Before you start
You will be able to
- Describe the main demographic variables (age, life stage and household, gender, income, occupation, education, ethnicity) and geographic units, and what each can and can’t tell a marketer.
- Explain how geodemographic classifications such as CACI’s Acorn assign neighbourhoods to types, and the ecological fallacy that limits them.
- Explain why demographic bases are popular (measurable and accessible) and why they rarely explain why people buy.
- Analyse a behaviour against a demographic with a cross-tab, the chi-square test and Cramér’s V, and judge the strength of the association, not only its significance.
- Calculate and compare penetration rates by area or neighbourhood type to decide where to market.
Review first
- Why segment? — Segments, segmentation bases and the five tests of a usable segment are defined there.
- Analysing survey data — Cross-tabs percentaged within the explanatory variable, and the base of every figure.
- Proportions and chi-square tests — The chi-square test of independence and Cramér’s V, used here without being derived again.
- What influences buying — Culture, subculture and social class, which several demographic variables stand in for.
Fictional case A fictional coffee business founded by Ana Lim. It starts as a catering and barista-training service, opens its own cafés and an online shop for beans, and plans a canned cold brew.
The idea
Fictional case
Lantern’s media agency has proposed a Cold Brew launch campaign aimed at “adults 25 to 44 in Central Harbour City”, because that is how it buys advertising. Before agreeing, Ana wants to know whether age and region separate coffee drinkers who behave differently. Her needs study records every respondent’s region (Lantern’s five: North, South, East, West and Central), age band, household and income band, alongside how many cups of coffee they drink a week, what they spend on coffee a month and where they mostly get it.
Demographic segmentation divides a market by characteristics of people: age, life stage and household, gender, income, occupation, education and ethnicity. Geographic segmentation divides it by where people live or buy: country, region, city, neighbourhood, climate, or how urban the area is. OpenStax calls these the “who” and the “where” of segmentation. They are usually the first bases a business tries, and the reasons are practical.
The demographic variables
- Age shapes needs through the body (energy, health), through life events and through the products people grew up with. It is usually recorded in bands, as in Ana’s study: 18–29, 30–44, 45–59 and 60 or over.
- Life stage and household: whether someone lives alone, with a partner, with children or in a shared house changes budgets, time and who influences a purchase, often more than age itself. Ana’s study records single, couple, family with children and sharing.
- Gender matters for some categories and not at all for many; asking about it respectfully, with inclusive answer options, and resisting stereotypes are part of doing it well.
- Income limits what people can buy. Household income in bands is the usual measure; many people decline to answer, and income is not the same as spending on the category.
- Occupation and education relate to income and to social class (M01), and to how and when people buy: a shift worker’s coffee day differs from an office worker’s.
- Ethnicity, self-identified, can signal a subculture with preferences of its own (foods, festivals, languages). It calls for particular care: segments should be used to serve people better, never to exclude them.
Geographic units
Geography works at every scale. A national brand may segment by region; a café chain by city and neighbourhood; a delivery business by postcode. Climate drives demand for some products (iced coffee sells in heat), and population density decides which channels work: a dense city centre supports cafés on every corner, a rural area needs supermarkets and home delivery. Geography also matters because a business’s own reach is geographic: Lantern can serve only where its cafés, its delivery and its retail listings are.
Who drinks coffee where? Lantern’s study
The figure crosses age with where respondents mostly get their coffee, percentaged within each age band, as M04 taught: the band is the explanatory variable. There are real differences. Respondents aged 60 or over home-brew most (45.9%, against 19.4% of 18–29s), and 18–29s buy most often from supermarkets or as ready-to-drink cans (33.5%). By region, Central respondents use cafés most (39.5%, against 31.9% overall).
Are the differences larger than chance? B04’s chi-square test of independence says yes for age (χ² = 29.6, 9 degrees of freedom, p < 0.001) and for region (χ² = 36.9, 12 df, p < 0.001). But Cramér’s V, which measures the strength of the association on a scale from 0 to 1, is only 0.12 for age and 0.14 for region: real but modest. Every age band and every region contains plenty of café buyers, home-brewers and can buyers. Household type shows no detectable association with source (p = 0.258).
