Hypotheses and research design

Introduction to Social Research Methodology

Author
Affiliation

Ben Stanley

Department of Social Sciences, SWPS University

Published

November 3, 2026

From questions to hypotheses

A research question tells you what you want to know; it does not yet tell you what you expect to find or how you will find it. This session covers the two steps that turn a question into a workable plan: formulating a testable hypothesis, and choosing a research design capable of testing it. The running example throughout is Meridian, a Polish retail-and-services firm with 450 employees, 32 stores, and a growing e-commerce arm, where annual staff turnover has risen from 14% to 27% in two years and the board wants evidence rather than hunches.

Research assumptions

Every study rests on assumptions — statements accepted as true for the purposes of the research without being tested directly. Three kinds recur:

Kind Meaning Meridian example
Theoretical Propositions taken over from theory or prior research Job dissatisfaction increases the likelihood of quitting
Methodological Beliefs about how the chosen methods behave Employees will answer an anonymous survey honestly
Practical Beliefs about the data and setting HR records of leavers are accurate and complete

Assumptions are unavoidable, because no study can test everything at once. What distinguishes good research is that assumptions are stated explicitly, so that readers can judge whether they are reasonable. A hidden assumption is a weakness; a stated one is a boundary of the study.

What is a hypothesis?

A hypothesis is a tentative, testable statement about the relationship between two or more variables. It is not a question but a proposed answer: a specific expectation that data can support or contradict. Hypotheses are the engine of deductive research — theory suggests the hypothesis, and the study is built to test it. Compare the two forms: the question “does the quality of store management affect staff turnover at Meridian?” becomes the hypothesis “stores with poorer-quality first-line management have higher staff turnover”. The hypothesis commits the researcher: by stating what is expected, it also specifies what evidence would prove the expectation wrong.

Figure 1 draws the move from question to hypothesis: the question becomes a proposed answer, an arrow from the independent variable (quality of first-line management) to the dependent variable (staff turnover), labelled with the expected direction. The three cards underneath name what makes it a hypothesis: two named variables in a stated relationship, a commitment to what would prove it wrong, and its place as the engine of deductive research.

Anatomy of a hypothesis At the top, a question: does the quality of store management affect staff turnover at Meridian? Below it, a proposed answer, the hypothesis: an arrow runs from the independent variable, quality of first-line management, to the dependent variable, staff turnover, labelled poorer to higher. In words: stores with poorer-quality first-line management have higher staff turnover. Three cards: at least two variables, named and linked by a stated relationship; a commitment, because it says what evidence would prove it wrong, namely no difference in turnover or the opposite one; and the engine of deduction, because theory suggests the hypothesis and the study is built to test it. Question: does the quality of store management affect staff turnover at Meridian? a proposed answer HYPOTHESIS INDEPENDENT VARIABLE Quality of first-line management DEPENDENT VARIABLE Staff turnover poorer → higher “Stores with poorer-quality first-line management have higher staff turnover.” At least two variables named, and linked by a stated relationship — not a description of one thing A commitment it says what evidence would prove it wrong: no difference in turnover, or the opposite one The engine of deduction theory suggests the hypothesis; the study is built to test it
Figure 1: Anatomy of a hypothesis.

Properties of a good hypothesis

A good hypothesis satisfies five criteria:

Property Meaning
Testable and falsifiable Observable evidence could, in principle, show it to be false
Clear and specific The variables and the expected relationship are stated precisely
A statement of relationship It links at least two variables rather than describing one thing
Grounded It follows from theory, prior research, or informed observation
Value-neutral It predicts what is, not what ought to be

“Bad management is unacceptable” fails every test: it is a value judgement, names no measurable variables, and no evidence could falsify it. “Stores whose managers score lower on supervisor support have higher turnover” passes them all.

Types of hypotheses

Two distinctions organise the types of hypothesis. The first concerns direction. A non-directional hypothesis states that a relationship exists without specifying which way it runs (“manager quality is related to staff turnover”). A directional hypothesis specifies the expected direction (“the poorer the manager quality, the higher the staff turnover”). Directional hypotheses are riskier and therefore more informative — they can fail in more ways, so surviving the test means more. A directional form is appropriate when theory or prior evidence justifies an expectation; a non-directional form is appropriate when the direction is genuinely unknown.

