You are viewing a free preview of this lesson.
Subscribe to unlock all 13 lessons in this course and every other course on LearningBro.
Spec mapping: AQA 7138 Unit 3.1.3 — Marketing Management (refer to the official AQA specification document for exact wording). This lesson develops sampling and data interpretation at A-Level depth — the formal logic of probability sampling, the analytic distinction between random, stratified, cluster and quota methods, confidence-interval interpretation and statistical-significance basics, correlation (Annex 8 #d3), extrapolation, and the evaluative judgement an examiner expects when a business is deciding whether a small-sample finding is trustworthy enough to act on.
Connects to:
A sample is a deliberately selected subset of a larger population (the target group the business actually cares about), drawn so that inferences from the sample can be generalised back to the population with known statistical confidence. Businesses sample because population-wide censuses are usually impossible, illegal under GDPR, or simply uneconomic. The sample is a trade-off device — it sacrifices completeness for tractability, on the condition that the inference back to the population remains defensible.
Definition: Sampling is the deliberate selection of a subset of a population, designed so that statistical inferences from the sample can be generalised — with known precision — back to the population the business wants to understand.
The four reasons businesses sample:
| Reason | A-Level depth |
|---|---|
| Cost | A census is typically several orders of magnitude more expensive than a sample large enough to support the decision being taken. |
| Time | A 2,000-respondent online survey can return fieldwork in 10 days; a population-wide census would take months or years. |
| Practicality | Many target populations cannot be enumerated — there is no comprehensive list of "UK speciality coffee buyers". |
| Sufficiency | A well-designed sample is genuinely sufficient — past a sample size threshold, additional respondents add precision at sharply diminishing marginal returns. |
The point is not that a sample is a regrettable second-best. A well-designed sample of 1,500 can support a defensible go/no-go national launch decision; a badly designed sample of 50,000 cannot. Design dominates size.
Every member of the target population has an equal and known probability of being selected. Requires a sampling frame (a complete enumeration of the population) — a database, an electoral roll, a customer list.
Strengths: statistically valid; confidence intervals defensible; eliminates systematic selection bias by construction. Weaknesses: a complete sampling frame often does not exist; pure-random samples can still produce demographically lopsided draws by chance; expensive if the population is geographically dispersed.
Divide the population into mutually exclusive strata (age, gender, region, customer-segment), then randomly sample within each stratum in proportion to its population share.
Strengths: guarantees representation of key subgroups; smaller required sample size for a given precision compared with simple random sampling; permits subgroup analysis. Weaknesses: requires accurate population-strata data; if the wrong stratification variables are chosen, the gains evaporate; more complex to administer.
Randomly select groups (geographic areas, branches, schools) and then sample everyone (or a sub-random sample) within selected clusters. Often used where the population is geographically dispersed and travel costs dominate.
Strengths: cheaper than simple random for geographically dispersed populations; useful when no individual-level sampling frame exists. Weaknesses: higher sampling error than stratified for the same total sample; clusters are often internally homogeneous, so within-cluster variation under-represents population variation.
Set quotas for subgroups (e.g. 100 men 18-30, 100 women 18-30, 100 men 31-50, 100 women 31-50) and fill them by non-random selection.
Strengths: cheap, fast, no sampling frame required, ensures subgroup coverage. Weaknesses: selection bias — interviewers approach approachable-looking respondents and avoid those in a hurry; not a probability sample, so confidence intervals and significance tests are technically invalid.
The researcher selects respondents who are convenient (in-store intercepts, social-media volunteers, mall-intercept). Used widely in pilot studies but should be flagged as such.
Strengths: cheapest method; rapid; useful for early-stage exploration. Weaknesses: highest bias risk; cannot support inferential generalisation; should never be the sole basis for a high-stakes decision.
| Criterion | Random | Stratified | Cluster | Quota | Convenience |
|---|---|---|---|---|---|
| Sampling frame required | Yes | Yes (with strata) | Cluster list only | No | No |
| Statistical validity | High | High | Moderate | Low | None |
| Cost | High | High | Medium | Low | Lowest |
| Speed | Slow | Slow | Medium | Fast | Fastest |
| Bias risk | Low | Low | Medium (intra-cluster) | High (interviewer) | Highest |
| Best for | Defensible inference | Subgroup-rich populations | Geographically dispersed targets | Rapid commercial research | Pilots and exploration |
The 7138 specification explicitly introduces confidence levels and intervals as A-Level-required statistical literacy.
Definition: A confidence interval is the range within which the true population value is estimated to lie, at a stated confidence level (typically 95 % or 99 %). A 95 % CI does not mean "95 % probability the true value lies in this interval" — it means "if we repeated this sampling process many times, 95 % of the resulting intervals would contain the true population value".
