Planning and conducting research — OCR A-Level Psychology
Test yourself on Planning and conducting research with OCR A-Level practice questions.
A week of Premium from your first lesson · Then the free plan continues · No card, no charge
Planning and conducting research explained
This part of the course is examined as a chain of decisions rather than as a glossary, because the paper hands you an unfamiliar aim and asks what you would actually do.
Read the full explanation
The order matters: an aim narrows into a testable prediction, the prediction fixes what must be measured and how, and only then do the sample and the design follow. Bocchiaro et al. (2012) shows why the planning itself is worth marks, since a separate comparison group was asked to predict how they would behave and their predictions bore little relation to what the participants in the procedure actually did, which is an argument for measuring behaviour rather than asking about it. Milgram (1963) shows the other half, standardising the prods, the setting and the learner's responses so that the only thing free to differ between participants was their own decision to continue.
Aims and hypotheses and how to formulate: research aim; research question; null hypotheses; alternative hypotheses; one-tailed (directional) hypotheses; two-tailed (non-directional) hypotheses.
A research aim states the study's purpose, whereas a research question phrases this as a specific query. An alternative hypothesis is a testable prediction. Operationalisation is crucial: predicting participants will recall more words when tested in their learning environment can be scored. The alternative hypothesis's form depends on existing evidence. If research suggests an outcome, a one-tailed (directional) hypothesis is defensible; an open area requires a two-tailed (non-directional) hypothesis. The null hypothesis is the position that there is no difference or relationship in the population, and any observed difference is due to chance.
Populations, samples and sampling techniques: target population and sample; random sampling; snowball sampling; opportunity sampling; self-selected sampling.
Recruitment bridges the sample tested and the target population a conclusion is about. Every technique is judged on who could never be included. Milgram (1963) advertised for men who chose to reply; this self-selected sampling invites volunteer bias, as respondents may be unusually compliant. Opportunity sampling relies on whoever is available, such as Piliavin et al. (1969) studying subway passengers, which is quick but skews the sample. Snowball sampling relies on participants recruiting others, useful for hard-to-reach groups. Random sampling gives every member of the target population an equal probability of inclusion. It requires a complete list of that group.
Experimental designs: repeated measures design; independent measures design; matched participants design.
Experimental design dictates how unwanted variation is managed. A repeated measures design puts the same people through every condition, removing individual differences. Testing participants in both noisy and silent conditions controls for baseline memory, but introduces order effects and reveals the procedure. An independent measures design uses different people in each condition, as Loftus and Palmer (1974) did when each group heard only one verb. This removes order effects, but participant variables now sit inside the comparison and must be spread out by allocating people at random. A matched participants design is the compromise: Bandura et al. (1961) rated children's existing aggression first and used those ratings to balance the groups.
Variables and how they are operationalised: independent variable (IV); dependent variable (DV); control of extraneous variables.
The independent variable (IV) is what a researcher manipulates or selects, the dependent variable (DV) is measured, and operationalising means defining each precisely enough for replication. In Loftus and Palmer (1974), the IV was the verb in the critical question, and the DV was speed in miles per hour; the mean estimate fell from 40.8 after 'smashed' to 31.8 after 'contacted'. Bandura et al. (1961) operationalised aggression by counting imitative acts in fixed intervals. Extraneous variables are factors other than the IV that could affect the DV. They become confounding if they vary systematically with the IV. Standardised instructions, counterbalancing, and random allocation help control them.
Designing observations: behavioural categories; coding frames; time sampling; event sampling.
A behavioural category turns a vague construct into something two observers can agree on, and a coding frame is the full list of those categories with the codes used to record them; categories have to be observable, mutually exclusive and wide enough to cover what is likely to happen. Time sampling records whatever is occurring at set moments, as when Bandura et al. (1961) scored each child through a one way mirror in five second intervals across a twenty minute session, while event sampling tallies every occurrence of a target act, as when Piliavin et al. (1969) logged each instance of help and the delay before it on New York subway trains. Bandura reported observer agreement of about 0.89, which is the evidence that a frame works rather than merely exists.
Designing self-reports: open questions; closed questions; rating scales: Likert rating scale, Semantic differential rating scale.
Open questions allow respondents to answer in their own words, yielding qualitative data. Closed questions restrict options, producing quantitative data for easy comparison. A Likert rating scale pairs a statement with an agreement scale (e.g., strongly agree to strongly disagree), summing scores across items. A semantic differential rating scale places opposing adjectives at either end of a line, asking respondents to mark a position between them. Chaney et al. (2004) used closed questions to compare the Funhaler with a standard spacer, finding more parents reported medicating their child the previous day. Wording is crucial; Loftus and Palmer (1974) demonstrated how changing one verb alters responses.
Your focus
- Apply principles of research design to novel scenarios by operationalising variables, selecting appropriate samples, and justifying specific experimental controls.
- Explain the purpose of a pilot study in planning research, detailing how it can be used to test procedures and implement necessary modifications.
- Evaluate proposed research procedures by identifying potential ethical or methodological weaknesses and suggesting workable safeguards or improvements.