Spending and drinking tell the same story. Mean monthly spend is higher in the upper income bands: $41.62 under $40k, $45.42 at $40k–80k, $60.35 at $80k–120k and $56.45 over $120k, a clear difference (F = 15.0, p < 0.001); yet income accounts for only 6.6% of the variation in spend between respondents (η², eta squared: the share of the total sum of squares between the bands). The study finds no evidence that cups a week differ by age band (p = 0.348) or by household (p = 0.919).
In depth
Why demographics are so widely used
Demographic and geographic bases pass two of the five tests from Why segment? almost automatically. They are measurable: national censuses count people by age, household and area, so the size of a demographic segment can be estimated without a survey, and a business’s own records often hold its customers’ ages and postcodes. They are accessible: much advertising is planned and bought by audience age, sex and region, and postcodes make direct mail and local media possible. They are also easy to explain and slow to change.
They fail most often on differentiable. Russell Haley argued in 1968 that segmentations built on descriptive characteristics of buyers are poor predictors of future buying, because they are associated with buying rather than its cause; France and Ghose’s review reports similar conclusions from later work. A 35-year-old can be a budget home-brewer or a café connoisseur, and age alone can’t tell them apart.
Lantern’s study shows how mixed a demographic group can be. Respondents with incomes over $120k rate “from a local, independent business” higher on average than those under $40k (4.52 against 3.25), and income explains 11.0% of the variation in that rating, more than for any other need. Yet of the 97 respondents in the top band, 37 rate local 6 or 7 while 36 rate it 3 or lower. Targeting “high earners” for a craft message would reach about as many people indifferent to it as people who care.
Using demographic variables well
Age, life stage and generation. Record age in bands that match the decision, not round decades by habit. Life stage often explains more than age: a 30-year-old couple and a 30-year-old parent of toddlers have different mornings. A generation is a cohort born in the same span of years; labels such as “millennials” have boundaries that vary between sources. A single survey can’t separate an age effect (people change as they age) from a cohort effect (a generation keeps its habits): the 18–29s’ taste for novelty may or may not last.
Gender and ethnicity. Ask only when the answer would change a decision, offer inclusive and self-described options, and report differences without turning them into stereotypes. In many countries anti-discrimination law restricts treating customers differently, in prices or access, on grounds such as sex, race or religion, so check the law before a segment defined this way affects what anyone is offered. Used well, these variables help a business serve groups it has overlooked.
Income, occupation and education. Income in bands is easier to answer than an exact figure, but expect refusals and treat them as missing, not as zero (M04). Household income suits household purchases; personal income suits personal ones. And income measures what people could spend, not what they do spend on the category, which is why Lantern’s spend varies far more within income bands than between them.
Geographic segmentation and geodemographics
Geographic segments are defined by place: regions, cities, neighbourhoods, climate zones, or urban against rural areas. They are often the natural unit for decisions that are themselves geographic: where to open a café, which shops to sell through, which areas get the leaflet drop. Their limitation is the same as demographics’: the people in one place differ widely.
Geodemographic segmentation combines the two. It starts from the observation that neighbours tend to resemble one another, classifies every small area into types using census and other data, and describes each type with survey data. CACI’s Acorn, a widely used UK system, assigns each of the 1.6 million UK postcodes to one of 7 categories, 22 groups and 65 types, built from CACI’s address data and the 2021 Census and described with survey data from YouGov, Ipsos FRS and Kantar’s TGI. Businesses use it to profile their customers’ postcodes, to find areas whose profile matches their best customers, and to size markets locally.
Geodemographics carries a specific risk, the ecological fallacy: assuming that a characteristic of an area holds for each person in it. A postcode classified as “young families” contains retired couples and students too, and a leaflet sent to every household in it reaches them all. Classifications also age between censuses as areas change. Use the types to choose where to concentrate effort, and measure the response, rather than as a description of each household.
Testing a demographic: cross-tabs, chi-square and Cramér’s V
The test of a demographic basis is whether the groups differ in something that matters: needs, behaviour or response. For a categorical outcome such as main source, build a cross-tabulation percentaged within each demographic group (M04) and test it with B04’s chi-square test of independence, checking that every expected count is at least 5. For a numerical outcome such as spend, compare the groups’ means with one-way ANOVA (B04) and report η², the share of the variation that lies between the groups.