Figure 2 shows why the directional form is riskier. Each panel asks which of three patterns of turnover against manager quality — falling, rising or flat — would support the hypothesis. The non-directional hypothesis is supported by either slope and refuted only by no relationship; the directional one is supported only by the falling slope, so it can fail in more ways and surviving the test means more.

Directional and non-directional hypotheses Two panels. Non-directional: manager quality is related to staff turnover; the two variables are joined by a line without an arrowhead. Of three possible patterns of turnover against manager quality, it is supported both when turnover falls as manager quality rises and when turnover rises, and refuted only by no relationship: it states that a relationship exists, not which way it runs; use it when you genuinely do not know. Directional: the poorer the manager quality, the higher the staff turnover; an arrow is labelled poorer to higher. Only the falling pattern supports it; a rising pattern or no relationship refutes it. It is riskier and more informative, because it can fail in more ways, so surviving the test means more; use it when theory or prior evidence justifies an expectation. Non-directional “Manager quality is related to staff turnover.” Manager quality Staff turnover related Which patterns would support it? ✓ falling ✓ rising ✗ flat manager quality → turnover → States that a relationship exists, not which way it runs. Use when you genuinely do not know which way it runs. Directional “The poorer the manager quality, the higher the staff turnover.” Manager quality Staff turnover poorer → higher Which patterns would support it? ✓ falling ✗ rising ✗ flat manager quality → turnover → Riskier and more informative: it can fail in more ways, so surviving the test means more. Use when theory or prior evidence justifies an expectation.
Figure 2: Directional and non-directional hypotheses.

The second distinction underpins statistical testing, which always works with a pair of hypotheses. The null hypothesis (H0) states that there is no relationship between the variables (“manager quality has no effect on staff turnover”). The alternative hypothesis (H1) states that a relationship does exist, and is usually the research hypothesis itself. The logic of testing is indirect: we ask how likely our data would be if the null were true, and reject the null only when the data make it implausible. We never “prove” the alternative — we accumulate evidence against the null. This cautious logic protects researchers from seeing patterns that are not there.

Figure 3 shows the pair of hypotheses and the logic of testing them: we ask how likely our data would be if H₀ were true, reject H₀ only when the data make it implausible, and never prove H₁.

Null and alternative hypotheses, and the logic of testing them Two hypotheses about manager quality and staff turnover. The null hypothesis, H0: manager quality has no effect on staff turnover; the link between the variables is crossed out. The alternative hypothesis, H1: stores with poorer manager quality have higher turnover; usually your research hypothesis. Below, the logic of testing: from our data we ask how likely these data would be if H0 were true. If very unlikely, we reject H0, which is evidence for H1; if plausible, we do not reject H0. We never prove H1; we accumulate evidence against H0, a cautious logic that protects us from seeing patterns that are not there. Null hypothesis (H₀) Manager quality Turnover ✗ “Manager quality has no effect on staff turnover.” Alternative hypothesis (H₁) Manager quality Turnover “Stores with poorer manager quality have higher turnover.” — usually your research hypothesis Our data How likely would these data be if H₀ were true? very unlikely reject H₀: evidence for H₁ plausible do not reject H₀ NEVER PROOF We never prove H₁; we accumulate evidence against H₀. The cautious logic protects us from seeing patterns that are not there.
Figure 3: Null and alternative hypotheses, and the logic of testing them.

From question to hypothesis at Meridian

The board’s question — why has staff turnover at Meridian doubled in two years? — is too broad to test directly. Theory and observation allow the research team to derive specific, testable expectations from it: (H1) employees who report lower supervisor support are more likely to quit within a year (directional); (H2) turnover is higher in stores than in the e-commerce division (comparative and directional); (H3) perceived pay fairness is related to intention to quit (non-directional). Each hypothesis carves off a piece of the big question that a single study can actually answer, and each demands the same three things: named variables, a stated relationship, and data that could prove it wrong.

Figure 4 draws this as carving: the broad question at the top, and three hypotheses cut from it, each with its type and its independent and dependent variables.

