From question to design: answer key
Introduction to Social Research Methodology
All model answers below follow the session-3 lecture (the notes’ sections on hypotheses, variables, units of analysis, designs and the Meridian mini-proposal), and they agree with the answers ticked in the session-3 class poll.
Task 1: from questions to hypotheses
There is no single correct hypothesis per question. Any answer earns full credit if the hypothesis is testable and falsifiable, clear and specific, states a relationship between named variables, has a plausible rationale and is value-neutral — and if the unit of analysis, the variables and the control are consistent with it. Figure 1 gives one good answer per question.
Q1. Does the quality of first-line management affect staff turnover across Meridian’s 32 stores?
- Hypothesis: Stores whose employees report lower supervisor support have higher annual turnover. This is the lecture’s own example: directional, falsifiable (no association, or the opposite one, would refute it), and it names both variables.
- Unit of analysis: stores. The question compares turnover across stores, and turnover is a property of a store, not of a person.
- Independent variable: quality of first-line management, operationalised for example as the average supervisor-support score reported by each store’s staff.
- Dependent variable: the store’s annual staff turnover rate.
- Control: pay level in the store — the lecture’s worked rival explanation. If weaker managers happen to sit in lower-paying stores, pay rather than management could produce the pattern. Local unemployment, store size and the age or tenure mix of staff are also good answers.
Q2. Is customer satisfaction lower in stores with higher staff turnover?
- Hypothesis: Stores with higher staff turnover have lower average customer satisfaction scores.
- Unit of analysis: stores.
- Independent variable: staff turnover. This is the lecture’s point that roles belong to the hypothesis, not to the variable: turnover was the dependent variable in Q1 and becomes the independent variable here.
- Dependent variable: the store’s average customer satisfaction score.
- Control: staffing level, because understaffed stores may both lose staff and serve customers badly, a common cause of both. Store location, footfall and local competition are also acceptable. Pairs who notice that causation might also run the other way (unhappy customers make the job worse, and staff leave) deserve credit: that is a reason to prefer a design with time in it.
Q3. Do employees who complete Meridian’s onboarding programme stay with the company longer than those who do not?
- Hypothesis: Employees who complete the onboarding programme stay with Meridian longer than those who do not. A version with a fixed horizon (are less likely to leave within twelve months) is equally good.
- Unit of analysis: individual employees — it is people who complete onboarding and people who leave.
- Independent variable: onboarding completed (yes or no).
- Dependent variable: length of stay, for example months until leaving.
- Control: contract type, because temporary or student staff may both skip onboarding and leave sooner by design. Role, store and prior experience are also good answers. The strongest pairs will see the deeper problem: people who complete onboarding may simply be the more committed ones, so self-selection is itself a rival explanation.
Marking guidance for Task 1
- Apply the falsification check from the worksheet. If the pair cannot say what result would prove the hypothesis wrong, it is not yet a hypothesis. “Management may or may not affect turnover” is the commonest unfalsifiable version.
- A control must come with its reason. Listing “age” without saying what rival story it rules out earns half credit at most.
- Watch the unit of analysis. A pair that answers Q1 at the individual level (“employees with worse managers quit more”) has written a different study. That is not wrong in itself, but it must then say so, and it should not draw store-level conclusions from individual data, or the reverse: that is the ecological fallacy.
- Directional and non-directional hypotheses are both acceptable, as long as the choice is deliberate.
Task 2: matching designs to managerial problems
Figure 2 gives the best-fitting design for each scenario, why it fits and the limitation to flag to the board.
| Scenario | Model design | Also defensible | Limitation to flag |
|---|---|---|---|
| A. Two Kraków stores | Comparative | A two-case case study | The stores still differ in many ways; each difference is a rival explanation |
| B. Unpaid overtime before pay talks | Cross-sectional | — | No trend, so it cannot test “now routine”; self-reports may be biased |
| C. Shift-swap app | Quasi-experimental | — | The pilot stores chose themselves |
| D. Sixty September hires | Longitudinal (cohort, followed as a panel) | — | Attrition: the leavers drop out |
A — comparative. Two stores alike in age, size and pay but opposite in turnover are a textbook comparative design: the similarities hold the obvious explanations constant, so the contrast points to whatever else differs (management style, staff mix, scheduling). A two-case case study, digging into how each store actually works, is a defensible second answer, and a good pair may combine the two. The limitation: compared cases differ in many ways at once, and every difference is a rival explanation that must be ruled out one by one.
B — cross-sectional. The finance director needs to know how common unpaid overtime is, store by store and role by role, within three weeks: a fast snapshot of all 32 stores is exactly what a cross-sectional design does best. The limitation: a snapshot shows no trend, so it cannot test the unions’ claim that overtime is now routine (that would need earlier data), and self-reported overtime may be over- or under-stated. Linking the survey to time-sheet records strengthens it.
C — quasi-experimental, made experimental next time. The existing evidence is quasi-experimental: five stores with the app, 27 without, but the five chose themselves, so their managers may simply be the more engaged ones, whose absences would have fallen anyway. The right next step is to make the next phase a true experiment by randomly choosing which stores get the app next: randomisation is, in the lecture’s words, a universal control variable. Two weaker answers are worth correcting in the plenary. Comparing the five stores before and after March is not enough, because absences change with the season and with everything else that happened in spring. And giving the app next to the stores with the most absences builds in regression to the mean.
D — longitudinal. Seeing how commitment develops and catching the point where it starts to slip requires measuring the same people repeatedly: a cohort (one September intake) followed as a panel. The limitation is attrition, and here it is acute, because the leavers, who matter most, drop out of the study. Keep their HR exit records, and if possible ask for a short exit interview.
Marking guidance for Task 2
- The justification matters more than the label, as the worksheet says. A design with a strong justification and an honest limitation beats the model label with neither.
- Each answer needs both halves: what the design does well for this problem and one limitation for the board. Answers without a limitation earn half credit.
- Watch for “experiment” as a default. Only scenario C involves an intervention; proposing to randomise pay rates (A) or overtime (B) is neither practical nor ethical.
Task 3: a one-page proposal skeleton
Any of the three Task 1 questions can be used. The lecture’s own mini-proposal worked Q1, so Figure 3 works Q3, as one good skeleton among several.
A complete skeleton has a problem with stakes (why the board should pay for the study), a focused question, a falsifiable hypothesis with named controls, a design matched to the question and the constraints, and a realistic data inventory that mixes new data with records Meridian already holds.
Survey mode. The worksheet asks for a mode and a reason. CAWI is the cheapest and fastest and suits the e-commerce division and head office, where everyone works at a screen; but store staff may have no company e-mail and little time at a computer during shifts. A mixed mode — PAPI (or a tablet) in the stores, CAWI elsewhere — is the strongest answer. A single mode is acceptable if the pair says whom it might miss.
Wrap-up
There is no model answer to “which move was hardest”. Use the answers diagnostically: pairs who name assigning the variables usually struggled with roles that change between hypotheses (Q1 and Q2); pairs who name choosing the design usually treated the typology as a menu rather than asking what each problem needed. Both are exactly what sessions 8 (measurement) and 9 (mixed methods) return to.