Then judge the strength, not only the p-value. With 640 respondents, small associations are statistically significant: Lantern’s age-by-source table gives p < 0.001 with V = 0.12. V has no universal thresholds: B04 gives none, and conventional benchmarks depend on the size of the table (the smaller of r − 1 and c − 1).
The practical question is better: would the difference change what Lantern offers or where it advertises? A band where 45.9% home-brew, against 28.8% overall, may justify a beans promotion in media that older readers use; it doesn’t make age a segmentation of coffee drinkers. And testing every demographic against every outcome will produce some “significant” results by chance, B04’s multiple-testing problem, so decide the comparisons that matter first.
Finally, remember where the data came from. Ana’s study is an opt-in panel with quotas on age band, a quota sample, so its age mix (32.2% aged 18–29, 13.3% aged 60 or over) was set by the quotas, not measured, and its regional mix (26.9% Central) needn’t match Harbour City’s. Comparisons between groups inside the sample are more trustworthy than its shares, but even they assume that panel members in each group resemble the coffee drinkers they stand for, which M04’s Sampling methods warned is an assumption, not a fact.
Formulas
Penetration rate of an area or neighbourhood type
- Customers (or buyers) living in area or neighbourhood type gcustomers
- Households (or people) in area or type ghouseholds
- Customers per 1,000 households in gper 1,000
Assumes. Customers and households are counted for the same areas and the same period; customers’ addresses are matched to the classification correctly.
Compare each area’s penetration with the overall rate, total customers ÷ total households × 1,000. Multiplying by 1,000 (or 100 for a percentage) only changes the scale.
Worked examples
Fictional figures
Kestrelton Fitness asked 950 members which time they would most like a new class. Does the preferred time differ by age band, and how strongly?
| Age band | Early morning | Lunchtime | Evening | Members |
|---|---|---|---|---|
| 18–34 | 60 | 40 | 200 | 300 |
| 35–54 | 120 | 90 | 190 | 400 |
| 55 and over | 110 | 70 | 70 | 250 |
| All members | 290 | 200 | 460 | 950 |
- Percentage within each age band, the explanatory variable: 18–34s 20.0%, 13.3%, 66.7%; 35–54s 30.0%, 22.5%, 47.5%; members 55 and over 44.0%, 28.0%, 28.0% (early morning, lunchtime, evening).
- Read across: two-thirds of 18–34s want an evening class, while members 55 and over split between early morning (44.0%) and evening (28.0%).
- Test: every expected count is at least 52.6, so the chi-square test applies. χ² = 82.6 with 4 degrees of freedom, p < 0.001.
- Strength: Cramér’s V = √(82.6 ÷ (950 × 2)) = 0.21, clearly stronger than Lantern’s age-by-source association (0.12).
Answer. Preferred time differs by age beyond chance (χ² = 82.6, p < 0.001, V = 0.21): evening dominates for 18–34s, while older members are more evenly split.
What it means. Age is a useful a priori basis here because the decision, when to timetable a class, is one that age bands do separate. An early-morning class marketed to members 55 and over, and an evening class to younger members, follows from the table. Even so, plenty of members in every band prefer other times.
Fictional figures
Copperwick Books has matched its 2,750 loyalty-card customers’ postcodes to a data supplier’s classification of its catchment into five neighbourhood types (both fictional). It can afford 20,000 leaflets, one per household, to win new customers. Which neighbourhood types should get them?
| Neighbourhood type | Households | Customers |
|---|---|---|
| City-centre renters | 9,000 | 540 |
| Young families on new estates | 12,000 | 360 |
| Established suburban owners | 15,000 | 1,350 |
| Retired village households | 6,000 | 420 |
| Student areas | 4,000 | 80 |
| Catchment | 46,000 | 2,750 |
- Penetration per 1,000 households: city-centre renters 540 ÷ 9,000 × 1,000 = 60.0; young families on new estates 360 ÷ 12,000 × 1,000 = 30.0; established suburban owners 1,350 ÷ 15,000 × 1,000 = 90.0; retired village households 420 ÷ 6,000 × 1,000 = 70.0; student areas 80 ÷ 4,000 × 1,000 = 20.0. Across the catchment, 2,750 ÷ 46,000 × 1,000 = 59.8.