Carving a broad question into testable hypotheses At the top, the board's question: why has staff turnover at Meridian doubled in two years? It is too broad to test directly. Dashed arrows carve it into three hypotheses. H1, directional: Employees who report lower supervisor support are more likely to quit within a year. Independent variable: supervisor support; dependent variable: quitting within a year. H2, comparative, directional: Turnover is higher in stores than in the e-commerce division. Independent variable: store or e-commerce; dependent variable: turnover. H3, non-directional: Perceived pay fairness is related to intention to quit. Independent variable: perceived pay fairness; dependent variable: intention to quit. Each slice has named variables, a stated relationship, and data that could prove it wrong. “Why has staff turnover at Meridian doubled in two years?” too broad to test directly: a topic wearing a question mark H1 Directional Employees who report lower supervisor support are more likely to quit within a year. IV: supervisor support DV: quitting within a year H2 Comparative, directional Turnover is higher in stores than in the e-commerce division. IV: store or e-commerce DV: turnover H3 Non-directional Perceived pay fairness is related to intention to quit. IV: perceived pay fairness DV: intention to quit Each slice: named variables, a stated relationship, and data that could prove it wrong
Figure 4: Carving a broad question into testable hypotheses.

Variables and units of analysis

Independent, dependent, and control variables

A variable is any characteristic that varies across the cases under study. Within a hypothesis, variables play defined roles. The independent variable (IV) is the presumed cause — the thing that does the influencing. The dependent variable (DV) is the presumed effect — the outcome to be explained. In “supervisor support → intention to quit”, support is the IV and intention to quit the DV. Importantly, the roles come from the hypothesis, not from the variables themselves: turnover is a dependent variable when we explain it, but an independent variable when we ask whether turnover damages customer satisfaction.

Figure 5 shows the two roles, and that they belong to the hypothesis rather than to the variable: turnover is the dependent variable when workload explains it, and the independent variable when we ask whether it damages customer satisfaction.

Independent and dependent variables, and how their roles depend on the hypothesis At the top: supervisor support, the independent variable or presumed cause, influences intention to quit, the dependent variable or presumed effect. Below: the roles come from the hypothesis, not from the variables. When we explain turnover, workload is the independent variable and turnover the dependent one; when we ask whether turnover damages customer satisfaction, turnover becomes the independent variable. INDEPENDENT VARIABLE Supervisor support DEPENDENT VARIABLE Intention to quit influences the presumed cause the presumed effect The roles come from the hypothesis, not from the variables: Workload DEPENDENT Turnover explaining turnover INDEPENDENT Turnover Customer satisfaction asking whether turnover damages satisfaction
Figure 5: Independent and dependent variables, and how their roles depend on the hypothesis.

A control variable is a third factor held constant, or measured and adjusted for, because it might otherwise distort the relationship of interest. Suppose stores with poor managers also pay less: is turnover driven by the manager, or by the pay? Without controlling for pay, the study risks a spurious conclusion — blaming managers for what wages are doing. Typical controls in a Meridian study would include pay level, the local unemployment rate, store size, and employee age and tenure. Identifying plausible controls is a test of how well the researcher understands the problem, since every control is a rival explanation taken seriously.

Figure 6 draws the pay example: the observed association between manager quality and turnover, and pay level pointing at both — stores with poor managers also pay less, and low pay drives people to leave — which is why, without the control, the conclusion would be spurious.

A control variable as a rival explanation We observe that stores with poor managers have high turnover: a dashed link with a question mark joins manager quality, the independent variable, to staff turnover, the dependent variable. A control variable, pay level, points at both: stores with poor managers also pay less, and low pay drives people to leave. Without controlling for pay, the conclusion is spurious: blaming managers for what wages are doing, and fixing the wrong thing. Typical controls in a Meridian study: pay level, local unemployment rate, store size, employee age and tenure, each a rival explanation taken seriously. INDEPENDENT Manager quality DEPENDENT Staff turnover ? observed: poor managers, high turnover CONTROL VARIABLE Pay level stores with poor managers also pay less low pay drives people to leave WITHOUT THE CONTROL a spurious conclusion: blaming managers for what wages are doing — and fixing the wrong thing TYPICAL CONTROLS pay level local unemployment rate store size employee age and tenure each a rival explanation taken seriously
Figure 6: A control variable as a rival explanation.

