Module 3: Research Design

← Previous: Module 2 · Subject index · Next: Module 4 →

Learning outcomes

After this module, you should be able to:

• state the role and importance of a research design;
• formulate and classify research hypotheses (directional/non-directional);
• choose between experimental and non-experimental, exploratory, descriptive, and causal designs;
• explain validity, reliability, and triangulation;
• design control groups and apply randomisation.

Prerequisites

Complete Module 2 first. Review its quick-revision section if any term below feels unfamiliar.

Study blocks

Study one block at a time. Work through its example and checkpoint before continuing.

Block	Topic	Suggested time
1	Start here: the simple idea	10-15 minutes
2	Research hypothesis: types and examples	10-15 minutes
3	Role and importance of research design	10-15 minutes
4	Experimental vs non-experimental designs	10-15 minutes
5	Exploratory, descriptive, and causal designs	10-15 minutes
6	Validity, reliability, and triangulation	10-15 minutes
7	Control groups and randomisation	10-15 minutes
8	Research-design case studies	10-15 minutes
9	Selecting a design for a proposal	10-15 minutes

Start here: the simple idea

A research design is the master plan that links the research question to the answer. Think of it as the blueprint of a house: it decides where every room goes so that the final structure is sound. A poor design cannot be rescued later by better statistics.

Analogy

Design = a cooking recipe: ingredients (data), steps (methods), temperature/time (design choices), and the tasting (analysis) that tells you if it worked. If the recipe is wrong, the dish fails no matter how good the stove is.

Research hypothesis: types and examples

A hypothesis is a testable statement about the relationship between variables.

Type	Form	Example
Directional (H₁)	predicts direction	Exercise increases daily step count
Non-directional	predicts a difference, no direction	Exercise affects stress scores
Null (H₀)	predicts no effect/difference	Exercise has no effect on stress

For our running study

• H₀: The 8-week exercise programme does not change mean stress scores.
• H₁ (directional): Mean stress score decreases after the exercise programme.

Role and importance of research design

The design:

• links operations to objectives;
• decides what data to collect and how;
• protects against bias and confounding;
• makes the study replicable;
• governs the strength of inference (correlation vs causation).

A good design is systematic, controlled, and matched to the question.

Experimental vs non-experimental designs

Feature	Experimental	Non-experimental
Manipulation of IV	Yes (researcher assigns treatment)	No (IV observed as it occurs)
Control / randomisation	Usually (control group, random assignment)	None/weak
Causal claim strength	Strong	Weak (association only)
Example	RCT of exercise vs control group	Survey of exercise habits and stress

True experimental design (our study)

Step 1  Recruit N students
Step 2  Randomise → Group A (exercise)  /  Group B (no exercise, wait-list)
Step 3  Measure stress (PSS-10) at baseline
Step 4  Run 8-week supervised programme for Group A only
Step 5  Measure stress again; compare post-score change A vs B

Randomisation here balances unknown confounders (fitness, motivation) across groups.

Exploratory, descriptive, and causal designs

• Exploratory: discover territory; qualitative interviews; hypothesis-generating.
• Descriptive: what/why/when/where; surveys, case studies; hypothesis-descriptive.
• Causal: establish cause-effect; experiments with manipulation + control.

Your study can be descriptive (report stress levels) and causal (test the exercise effect) at the same time: descriptive of the phenomenon, causal in the comparison.

Validity, reliability, and triangulation

Validity (are you measuring the right thing?)

• Construct validity: the measure captures the theoretical construct (stress).
• Internal validity: the IV truly causes the DV (ruling out confounds).
• External validity: results generalise beyond the study.
• Statistical conclusion validity: the stats correctly support the conclusion.

Reliability (are your measurements stable?)

• Test-retest: same score on re-measurement.
• Internal consistency: Cronbach's α (≥ 0.7 acceptable for scales).
• Inter-rater: different raters agree.

Triangulation

Using multiple methods/data sources to study the same thing. For our study: PSS-10 stress scores + weekly mood diaries + weekly step counts → stronger validity than any single measure.