- Rank by penetration: established suburban owners (90.0) and retired village households (70.0) are well above the average; city-centre renters (60.0) are at it; young families (30.0) and student areas (20.0) are below.
- Fit the budget: the two strongest types hold 21,000 households, 1,000 more than the 20,000 leaflets. Cover all 15,000 suburban households and 5,000 of the 6,000 village ones, choosing the villages nearest the shop.
- Question the logic before printing. High penetration means the shop already does well there, which suggests a good fit but also that many willing buyers are already customers. A low-penetration type, such as young families, may hold untapped demand that a different message (a children’s section) could unlock. And each leaflet reaches a household, not a type: some suburban households will never buy books.
Answer. Leaflet the established suburban owners (90.0 customers per 1,000 households) and most of the retired village households (70.0), and consider a separate test for young families rather than ignoring them.
What it means. Geodemographic types make a geographic budget decision easy to reason about, but penetration measures past success, not future response. A sensible plan spends most of the budget where the profile fits and keeps a small test where it doesn’t, then compares the response per 1,000 leaflets.
Fictional figures
Use the needs study (640 respondents) to judge the agency’s proposal: are age and region good bases for segmenting Lantern’s market, or only useful descriptors?
- Source by age: χ² = 29.6 (9 df, p < 0.001), V = 0.12. By region: χ² = 36.9 (12 df, p < 0.001), V = 0.14. By income: V = 0.10 (p = 0.027). By household: p = 0.258, no evidence of an association. All the significant associations are weak.
- Spend and cups: income explains 6.6% of the variation in monthly spend; the study finds no evidence that cups a week differ by age or household (p = 0.348 and p = 0.919).
- Needs: the strongest demographic link is income with the importance of local (η² = 0.110), followed by income with taste (0.096); age’s strongest link is with novelty (0.062): 18–29s rate new flavours highest. Even the strongest leaves nearly nine-tenths of the variation within the groups.
- Apply the five tests: age and region are measurable and accessible, but the groups barely differ in behaviour, so they fail differentiable as segments of coffee drinkers.
Answer. Age and region are weak bases for Lantern: they are associated with where people buy and what they spend, but every age band and region mixes all kinds of coffee drinkers. Use them as descriptors to reach segments formed on needs, not as the segments themselves.
What it means. The agency’s “25 to 44 in Central” isn’t wrong as a media buy if the segment Lantern targets turns out to be concentrated there. But that has to be shown after the segments are formed (module 2’s profiling lesson), not assumed before.
Explore
Common mistakes
“The difference is significant, so age is a good segmentation base.”
With hundreds of respondents, weak associations are significant: Lantern’s age-by-source table has p < 0.001 but V = 0.12. Judge the strength (V, η², the size of the gaps in percentage points) and whether it would change a decision.
“High earners want premium, local coffee.”
On average they rate local higher, but 36 of the 97 top-band respondents rate it 3 or lower. An average difference between groups is not a description of every member: test for spread within the group before targeting it.
“Everyone in a ‘young families’ postcode is a young family.”
That is the ecological fallacy. A neighbourhood type describes the area’s typical profile, not each household. Use geodemographic types to choose where to concentrate effort, and measure the response.
“The survey shows that a third of Harbour City’s coffee drinkers are 18–29.”
The age mix of a quota sample is set by the quotas. Use the sample to compare groups, and take population shares from census data or a probability sample.
“Income tells us what people spend on coffee.”
Income bounds what people can spend; it explains only 6.6% of the variation in Lantern’s respondents’ monthly coffee spend. Measure category spending directly when it matters.
Guided practice
Work through these with hints when you need them. Every hint and every worked solution is there to be used.
Loading four questions…
On your own
No hints this time. Answer, check, and read the feedback; the solution appears once you have answered or if you ask for it. Your first answers count toward mastery.
Loading five questions…
In practice
Fictional case
Fernhollow Travel, a fictional coach-holiday company, wants to fill its off-season tours. Owen Pryce, its marketing analyst, has the bookings of the last three years and access to census counts and a geodemographic classification for the region.
He starts with what the business already knows. Matching past bookers’ postcodes and ages shows that retired couples book most off-season tours. That is an a priori demographic segment, and it passes two tests at once: census tables give the number of retired couples in each area (measurable), and the company can reach them by post and through local newspapers (accessible).