Units of analysis

The unit of analysis is the what or whom the study describes — the entities about which data are collected and conclusions drawn.

Unit Organisational examples
Individuals Employees, customers, voters
Groups Teams, departments, shifts
Organisations Stores, firms, branches
Social artefacts Exit interviews, complaint records, job adverts

The same topic can be studied at different units: “which employees quit?” is an individual-level question, while “which stores have high turnover?” is an organisational-level one. Researchers must beware the ecological fallacy: conclusions about one unit do not automatically transfer to another. Knowing that a store has high turnover does not tell you which individual employees within it are at risk of leaving.

Figure 7 draws the units as nested boxes — individuals within groups within organisations — with social artefacts as a separate kind of unit, the same topic posed at two units, and the ecological fallacy underneath.

Units of analysis and the ecological fallacy Nested boxes: organisations such as stores, firms and branches contain groups such as teams, departments and shifts, which contain individuals such as employees, customers and voters. Beside them, social artefacts, human products studied as units: exit interviews, complaint records, job adverts. The same topic at two units: which employees quit is a question about individuals, because people quit; which stores have high turnover is a question about organisations, because turnover is a property of a store. The ecological fallacy: a store with 40% turnover does not tell you which of its employees are at risk, since the churn may all be among weekend staff; conclusions about one unit do not transfer automatically to another. Organisations stores, firms, branches Groups teams, departments, shifts Groups teams, departments, shifts Individuals: employees, customers, voters Social artefacts human products studied as units: exit interviews, complaint records, job adverts “Which employees quit?” unit: individuals — people quit “Which stores have high turnover?” unit: organisations — turnover is a store's property THE ECOLOGICAL FALLACY A store with 40% turnover does not tell you which of its employees are at risk — perhaps the churn is all among weekend staff. Conclusions about one unit do not transfer automatically to another.
Figure 7: Units of analysis and the ecological fallacy.

Putting the pieces together

For hypothesis H1 — employees who report lower supervisor support are more likely to quit within a year — the components line up as follows:

Component Choice
Unit of analysis Individual employees
Independent variable Perceived supervisor support
Dependent variable Quitting within 12 months
Control variables Pay level, tenure, contract type, store location

Once this table is filled in, the study almost designs itself: we know whom to study, what to measure, and which rival explanations to rule out. This is the discipline the session exercise requires — for every hypothesis, name the unit, the IV, the DV, and at least one control.

Research designs

What is a research design?

A research design is the overall structure of a study: the plan that determines when, where, and on whom data are collected. It is the blueprint connecting the hypothesis to the evidence needed to test it — Creswell & Creswell describe it as the plan running from broad assumptions to detailed methods. Design is not the same as method: a survey (a method) can sit inside a cross-sectional, longitudinal, or comparative design. Five designs cover most organisational research, and the design question is always the same: what structure of evidence would allow this hypothesis to fail?

The five main designs

Cross-sectional designs collect data from many cases at a single point in time — a snapshot. They are the workhorse of organisational research: one employee survey across all 32 Meridian stores, fielded in a single week. They are fast, relatively cheap, and good for describing patterns and comparing groups at one moment, but weak on causal order: if dissatisfied employees quit, does dissatisfaction cause quitting, or do soon-to-quit employees become dissatisfied? A cross-sectional design can show that support and turnover intention are associated; it struggles to show which comes first.

In Figure 8, ten cases are drawn as timelines, all observed in one vertical band: a single moment. The boxes on the right give the Meridian version, the strengths and the weaknesses.

Cross-sectional design Pictogram: ten cases drawn as horizontal timelines, all observed at a single moment, a vertical band across them. At Meridian: one employee survey across all 32 stores, fielded in a single week. Strengths: fast and relatively cheap; good for describing patterns and comparing groups at one moment. Weaknesses: weak on causal order, since we cannot tell whether dissatisfaction causes quitting or soon-to-quit staff become dissatisfied; it shows association, not which comes first. cases time → one moment AT MERIDIAN One employee survey across all 32 stores, fielded in a single week. STRENGTHS Fast and relatively cheap; good for describing patterns and comparing groups at one moment. WEAKNESSES Weak on causal order: does dissatisfaction cause quitting, or do soon-to-quit staff become dissatisfied? It shows association, not which comes first.
Figure 8: Cross-sectional design.

