Define the decision and the population first
A fictional campus service is comparing two laundry-pickup concepts. The business question is which offer should receive further development. The research measure is stated interest under a specified description, not actual purchase, willingness to pay or future retention. Keep those distinctions in the brief.
Specify who the intended customers are, how participants would be recruited and which offer details remain constant. Price, pickup location and turnaround time can all change an answer. A test that varies several elements at once needs a design that supports interpreting those changes; otherwise you cannot isolate which feature drove the result.
AAPOR’s research guidance stresses careful questions, suitable sampling and transparent reporting. Use that as a foundation for a small project, then make the assumptions of your own exercise visible. The numbers here are synthetic and support a calculation lesson, not a claim about student preferences.
Sources[1]
Inspect the four rows before calculating one total
Concept A was shown to 20 commuters and 80 residential students. Concept B was shown to 80 commuters and 20 residential students. Within both groups, A has a ten-percentage-point advantage: 70% versus 60% for commuters and 30% versus 20% for residential students.
| Concept | Segment | Respondents | Interested | Segment interest rate |
|---|---|---|---|---|
| A | Commuter | 20 | 14 | 70% |
| A | Residential | 80 | 24 | 30% |
| B | Commuter | 80 | 48 | 60% |
| B | Residential | 20 | 4 | 20% |
However, the pooled numbers look different. A has 38 interested respondents out of 100, while B has 52 out of 100. Declaring B the winner from the 52% versus 38% totals would ignore that its sample contains many more of the high-interest commuter group.
Compare the same segment mix without calling it a market estimate
For an illustrative 50–50 segment mix, A’s standardized rate is half of 70% plus half of 30%, or 50%. B’s is half of 60% plus half of 20%, or 40%. The sign of the comparison now matches both segment-level contrasts.
The equal mix is a stated teaching assumption. It is not an estimate of the campus population. To estimate a target population, you would need credible composition information, an appropriate sample and a defensible adjustment method. Reweighting observed segments does not fix every selection problem or account for people missing from the recruitment process.
Also distinguish an observational comparison from a randomized test. This four-row table does not establish random assignment within segments or equivalent recruitment. It illustrates how composition affects aggregation; it does not prove that concept A causes higher interest.
Write questions that do not advertise the preferred answer
Replace a question such as “Would you use our convenient, affordable service?” with a neutral description of what the concept actually offers, followed by a clear response question. The adjectives already tell the participant which conclusion the writer hopes to obtain.
An original short instrument could ask about the person’s recent laundry behavior, which collection options they have used, what makes the process difficult, and how likely they would be to consider the precisely described offer. Provide a way to say the concept is irrelevant or unclear where that is a meaningful response.
Keep one question from silently combining different judgments. Liking the pickup location and accepting the turnaround time are not necessarily the same answer. If the decision requires knowing which constraint matters, ask in a way that lets you distinguish them. Do not lengthen a survey simply because another variable would be interesting.
Write the finding with its denominator and limitation
A supported headline for the exercise is: “The pooled comparison favors B because its sample contains more commuters; within each supplied segment, A has the higher stated-interest rate.” Follow it with the actual counts and the 50–50 sensitivity, explicitly labelled as a standardization example.
The next decision might be to run a more comparable concept test or investigate why the groups respond differently. Choose it based on what uncertainty blocks the business decision. A small set of interviews can investigate reasons, but quotations from those interviews do not establish population percentages.
Do not attach a conventional random-sample margin of error to an unspecified convenience sample, and do not convert stated interest into a sales forecast. Those would require additional assumptions and evidence not supplied by this exercise.
Deliver the instrument, data and memo together
A useful internship work sample includes the decision brief, proposed questions, recruitment approach, data dictionary, four-row calculation and a one-page findings memo. The memo should separate the fictional results from the proposed next test. A reviewer should be able to recalculate the pooled and standardized figures without opening a presentation.
Add a sensitivity using a 25% commuter and 75% residential mix. A becomes 40% and B 30%, again under that assumed mix. Explain why you should not select the weights solely to make a preferred answer appear.
In the application, describe the task accurately: diagnosed an aggregation problem in a synthetic concept test and wrote a neutral follow-up research plan. That is substantive work. Claiming that you validated product-market fit or predicted real adoption would overstate it.