The diagnostic interpretation:
Three factors determine width:
| Factor | Direction | Mechanism |
|---|---|---|
| Sample size | Larger → narrower | Reduces sampling variability |
| Confidence level | Higher → wider | 99 % requires more room than 95 % to be more certain |
| Underlying variability | Higher → wider | More variation in responses requires more room |
Doubling sample size from 400 to 1,600 roughly halves the confidence-interval width, but quadruples the fieldwork cost. The right sample size is the minimum large enough to support the decision, not the largest the budget can buy.
Statistical significance at A-Level depth: a finding is statistically significant if it is unlikely to have arisen by random sampling variation alone. A 12-point favourability gap between two brands in a sample of 200 may be statistically significant (unlikely to be chance) and yet commercially trivial (the gap is too small to change the marketing plan). Significance and commercial materiality are separate tests; serious A-Level evaluation respects both.
Correlation (Annex 8 analytical concept #d3) measures the strength and direction of the relationship between two variables. The correlation coefficient r ranges from −1 (perfect negative) through 0 (no linear relationship) to +1 (perfect positive).
| r value | Interpretation |
|---|---|
| ±0.9 to ±1.0 | Very strong |
| ±0.7 to ±0.89 | Strong |
| ±0.4 to ±0.69 | Moderate |
| ±0.1 to ±0.39 | Weak |
| 0 | No linear relationship |
The crucial A-Level discipline: correlation does not establish causation. A correlation between two variables can reflect:
The textbook illustration: ice-cream sales and drowning deaths are positively correlated, but neither causes the other — hot weather drives both. The commercial-research illustration: social-media follower growth and revenue growth often correlate, but both may be driven by a third variable (a successful TV campaign) rather than the followers driving the revenue.
flowchart TD
Pop["Target population<br/>(business cares about this)"] --> Sample["Sample design<br/>(random / stratified / cluster /<br/>quota / convenience)"]
Sample --> Data["Sample data"]
Data --> Stats["Confidence interval<br/>+ significance test"]
Stats --> Infer["Inference back to population<br/>(with known precision)"]
Data --> Corr["Correlation analysis"]
Corr --> Cause{"Causal claim<br/>justified?"}
Cause -->|"No (most cases)"| Hyp["Treat as hypothesis"]
Cause -->|"Yes (with controlled design)"| Act["Act on causal interpretation"]
Infer --> Decision["Commercial decision"]
Hyp --> Decision
style Sample fill:#1d4ed8,color:#fff
style Decision fill:#15803d,color:#fff
style Cause fill:#a16207,color:#fff
The crucial node is the causal claim branch. Most commercial research never earns the right to a causal interpretation — controlled experiments do, observational research does not. Treating a correlation as causation is the most common A-Level error and the most common commercial mistake.
Highmoor Apparel is a UK mid-market clothing brand considering a major spring launch of a new £85 women's outerwear range targeting urban professionals aged 28-42. The marketing team commissioned a research study to assess purchase intent. The study used two methods. Method A — an online stratified survey of 1,800 respondents stratified by age, region and household income, generating purchase-intent estimates of 23 % ± 2.1 % at 95 % confidence. Method B — a single in-store intercept survey of 120 shoppers at one flagship London store on a Saturday afternoon, reporting purchase intent of 41 %. The two findings appear to conflict. The marketing director has a £1.2 million launch budget riding on the decision and must decide which finding to trust.
Figures fabricated for illustrative purposes; not affiliated with any actual business.
Assess whether Highmoor Apparel should trust the small-sample in-store result (Method B) over the stratified online survey result (Method A) when deciding whether to commit the £1.2 million launch budget. (9 marks)
| AO | What the question rewards | Mark weighting on this 9-mark item |
|---|---|---|
| AO1 | Knowledge of stratified vs convenience sampling, confidence intervals, statistical significance | ~2 marks |
| AO2 | Application to Highmoor's specific context — £1.2m launch decision, urban-professional target, online vs single-store findings | ~2 marks |
| AO3 | Analytical chain-of-reasoning linking sample design to the trustworthiness of each finding | ~3 marks |
| AO4 | Evaluative judgement weighing the two methods against the decision being taken | ~2 marks |
The platform's general guidance: 9-mark Assess questions reward a structured "for / against / on balance" build supported by chain-of-reasoning, not exhaustive coverage. Pick two strong arguments per side and develop them in depth.
Subscribe to continue reading
Get full access to this lesson and all 13 lessons in this course.