Longitudinal designs collect data over time, allowing change to be observed directly and causes to be placed before effects. Babbie distinguishes three variants: trend studies (the same population, different samples — annual engagement surveys of Meridian staff), cohort studies (following a defined group, such as everyone hired in 2024), and panel studies (the same individuals measured repeatedly — surveying the same 200 employees every six months). Longitudinal work is slow and expensive, and panels suffer attrition; in a turnover study, the very people we most want to follow are the ones who leave.

Figure 9 shows the three variants across three waves: a new sample each wave (trend), one defined group followed over time (cohort), and the same individuals measured every time (panel).

Longitudinal designs: trend, cohort and panel Pictogram: three rows of data across three waves. Trend: a new sample each wave from the same population, drawn as differently coloured dots each time, as in annual engagement surveys. Cohort: one defined group, drawn as a box, followed across the waves, such as everyone hired in 2024. Panel: the same individuals, the same four coloured dots, measured at every wave, such as the same 200 staff every six months. Strengths: change can be observed directly, and causes can be placed before effects. Weaknesses: slow and expensive; panels suffer attrition, and in a turnover study the people we most want to follow are the ones who leave. wave 1 wave 2 wave 3 Trend new sample each wave annual engagement surveys Cohort one defined group everyone hired in 2024 Panel the same individuals the same 200 staff, every six months AT MERIDIAN Trend: annual engagement surveys. Cohort: everyone hired in 2024. Panel: the same 200 employees surveyed every six months. STRENGTHS Change can be observed directly, and causes can be placed before effects. WEAKNESSES Slow and expensive; panels suffer attrition — in a turnover study, the very people we most want to follow are the ones who leave.
Figure 9: Longitudinal designs: trend, cohort and panel.

Case study designs investigate a single case in depth — one store, one team, one organisation — using multiple sources of evidence such as interviews, observation, and documents. They suit questions of how and why: why does the Katowice store keep its staff when comparable stores cannot? Their strength is depth, context, and the ability to surface mechanisms and generate hypotheses; their limit is that findings cannot be statistically generalised. One exceptional store proves what is possible, not what is typical.

Figure 10 shows the case study as a magnifying glass over one store in its context, with interviews, observation and documents as its sources of evidence.

Case study design Pictogram: a magnifying glass over one store in its context, with three sources of evidence around it: interviews, observation and documents. At Meridian: why does the Katowice store keep its staff when comparable stores cannot? Strengths: depth and context; it surfaces mechanisms and generates hypotheses for wider studies to test. Weaknesses: no statistical generalisation, since one exceptional store shows what is possible, not what is typical. one store in context interviews observation documents AT MERIDIAN Why does the Katowice store keep its staff when comparable stores cannot? STRENGTHS Depth and context; surfaces mechanisms and generates hypotheses for wider studies to test. WEAKNESSES No statistical generalisation: one exceptional store shows what is possible, not what is typical.
Figure 10: Case study design.

Comparative designs systematically compare two or more cases, groups, or settings on the same dimensions. Meridian offers natural comparisons: stores versus e-commerce, high-turnover versus low-turnover stores, region against region. The logic is to let contrast do the explanatory work: if high- and low-turnover stores differ systematically in management style but not in pay, that is evidence worth having. Comparison approximates the logic of experiment where experiments are impossible, but compared cases differ in many ways at once, so rival explanations must be ruled out one by one.

In Figure 11, a high-turnover and a low-turnover store are compared on the same dimensions; where everything else is the same, the contrast points to what differs.