He sizes it before he targets it. The census gives about 41,000 retired couples within reach of the company’s pick-up points (fictional). At last year’s rate of 1.2% booking an off-season tour, that is about 492 couples a year: a substantial segment for a coach company, but one Fernhollow already reaches well.
He checks whether the demographic explains the behaviour. A short survey of past bookers asks why they chose an off-season tour. 65.0% of the retired couples say the lower price, but many mention quieter destinations or a hotel included; younger bookers without children give the same reasons. The demographic describes most of the bookers, but the reasons cut across it, so Owen recommends messages built around the reasons (“quieter, cheaper, all included”) placed in media that reach retired couples first.
He uses geography for the media plan, with a check. The geodemographic types with the most past bookers per 1,000 households get the brochure drop; two lower-scoring types get a smaller test drop, so the company learns whether it has been missing demand rather than assuming the classification is the whole story.
Knowledge check
Six questions, one attempt each, no hints. Score 80% or more to complete the lesson; you can take it again at any time.
Loading six questions…
Summary
- Demographic segmentation divides a market by people’s characteristics (age, life stage and household, gender, income, occupation, education, ethnicity); geographic segmentation by place (region, city, neighbourhood, climate, urban or rural).
- They are popular because they are measurable (censuses, customer records) and accessible (media and data are sold by them), and easy to explain.
- They describe who buys far better than why: within every demographic group there are buyers of every kind, so they often fail the differentiable test as segments, while making excellent descriptors.
- Geodemographic systems such as CACI’s Acorn classify every postcode into types (Acorn: 7 categories, 22 groups, 65 types); beware the ecological fallacy of treating an area’s type as each household’s.
- Test a demographic with a cross-tab percentaged within the group, B04’s chi-square test and Cramér’s V, or ANOVA and η² for a numerical outcome; judge strength and usefulness, not only the p-value.
- Penetration = customers ÷ households × 1,000 compares areas or neighbourhood types; it measures past success, not future response.
- In Lantern’s study, age, region and income are associated with where people buy and what they spend, but weakly (V about 0.10–0.14; income explains under 7% of spend).
Key formulas
Terms introduced
- Demographic segmentation
- Dividing a market by characteristics of people such as age, life stage and household, gender, income, occupation, education and ethnicity.
- Geographic segmentation
- Dividing a market by place: country, region, city, neighbourhood, climate or how urban an area is.
- Life stage
- A person’s position in the course of life as shaped by household and family, such as living alone, living with a partner, raising children or retired; it often explains needs better than age alone.
- Generation
- A cohort of people born in the same span of years, such as those born in the 1980s; the boundaries of named generations vary between sources.
- Geodemographic segmentation
- Classifying small areas, such as postcodes, into types of neighbourhood using census and other data, and describing each type with survey data, on the principle that neighbours tend to resemble one another.
- Ecological fallacy
- The error of assuming that a characteristic of a group or area, such as a neighbourhood’s geodemographic type or average income, holds for each individual in it.
- Penetration rate
- The number of a business’s customers in an area or group per 1,000 (or per 100) households or people there; it compares how well the business has done in different areas.
Sources
- Gomez Albrecht, M., Green, M. & Hoffman, L. (2023). Principles of Marketing, section 5.1, “Market Segmentation and Consumer Markets” (geographic, demographic, psychographic and behavioural segmentation). OpenStax. openstax.org/books/principles-marketing/pages/5-1-market-seg…
- CACI. “What is Acorn?” (postcodes, categories, groups and types; data sources and uses). acorn.caci.co.uk/what-is-acorn
- CACI. “How Acorn works.” acorn.caci.co.uk/how-acorn-works
- Haley, R. I. (1968). Benefit segmentation: a decision-oriented research tool. Journal of Marketing, 32(3). doi:10.1177/002224296803200306 (abstract: descriptive characteristics of buyers are poor predictors of future buying). doi.org/10.1177/002224296803200306
- France, S. L. & Ghose, S. (2019). Marketing analytics: methods, practice, implementation, and links to other fields. Expert Systems with Applications; preprint arXiv:1801.09185 (section 4 on segmentation bases). arxiv.org/abs/1801.09185