Non-Experimental Methods

Not every question in psychology can — or should — be answered with an experiment. Some behaviour cannot ethically or practically be manipulated; some occurs naturally and would be destroyed by the artificiality of a lab; and some questions are about meaning and experience rather than a single measured outcome. For these, psychologists turn to non-experimental methods: observations, self-report (questionnaires and interviews), correlations, case studies, and content analysis. Each describes, measures or explores behaviour without manipulating an independent variable, and so — with the sole exception nobody should forget — none of them can establish cause and effect on its own. Edexcel Paper 3 tests these methods heavily, and almost always in context: you will be asked to identify a method, design an element of it (a set of behavioural categories, a questionnaire item), or evaluate its reliability, validity or ethics for a specific study. This lesson equips you for all of those.

Key Definition: Non-experimental methods are research approaches that do not involve the direct manipulation of an independent variable. They describe, measure and explore relationships between variables, but do not — in themselves — establish causation.

By the end of this lesson you will be able to:

• Classify an observation along its four dimensions and design operationalised behavioural categories with an event- or time-sampling procedure.
• Distinguish questionnaires from interviews, evaluate response biases, and design clear, unbiased self-report items.
• Interpret a correlation coefficient and explain rigorously why correlation cannot establish causation.
• Evaluate the strengths and limitations of case studies and content analysis, and select the method that fits a given research question.

---

Spec Mapping

Edexcel 9PS0 — Paper 3: Psychological Skills (Research Methods). This lesson develops the non-experimental methods strand assessed in Section A of Paper 3. Our teaching sequence groups the methods by what the researcher is doing — watching (observation), asking (self-report), relating (correlation), and analysing existing material (case studies, content analysis) — rather than following the specification's own order.

Our lesson covers	Edexcel 9PS0 research-methods area
Naturalistic/controlled, covert/overt, participant/non-participant observation; behavioural categories; event/time sampling	Observational methods and techniques
Questionnaires (open/closed) and interviews (structured/unstructured/semi-structured); item design	Self-report methods
Positive, negative and zero correlations; the correlation coefficient; the causation caveat	Correlational analysis
Case studies; content analysis and coding	Case studies and analysis of qualitative material

Assessment Objectives. These items are AO2-led (identify the method in a scenario; design behavioural categories or a self-report item for it; sketch a scattergram) and AO3-heavy (evaluate ecological validity, response bias, the causation limit, or generalisability in context), on an AO1 base of accurate definitions. Reliability, validity and ethics — which apply across all these methods — have their own dedicated lessons in this course; here they are touched on only where a method-specific point requires it.

Connects to…

• Social/developmental psychology (aggression). Bandura's (1961) Bobo-doll study is a structured observation using operationalised behavioural categories scored for inter-observer reliability — the standard worked example for the observational method.
• Cognition (memory). The case of HM (Henry Molaison) (Scoville & Milner, 1957) is the classic case study, providing irreplaceable evidence for the role of the hippocampus in forming new long-term memories.
• Clinical / individual differences. Correlational designs dominate research relating, for example, life-stress scores to illness or symptom severity, where the variables cannot ethically be manipulated — the natural home of "correlation is not causation".

---

Observations

Observational research means watching and recording behaviour as it occurs. The categories below are not mutually exclusive — a single study could be, for example, a covert, non-participant, naturalistic, structured observation.

graph TD
    A[Observational techniques] --> B[Setting]
    A --> C[Awareness]
    A --> D[Researcher role]
    A --> E[Structure]
    B --> B1[Naturalistic]
    B --> B2[Controlled]
    C --> C1[Covert]
    C --> C2[Overt]
    D --> D1[Participant]
    D --> D2[Non-participant]
    E --> E1[Structured: behavioural categories]
    E --> E2[Unstructured]

Type	Description	Strength	Limitation
Naturalistic	Behaviour observed in its natural setting, no intervention	High ecological validity; genuine behaviour	Low control; extraneous variables; hard to replicate
Controlled	Behaviour observed in a structured, set-up environment	Greater control; replicable	Lower ecological validity; behaviour may be artificial
Covert	Participants unaware they are observed	Reduces demand characteristics	No informed consent; privacy concerns
Overt	Participants know they are observed	Ethically preferable; consent possible	Behaviour may change (the Hawthorne effect)
Participant	Researcher joins the group	Rich, insider data	Loss of objectivity; "going native"; ethical issues
Non-participant	Researcher observes from outside	More objective; less influence on the group	May miss subtle, context-dependent behaviour
Structured	Uses pre-set behavioural categories and a sampling method	Systematic, quantitative; better inter-observer reliability	May miss unanticipated behaviours
Unstructured	Records all relevant behaviour, no framework	Captures richness and complexity	Hard to analyse; observer bias; lower reliability

