Quantitative market research: measure what the business needs to decide
Quantitative market research measures defined questions across an audience or condition. Specify what counts, who each percentage represents and which comparison changes the decision. Keep understanding, stated preference and observed behavior as separate measures.
The first quantitative decision is what to measure. For the grocery offer, deadline understanding can change wording, usefulness can change fit and observed ordering can evaluate a commercial intervention. Mixing them into one score leaves marketing unsure what to do. The six records below are fictional teaching values.
Define a variable that can change an action
This fictional study asks whether a grocery delivery proposition is clear enough to progress to a usability test. The proposition reads: “Select a two-hour delivery window at checkout. Changes are allowed until the evening before delivery.”
Six records reveal more than one average
Record | Deadline understood | Usefulness |
|---|---|---|
Q1 | Yes | 5 |
Q2 | Yes | 4 |
Q3 | No | 4 |
Q4 | Yes | 3 |
Q5 | No | 2 |
Q6 | Yes | Cannot judge |
Fictional exercise. Understanding: 4/6. Ratings 4–5: 3/5 substantive ratings, with one Cannot judge. Q3 is an appealing misunderstood offer; Q4 is understood but less useful.
The audience is recent grocery-delivery users. The decision is whether to revise this wording. The primary measure is correct understanding of the change deadline, assessed before showing answer choices. A separate relevance rating explores usefulness. Neither measure tests operational delivery performance.
For a real study, define recent use, geography, recruitment and exclusions. Decide whether the evidence concerns eligible respondents, a target human population or a modeled population. Do not let the same chart quietly stand for all three.
Start with the unit and opportunity: an eligible buyer shown a particular offer, an account facing renewal or a customer able to place an order. Then define the response or behavior. A measure is more useful when the decision owner knows what a high or low result would change.
For a brand question, that may be awareness under a specified prompt. For a concept, it may be correct unaided interpretation. For an intervention, it may be an observed outcome under assigned conditions. Those measures can all be quantitative while serving different claims.
Specify the questionnaire and the stored values together
- Have you received a grocery delivery in the past 30 days? Yes; No; Prefer not to answer. Continue eligible respondents only.
- Show the proposition exactly as written. When can you last change the delivery window? Open response.
- How useful would this arrangement be for your household? 1 Not at all; 2 Slightly; 3 Moderately; 4 Very; 5 Extremely; Cannot judge.
- What information is missing from this description? Open response.
Variable | Type and allowed values | Rule |
|---|---|---|
record_id | Unique text | One eligible record per participant |
recruitment_source | Text | Preserve actual source, not an inferred demographic |
deadline_text | Original text | Store before coding |
comprehension | 1 correct; 0 incorrect; missing unanswerable | Correct requires the evening before delivery |
usefulness | Integer 1–5; Cannot judge; missing | Do not encode “Cannot judge” as 3 |
information_needed | Text | Code separately from comprehension |
The data dictionary prevents an interface choice from quietly changing the measure. If Cannot judge becomes 3, uncertainty becomes moderate usefulness. If missing comprehension becomes 0, unanswerability becomes an incorrect answer. The analysis must preserve the distinction chosen in the instrument.
Plan any group cuts before collection. A new/repeat split needs a defined history; a service zone needs the relevant geography. Collect only the context that can change a planned comparison or help explain a diagnostic.
Calculate the result from records, not an overall average
The following six records are fictional demonstrations, not survey findings.
Record | Comprehension | Usefulness |
|---|---|---|
Q1 | 1 | 5 |
Q2 | 1 | 4 |
Q3 | 0 | 4 |
Q4 | 1 | 3 |
Q5 | 0 | 2 |
Q6 | 1 | Cannot judge |
Correct comprehension is 4/6. Top-two usefulness is 3/5 among substantive ratings. Its denominator is five because “Cannot judge” is reported separately. An ordinal scale’s distribution is more informative here than a mean that could conceal misunderstanding.
If a provisional gate required all six demonstration records to understand the deadline, the wording would fail. That is an instructional decision rule, not an industry standard. Inspect incorrect responses and revise the deadline sentence before testing again.
Read an incorrect answer before revising
Fictional record Q3 says: “I can change it during the two-hour slot.” Under the stated evening-before rule, code comprehension as 0 even though usefulness is 4. Q1 says: “No later than the evening before it arrives,” which receives 1. The favorable rating does not repair the deadline misunderstanding. Test a revised sentence that puts the deadline first, while holding the offer and exposure conditions constant.
An ordinal usefulness mean could conceal Q3: it rates 4 but misunderstands the deadline. It could also conceal Q4: it understands but rates 3. The record pattern gives the team two different problems rather than a single low score.
Use the distribution and cross-pattern that address the action. It is unnecessary to choose the most elaborate statistic if a count and an original answer identify the information gap. Complexity should answer a question the simpler readout cannot.
Compare audiences under the same complete offer
A group difference can help choose an offer when exposure, eligibility and response base are comparable. If the first-time group sees one delivery condition and repeat users see another, audience and offer changed together. Label the comparison accordingly.
Here comprehension uses six records, usefulness uses five substantive ratings and one Cannot judge answer is separate. A percentage on all six answers a different question from the percentage among five ratings. Put the base beside every result so the reader can see which question is being answered.
For a real sample, assess who is reachable, who responds and how the intended inference is supported. A whole-study total does not guarantee useful subgroup bases. For modeled comparisons, record audience grounding and relevant validation rather than assigning a human sampling margin from the number of generated profiles.
Let the readout choose the next investigation
Ask for the question-level distributions, valid bases and the records that explain incorrect answers. Keep routing, missing values and recruitment coverage in the supporting analysis. A number without its question and audience cannot tell marketing what to change.
Sampling uncertainty depends on the human design and estimator. For modeled comparisons, disclose grounding and task-relevant validation; generating more profiles does not create more independent human observations. The AAPOR disclosure standards provide a reference for reporting the design.
Use the analysis walkthrough for the data and survey guide for fieldwork. Discuss a structured comparison with Sapien when the result can choose a subsequent offer or validation study.
The completed wording decision is a deadline-first candidate with unchanged offer terms. Retest unaided understanding. If clarity is restored but usefulness is still weak, inspect the household occasion or compare another complete arrangement. If the team needs ordering effect, use an appropriate behavioral design after the offer is understood.
Quantitative work earns its place by making those branches visible. A tidy dashboard full of percentages is insufficient when nobody knows which product, message or service choice each measure controls.
