promise not to and you control who sees what. Most surveys described as anonymous are confidential. Saying the wrong one is a broken promise, and respondents find out.

Demographic fields are what turns the second into the first by accident. No single field identifies anybody. Four of them together often do. Latanya Sweeney showed the shape of the problem using the 1990 US census: 87 per cent of the population was unique on the combination of five-digit ZIP code, gender and date of birth alone.¹⁵ A department, a job level, a tenure band and a location will do the same job inside a company of two hundred people.

Four rules that hold

• Collect the coarsest version that still answers your question. A region instead of a postal code. A band instead of a year. A year instead of a date.
• Set a suppression threshold before you look at the data. Do not display a subgroup result below n = 5, or whatever number you choose, and publish the threshold next to the table so nobody thinks a gap is a finding.
• Write down when the raw responses get deleted, and delete them then. A retention rule nobody enforces is worse than none, because it was promised.
• Report the opt-outs. "Prefer not to say" is a result. One respondent in six declining an income question tells you something; silently dropping those rows tells you nothing and biases everything after it.

Demographic analysis, step by step, with an example

1. Count who answered, opt-outs included

   A frequency table per field, as counts and as percentages, with "Prefer not to say" kept as a visible row rather than dropped. This is the table you will be asked for first and the one that catches a broken question fastest: an option nobody picked usually means the wording, not the population.

2. Compare that against who you meant to reach

   Put your distribution next to the population you sampled from: the employee roster, the customer list, the member database. Where there is no roster, a national statistics table using the same age bands will do. Name the gaps in the write-up. A survey that over-represents recent joiners by twenty points is still useful, but only to somebody who knows it does.

3. Cross-tab one or two outcomes, not all of them

   Pick the outcomes you would act on, put the groups down the rows, and print the group size beside every percentage. Three demographics against three outcomes is nine comparisons, which is already enough for something to look significant by chance. The mechanics of coding the answers and building the table are in the guide to how to analyse demographic groups in quantitative survey data.

   An illustration of what that buys, with made-up numbers: 400 employees, 71 per cent agreeing they have what they need to do the job. Fine, until the tenure cut: 82 per cent past three years, 54 per cent in the first six months. That is an onboarding finding, and one demographic question is the whole difference between seeing it and not.

4. Check the subgroup is big enough to say anything

   Small groups swing hard, and a twelve-point gap between two groups of eleven people is noise wearing a suit. Decide the minimum group size while you are still designing the sample, and suppress anything under your threshold in the report. Before a gap goes into a slide, assess differences between groups with a test and an interval rather than by eye.

Segments earn their keep when you can serve them differently. A tenure band that changes the onboarding programme is a segment. A gender split you report because the field was there is a table.

Frequently asked questions

How we counted

Every corpus figure here comes from one pass over the live SuperSurvey template library on 10 September 2026: 109 published templates, 1,422 questions. Question text, type, options and position were read from the published markup rather than typed in, so the counts are reproducible.

A question was counted as demographic when its wording matched a fixed keyword list: age, gender, race, ethnicity, income, education, employment, tenure, "how long have you", department, job title or role, location, country, region, marital, household, disability, veteran. That returned 41 questions. Each of the 41 was then assigned to exactly one topic in priority order, so the chart totals 41 rather than double-counting the two questions that mention two topics. Position is the question index as a share of the way through its own template, 0 for the first question and 100 for the last.

Every percentage quoted from the mailed experiment is recomputed from the raw counts in its Tables 2 and 3 (for example 489 of 1,480 delivered gives 33.04 per cent) rather than copied from the abstract, which lists the integrated and standalone arms in the opposite order from the paper's own table. The banded questions in the bank were checked mechanically: every boundary value fits exactly one option, across 30 bands in 5 questions.

What the base excludes: drafts, redirect rows and unpublished templates. What it cannot tell you: anything about surveys other organisations run. It describes what one survey vendor publishes as recommended practice, which is why every figure here is written as a share of our own library and never as a fact about surveys in general. The two-question race and ethnicity count is the clearest example: it says we rarely ask, and nothing at all about whether you should.