Behavioural categories and sampling

To make observation systematic, researchers operationalise behaviour into behavioural categories — a checklist of clearly defined, observable, mutually exclusive actions (e.g. for "aggression": hits, kicks, pushes, shouts). Good categories are objective (require no inference — "hits another child" rather than "is being mean"), mutually exclusive (a single act falls into only one category) and exhaustive enough to capture the behaviour of interest. Researchers then decide how to sample behaviour over time:

• Event sampling — counting every occurrence of a target behaviour throughout the observation. Best when the behaviour is relatively infrequent, but it can overwhelm the observer if many things happen at once.
• Time sampling — recording behaviour only at fixed intervals (e.g. what is happening every 30 seconds). This keeps the workload manageable for frequent behaviour, but may miss behaviours occurring between the sampled moments, reducing validity.

For example, an observer studying playground aggression might use event sampling during a 20-minute break, tallying each instance of hitting, kicking, pushing or shouting against the agreed categories — a concrete procedure that also makes inter-observer reliability checkable.

The categories interact in practice. A study might be covert and participant and naturalistic — as in a researcher who secretly joins a group to observe it in its everyday setting. Each choice carries its own trade-off: covert observation gains natural behaviour but loses informed consent; participant observation gains insider depth but risks the researcher "going native" and losing objectivity; naturalistic observation gains ecological validity but loses control. Strong evaluation discusses these combined implications rather than treating each label in isolation.

Key Definition: Inter-observer reliability is the extent to which two or more observers, recording the same behaviour independently, produce the same records. It is assessed by correlating their tallies; a coefficient of $r \geq +0.80$r≥+0.80 (or an agreement of 80% or more) is generally taken as acceptable. If agreement is low, the behavioural categories are redefined and the observers retrained.

A recurring threat unique to observation is observer bias — the tendency for what an observer expects or interprets to colour what they record, especially with vague, inferential categories. Two observers watching the same rough-and-tumble play might code it very differently ("aggression" vs "play-fighting") if the categories leave room for judgement. This is precisely why operationalised, low-inference categories and inter-observer reliability checks matter: they convert a subjective act of watching into a systematic, checkable measurement. A related concern is reactivity (the observer effect): if participants know they are watched (an overt observation), they may alter their behaviour — the Hawthorne effect — which is one motivation for covert observation, at the cost of the consent and privacy issues discussed in the ethics lesson.

Exam Tip: When evaluating any observation, discuss inter-observer reliability and how to improve it — operationalise categories clearly, train observers together, and use video so records can be re-checked. (Reliability is developed fully in this course's dedicated lesson.)

Worked example — designing an observation from scratch

It is worth walking through the whole design decision once, because Edexcel items often ask you to build one element and it helps to see how the pieces fit. Suppose a researcher wants to study helping behaviour among nursery-age children during free play.

1. Choose the design along the four dimensions. A naturalistic setting (their own nursery, not a lab) preserves ecological validity; non-participant observation keeps the researcher objective and avoids influencing the young children; covert recording — from behind a one-way screen or via an unobtrusive camera, with staff and parental consent obtained in advance — reduces reactivity so the children behave normally. The observation is structured, because the researcher wants countable data on a defined behaviour.
2. Operationalise the behaviour into categories. "Helping" is too vague to count, so it is broken into observable, mutually exclusive units: shares a toy, comforts a crying child, helps another child complete a task, fetches an object for someone. Each is defined so that no inference about intention is required — the observer records what is seen, not what is assumed.
3. Choose a sampling procedure. Because helping is relatively infrequent, event sampling fits: every instance of a listed behaviour is tallied across a 30-minute free-play session. (If the behaviour were near-continuous, time sampling — checking every 20 seconds — would keep the workload manageable, at the cost of missing acts between the sampled points.)
4. Build in a reliability check. Two observers record the same session independently, and their tallies are correlated. An agreement of $r \geq +0.80$r≥+0.80 is treated as acceptable; if it is lower, the categories are ambiguous and must be tightened and the observers retrained before real data collection begins. Video allows the session to be re-scored against identical footage.