Comparative design Pictogram: a high-turnover and a low-turnover store compared on the same dimensions; pay, store size and region are the same, management style differs, so the contrast points to what differs. At Meridian: stores against the e-commerce division, high- against low-turnover stores, Poland's regions against one another. Strengths: contrast does the explanatory work, approximating the logic of experiment when experiments are impossible. Weaknesses: compared cases differ in many ways at once, so every other difference is a rival explanation to rule out, one by one. High-turnover store Low-turnover store = pay same = store size same = region same ≠ management style differs the contrast points to what differs AT MERIDIAN Stores against the e-commerce division; high- against low-turnover stores; Poland's regions against one another. STRENGTHS Lets contrast do the explanatory work; approximates the logic of experiment when experiments are impossible. WEAKNESSES Compared cases differ in many ways at once, so every other difference is a rival explanation to rule out, one by one.
Figure 11: Comparative design.

Experimental designs are the strongest for establishing cause: the researcher manipulates the independent variable and randomly assigns cases to treatment and control groups. Randomisation makes the groups equivalent on average, so a difference in outcomes can be attributed to the treatment — for instance, Meridian pilots a new onboarding programme in 16 randomly chosen stores while the other 16 continue as before. A quasi-experiment keeps the comparison but lacks random assignment (the programme goes to the stores whose managers volunteered). Quasi-experiments are common in organisations, where randomisation is often impractical, but self-selected groups may differ from the start, so causal claims require more caution.

Figure 12 contrasts random assignment, which makes the groups equivalent on average, with volunteering, where the groups may differ from the start.

Experimental and quasi-experimental designs Pictogram, top: in an experiment, 32 stores are randomly assigned, 16 to a new onboarding programme and 16 to continue as before, and turnover is compared; the groups are equivalent on average. Bottom: in a quasi-experiment, managers volunteer, so 16 volunteer stores get the programme and the other 16 do not; the groups may differ from the start. At Meridian: a new onboarding programme piloted in 16 randomly chosen stores; a quasi-experiment if it goes to the stores whose managers volunteered. Strengths: the strongest design for establishing cause, because random assignment makes the groups equivalent on average. Weaknesses: often impractical or unethical in organisations; self-selected groups may differ from the start, so causal claims need more caution. Experiment 32 stores random assignment 16: new onboarding 16: as before compare turnover groups equivalent on average Quasi-experiment 32 stores managers volunteer 16 volunteers: programme the other 16 ? groups may differ from the start AT MERIDIAN A new onboarding programme piloted in 16 randomly chosen stores while the other 16 continue as before; a quasi-experiment if it goes to the 16 stores whose managers volunteered. STRENGTHS The strongest design for establishing cause: random assignment makes the groups equivalent on average. WEAKNESSES Often impractical or unethical in organisations; in a quasi-experiment, self-selected groups may differ from the start, so causal claims need more caution.
Figure 12: Experimental and quasi-experimental designs.

Choosing among designs

Design Best for Main limit
Cross-sectional Describing patterns now; comparing groups Weak on causal order
Longitudinal Tracking change; ordering cause and effect Slow, costly, attrition
Case study Understanding how and why in depth No statistical generalisation
Comparative Learning from contrasts between units Many differences at once
Experimental Establishing cause Often impractical or unethical

No design is best in the abstract: the design must fit the question, the hypothesis, and the constraints of time, money, and access. Managerial problems usually signal the appropriate design in their own wording — now points to a cross-sectional snapshot, whether it works to an experiment or quasi-experiment, why this case to a case study, and changes over time to a longitudinal design. Managers rarely ask for a “research design”; translating their problem into one is precisely the researcher’s job.

Planning the research process

The stages of a research project

Babbie pictures a research project as a connected structure flowing from idea to application, with feedback loops throughout:

Stage What happens Covered in
Interest, idea, theory The problem is identified and hunches formed Sessions 1–2
Conceptualisation The meaning of the key concepts is specified Session 8
Choice of research method Survey, field research, existing data, and so on Sessions 11–12
Operationalisation How the variables will actually be measured Session 8
Population and sampling Whom the conclusions are about, and who will be observed Session 10
Observations The data are collected Sessions 11–12
Data processing and analysis Data are transformed, analysed, and conclusions drawn Later sessions
Application Results are reported and their implications assessed —

The stages loop as well as flow: the results of analysis feed back into the initial interests, ideas, and theories, and often this feedback begins another cycle of inquiry.