Control groups and randomisation

• Control group: no treatment (or standard treatment); baseline for comparison.
• Randomisation: assign units to conditions by chance → equalises groups on expected confounders in expectation.
• Blocking / stratification: randomise within a stratum (e.g., by sport vs non-sport students) to balance a known factor.

Randomisation methods:

• Simple random (lottery).
• Block random (small blocks to keep groups balanced over time).
• Stratified random (by a known prognostic factor).

Research-design case studies

1. RCT of a drug — experimental, causal, high internal validity.
2. Cohort study of air pollution and asthma — observational, can show association; causal via strong design + analysis, but weaker.
3. Cross-sectional survey of smartphone use and sleep — descriptive/observational.
4. Case-control study of rare disease — efficient for rare outcomes.

Selecting a design for a proposal

Choose by matching the question to the method:

1. Want to test causality? → experiment.
2. Exploring a new construct? → qualitative/exploratory.
3. Describing prevalence? → cross-sectional survey.
4. Studying a rare outcome? → case-control.
5. Need to control confounding? → randomisation or matching.

PDF-aligned additions

Hypothesis forms

• Null hypothesis (H0): states no effect, difference, or relationship.
• Alternative hypothesis (H1): states the effect or relationship expected by the researcher.
• Directional hypothesis: predicts the direction of an effect.
• Non-directional hypothesis: predicts a difference without specifying its direction.

A hypothesis should identify measurable variables and be testable with the selected design.

Validity, reliability, and triangulation

Validity asks whether a study or instrument measures what it claims to measure. Reliability asks whether measurement is consistent. A measure may be reliable but invalid. Triangulation strengthens conclusions by combining methods, data sources, investigators, or theoretical perspectives.

Randomisation reduces systematic assignment bias; a control group provides a comparison; neither automatically removes every confounder.

Common mistakes

• Confusing variables (properties) with attributes (values).
• Choosing a design after seeing the data (researcher degrees of freedom bias).
• Ignoring threats to internal validity (confounding, selection, history).
• Believing randomisation removes all bias (it balances expected confounders, but a single experiment can still be unbalanced by chance).

Memory rules

• H₀ = no effect; H₁ = effect (directional = predicts direction; non-directional = predicts difference).
• Experimental = manipulate + control; non-experimental = observe as-is.
• Validity (right thing) vs reliability (stable); triangulation strengthens validity.
• Randomisation balances groups in expectation; blocking balances known factors.
• Exploratory (discover) → Descriptive (what) → Causal (why).

Check your understanding

1. Write H₀ and a directional H₁ for "exercise reduces stress."
2. Name one internal-validity threat and one way to address it.
3. Is a cross-sectional survey experimental or non-experimental?
4. Why block on a variable when randomising?

Answers

Reveal answers after attempting the questions

1. H₀: no change in mean stress; H₁: mean stress decreases after exercise.
2. e.g., confounding (students with higher fitness may exercise more) → randomise to balance.
3. Non-experimental.
4. To ensure equal representation of that factor across conditions (balance known prognostic factor).

Quick revision box

• Hypothesis: H₀ (null, no effect) vs H₁ (directional or non-directional).
• Design = blueprint; match it to the question: experiment (causal) vs observational (association).
• Experimental = manipulate + control group + randomisation; non-experimental = as-observed.
• Validity (construct/internal/external/statistical), reliability (test-retest, α, inter-rater); triangulation.
• Randomisation (simple/block/stratified) balances confounders.
• Exploratory → Descriptive → Causal (design types), not a strict sequence.

Practice ladder

1. Easy - Recall: Define the module's central idea in one or two sentences.
2. Easy - Recognize: Identify the correct method for a small example and explain why it fits.
3. Medium - Apply: Work through one representative problem without copying the example.
4. Medium - Compare: Contrast two methods or concepts from the module.
5. Hard - Integrate: Solve a university-style scenario and justify every major step.

Reveal self-evaluation guide

A complete response uses correct terminology, shows intermediate steps, connects the result to the scenario, and states one assumption or limitation.

---

← Previous: Module 2 · Subject index · Next: Module 4 →