The legal material was read from primary text: the directive, the regulation, the statute, the rule. Where a source is a national instrument rather than a law, the reference links to the instrument itself (the questionnaire or the directive text) and the country it belongs to is named in the copy block beside it. Dates given are the dates on those documents as of 11 September 2026, and at least one of them changes ten days later.

References

1. SuperSurvey Question Corpus, 109 live templates and 1,422 questions, read 11 September 2026. Question text, type, options and position parsed from the published template markup.
2. Ziegenfuss, J. Y., Easterday, C. A., Dinh, J. M., JaKa, M. M., Kottke, T. E., and Canterbury, M. (2021). Impact of demographic survey questions on response rate and measurement: A randomized experiment. Survey Practice, 14(1). Tables 2 and 3. surveypractice.org, full text; doi.org/10.29115/SP-2021-0010
3. Office of Management and Budget (2024). Revisions to OMB Statistical Policy Directive No. 15: Standards for Maintaining, Collecting, and Presenting Federal Data on Race and Ethnicity. 89 FR 22182, 29 March 2024. Directive text with the combined question and its minimum categories. federalregister.gov, 2024-06469
4. Office for National Statistics (2021). Census 2021 paper questionnaire, England household version (PDF). The ethnic group, religion and sexual orientation questions as put to respondents. ons.gov.uk, household questionnaire PDF
5. Office for National Statistics (2022). Religion, England and Wales: Census 2021. 94.0 per cent of the population answered the voluntary religion question. ons.gov.uk
6. Washington Group on Disability Statistics (2022). The Washington Group Short Set on Functioning (WG-SS), questionnaire PDF, October 2022 edition. washingtongroup-disability.com, WG-SS questionnaire
7. Regulation (EU) 2016/679 (General Data Protection Regulation), Article 9. OJ L 119, 4 May 2016. Retained in UK law as the UK GDPR. gdpr-info.eu/art-9-gdpr
8. Commission nationale de l'informatique et des libertes (2025). Recommandation relative aux enquetes de mesure de la diversite au travail, adopted 10 April 2025, published June 2025 (PDF, in French). cnil.fr, recommendation PDF
9. California Civil Code, sections 1798.140(ae) and 1798.121 (CCPA as amended by the CPRA). leginfo.legislature.ca.gov
10. US Equal Employment Opportunity Commission. EEO-1 Component 1 Data Collection: scope, authority and the annual Instruction Booklet, which the EEOC publishes only while a filing window is open. eeoc.gov/data/eeo-data-collections
11. Office of Federal Contract Compliance Programs. Voluntary Self-Identification of Disability (CC-305), and the final rule of 21 August 2026, RIN 1250-AA18, discontinuing the form on 21 September 2026. dol.gov/agencies/ofccp/self-id-forms
12. Federal Trade Commission. Children's Online Privacy Protection Rule, 16 CFR Part 312, sections 312.2, 312.3 and 312.5; amendments published 22 April 2025, compliance due 22 April 2026. ecfr.gov
13. American National Election Studies (2021). ANES 2020 Time Series Study: Questionnaire Specifications (PDF), party identification items. electionstudies.org, 2020 questionnaire PDF
14. European Social Survey (2020). ESS Round 10 Source Questionnaire (PDF), item B26, left-right self-placement. europeansocialsurvey.org, Round 10 source questionnaire PDF
15. Sweeney, L. (2000). Simple Demographics Often Identify People Uniquely. Carnegie Mellon University, Data Privacy Working Paper 3. dataprivacylab.org, paper PDF

What to read next

• How to write survey questionsThe rules that apply before you pick a single answer option.
• Multiple choice questionsBuilding option lists that are exhaustive and mutually exclusive.
• How many responses you needBefore a subgroup gap is worth putting in a slide.