Laying the design out this way exposes the trade-offs a scenario question expects you to discuss: covert recording buys natural behaviour but raises consent and privacy issues (handled here by obtaining staff and parental permission); tight categories buy reliability but may miss forms of helping no one anticipated; event sampling captures every instance but can overwhelm one observer if several children help at once. This is precisely the reasoning that separates a described procedure from an evaluated one.

---

Self-Report Methods

Self-report means asking participants to report their own thoughts, feelings, attitudes or behaviour, via questionnaires or interviews.

Questionnaires

A questionnaire is a pre-set list of written items. Items may be:

• Closed questions — fixed response options (yes/no, Likert scales, ranking). They produce quantitative data that is easy to analyse and compare across large samples, but restrict respondents to the options provided.
• Open questions — answered in the respondent's own words. They produce qualitative data — rich and detailed, able to capture unexpected responses — but harder to analyse and compare.

The choice is therefore a trade-off between breadth and standardisation (closed) and depth and authenticity (open), and many questionnaires deliberately mix the two: closed items for the variables of central interest, plus a few open items to capture nuance.

Strengths	Limitations
Reach large samples quickly and cheaply	Social desirability bias — answers people think are acceptable rather than true
Easy to replicate (standardised format)	Acquiescence bias — a tendency to agree regardless of content
Closed-question quantitative data allow statistical analysis	Respondents may rush or misinterpret items with no one to clarify
Anonymity may improve honesty	Closed questions restrict depth

Interviews

Type	Description	Strength	Limitation
Structured	Fixed questions in a set order	Standardised; replicable; easy to compare	Inflexible; may miss important points
Unstructured	Conversation develops freely	Rich, detailed, flexible	Hard to replicate/analyse; interviewer bias
Semi-structured	Set questions plus follow-ups	Balances structure and flexibility	Some loss of comparability; skill-dependent

Key Definition: Social desirability bias occurs when participants give answers that present them in a favourable or socially acceptable light rather than truthful ones — a central threat to the validity of self-report data.

Designing good self-report items

Edexcel expects you to be able to design self-report materials, so it is worth knowing what separates a good item from a poor one. Good questionnaire and interview questions are:

• Clear and unambiguous — no jargon, no double negatives, and no double-barrelled questions (which ask two things at once, e.g. "Do you find your job stressful and poorly paid?" — a participant cannot answer if only one applies).
• Unbiased / non-leading — they must not steer the respondent towards an answer (e.g. "How annoying is the new policy?" presupposes annoyance).
• Appropriately formatted — closed items (Likert scales, rating scales, fixed choices) for easily analysed quantitative data; open items where depth and the respondent's own words matter.
• Sensitively sequenced — sensitive questions placed later, with filler questions included to disguise the aim and reduce demand characteristics.

A Likert scale (strongly agree → strongly disagree) is the most common closed format; note that the data it yields are ordinal, because the psychological distance between "agree" and "strongly agree" is not guaranteed to equal that between "neutral" and "agree".

Exam Tip: Good self-report items are clear (no jargon, no double-barrelled questions), unbiased (no leading questions), and use filler items to disguise the aim. When evaluating, weigh social desirability bias, the data type (quantitative vs qualitative), depth, and whether a researcher is present to influence responses.

---

Correlations

A correlation measures the strength and direction of the relationship between two co-variables. It is a technique of analysis, not strictly a method — the data can be gathered by questionnaire, observation or archival record.

Key Definition: A correlation is a measure of the relationship between two co-variables. A positive correlation means both rise together; a negative correlation means one rises as the other falls; a zero correlation means there is no systematic relationship.

Type	Description	Scattergram pattern
Positive	Both co-variables increase together (e.g. study hours and exam marks)	Points slope up, left to right
Negative	One rises as the other falls (e.g. stress and immune function)	Points slope down, left to right
Zero	No systematic relationship	Points scattered randomly

The correlation coefficient

Strength is expressed as a correlation coefficient, a number on the scale

$-1.0 \leq r \leq +1.0$−1.0≤r≤+1.0

where $+1.0$+1.0 is a perfect positive relationship, $0$0 is no relationship, and $-1.0$−1.0 is a perfect negative relationship. The closer to $\pm 1$±1, the stronger the association: roughly, $|r| > 0.7$∣r∣>0.7 is strong, $0.3$0.3–$0.7$0.7 moderate, and $< 0.3$<0.3 weak. Crucially, a negative coefficient is not a weak one — $-0.85$−0.85 is a strong (negative) relationship.