Schedule, resources, and practicalities

A plan is not just a list of stages — it attaches time, money, and people to each one. Building a schedule, even a rough timeline per stage, exposes whether the project fits its deadline: a board that wants answers in six weeks has just ruled out an eighteen-month panel study. The budget should cover the real costs of research: staff time, incentives for participants, software, travel, and data access. Practicalities need checking early — access to the stores, managers’ willingness to release staff for interviews, and the gatekeepers whose approval the project requires. Most research failures are planning failures: the design was fine, but the time and access it required were never there.

The research proposal

Before a study runs, its plan is usually written down for someone else’s review — a supervisor, a client, a board, or a funder. The research proposal lays out what you want to study, why it matters, and exactly how you will proceed. Writing it is not bureaucracy but the discipline of committing to decisions before data collection makes them irreversible; a proposal that cannot be written clearly is a study that has not been thought through clearly. For Meridian’s board, the proposal is the product: it turns “we should look into turnover” into a project someone can approve, fund, and hold the team to.

A short proposal typically contains the following sections:

Section Question it answers
Problem / objective What exactly will you study, and why is it worth studying?
Literature review What is already known, and what remains unresolved?
Research question & hypothesis What do you ask, and what do you expect to find?
Subjects for study Whom or what will you study, and how will they be selected?
Measurement What are the key variables, and how will you measure them?
Data-collection method How will the data actually be gathered?
Analysis How will the data answer the question?
Schedule & budget How long will it take, and what will it cost?

In this course, the session exercise asks for a one-page skeleton — problem, question, hypothesis, design, and data needed — which later sessions will build on.

A Meridian mini-proposal

A worked example shows how compactly the whole plan can be stated. Problem: turnover has doubled to 27% in two years, and the board needs to know why before budgeting counter-measures. Question: does the quality of first-line management explain differences in turnover across Meridian’s stores? Hypothesis: stores whose employees report lower supervisor support have higher annual turnover, controlling for pay and local labour-market conditions. Design: a cross-sectional survey of all store employees (CAWI), linked to store-level HR turnover records. Data needed: survey measures of supervisor support and quit intention; HR data on leavers, pay, and tenure; regional unemployment figures. Five short entries — and the board can see the whole study, judge it, and decide whether to fund it.

Figure 13 sets out the same five entries as blocks.

A Meridian mini-proposal in five entries Five entries of a short research proposal. Problem: Turnover has doubled to 27% in two years; the board needs to know why before budgeting counter-measures. Question: Does the quality of first-line management explain differences in turnover across Meridian's stores? Hypothesis: Stores whose employees report lower supervisor support have higher annual turnover, controlling for pay and local labour-market conditions. Design: A cross-sectional survey of all store employees (CAWI), linked to store-level HR turnover records. Data needed: Survey measures of supervisor support and quit intention; HR data on leavers, pay and tenure; regional unemployment figures. Problem Turnover has doubled to 27% in two years; the board needs to know why before budgeting counter-measures. Question Does the quality of first-line management explain differences in turnover across Meridian's stores? Hypothesis Stores whose employees report lower supervisor support have higher annual turnover, controlling for pay and local labour-market conditions. Design A cross-sectional survey of all store employees (CAWI), linked to store-level HR turnover records. Data needed Survey measures of supervisor support and quit intention; HR data on leavers, pay and tenure; regional unemployment figures.
Figure 13: A Meridian mini-proposal in five entries.

Conclusion

A hypothesis converts a research question into a testable expectation — specific, falsifiable, and built from named variables. Variables have roles: the independent variable does the explaining, the dependent variable is explained, and control variables guard against spurious conclusions, all relative to a stated unit of analysis. Research designs — cross-sectional, longitudinal, case study, comparative, and experimental — are structures of evidence, and the craft lies in matching the design to the problem rather than defending a favourite technique. Finally, planning turns method into management: stages, schedule, resources, and a proposal that commits the plan to paper before the data commit you. The session followed Meridian’s turnover crisis from a board-room worry to a fundable one-page plan; the exercise, and later your own projects, make the same journey.