Show all 21 objectives
- Distinguish between a research aim and a research question.
- Formulate operationalised one-tailed and two-tailed alternative hypotheses.
- Define the null hypothesis as the position that there is no population difference and any observed difference is due to chance.
- Differentiate between a target population and a sample.
- Identify and evaluate random, snowball, opportunity, and self-selected sampling techniques.
- Apply knowledge of sampling techniques to assess the generalisability of research findings.
- Identify repeated measures, independent measures, and matched participants designs from a research scenario.
- Explain the strengths and weaknesses of each experimental design, including order effects and participant variables.
- Describe how to control for participant variables using random allocation or matching, and order effects using counterbalancing.
- Identify and operationalise independent and dependent variables in a given research scenario.
- Distinguish between extraneous and confounding variables.
- Explain appropriate methods for controlling extraneous variables, such as standardisation and counterbalancing.
- Construct observable and mutually exclusive behavioural categories and coding frames to reliably record target behaviours.
- Compare time sampling and event sampling techniques, explaining which method is most appropriate for continuous versus rare behaviours.
- Evaluate the reliability of observational designs by explaining how inter-rater reliability provides evidence that a coding frame is usable.
- Design appropriate open and closed questions for a given research scenario.
- Differentiate between Likert and semantic differential rating scales.
- Evaluate the strengths and weaknesses of different self-report designs, including issues of bias.
Planning and conducting research exam tips
Marking Points
- Any design answer needs the variables operationalised, the controls named, and a reason each control is there rather than a list of controls in general.
- Say what a pilot study would test in this scenario, such as whether the instructions are understood or the task is too easy, and what you would change as a result.
- Ethical planning is credited when it names the issue this procedure raises and offers a workable safeguard, for example presumptive consent and a full debrief where deception is unavoidable.
- Where you identify a weakness, the mark follows the improvement plus the explanation of why it removes the problem you named.
- A credited alternative hypothesis names both conditions or co-variables and states the measure, so a marker can see what counts as evidence against it.
- A one-tailed (directional) hypothesis uses comparative words like more, fewer, or faster; a two-tailed (non-directional) hypothesis states there will be a difference but refuses to specify the direction.
- The null hypothesis must be written for the same operationalised variables, stating that there is no difference or relationship in the population and any observed difference is due to chance.
- Justifying the choice of a one-tailed or two-tailed hypothesis by citing previous findings, or the absence of them, is where justification marks are awarded.
- Name the target population the study wants to describe before criticising the sample, because a criticism only works against a stated population.
- Random sampling means a procedure giving everyone an equal probability, such as numbering a full list and drawing with a random number generator; it is not haphazard selection.
- Snowball sampling earns credit for reaching groups with no available list, but loses representativeness because participants recruit people similar to themselves.
- Arguments about generalisability are stronger when the limit is named, for example, an all-male sample or participants from a single city.
- Name the design and justify it from the scenario, for example, that an independent measures design is needed because the task cannot be given twice without answers being remembered.
- Counterbalancing must be described as a procedure, with half the participants taking the conditions in one order and half in the reverse, not simply named.
- Order effects should be specified as practice, fatigue, or boredom, and matched to the task actually described in the stem.
- Allocating participants at random is the control for participant variables in an independent measures design, and describing how it would be done earns the mark.
- Define the IV as the variable manipulated or selected by the researcher, and state its conditions (e.g., the verb in Loftus and Palmer with 'smashed', 'collided', 'bumped', 'hit', 'contacted').
- Define the DV with its unit, such as a speed estimate in miles per hour rather than just 'memory', or counting imitative aggressive acts in fixed intervals.
- Operationalise variables by providing a measurable indicator, unit, or scale, ensuring another researcher could replicate the procedure exactly.
- Distinguish an extraneous variable (which could affect the DV) from a confounding variable (which varies systematically with the IV conditions, offering a rival explanation).
- Explain controls by naming the technique and its purpose: standardised instructions ensure all participants hear the same information, counterbalancing mitigates order effects, and random allocation distributes participant variables.
- A behavioural category earns credit when it is observable and unambiguous, so pushes another child is creditworthy and is being naughty is not, and categories together should not overlap.
- Describe time sampling as recording what is happening at set moments, using Bandura et al. (1961), who scored each child in five second intervals across a twenty minute session behind a one way mirror.
- Describe event sampling as tallying every occurrence of the target behaviour, using Piliavin et al. (1969), who recorded whether help was given, how long it took and who gave it, with the victim carrying a cane helped on far more trials than the apparently drunk victim.
- Link the method to the behaviour: event sampling suits rare acts because it misses nothing, and time sampling suits behaviour too continuous or too frequent for one observer to keep up with.
- Inter-rater reliability is the evidence a coding frame is usable, and Bandura reported agreement of about 0.89 between observers scoring the same children.
- Contrast open questions, which provide depth but are harder to analyse, with closed questions, which yield easily tallied quantitative data but restrict respondent choices.
- Define a Likert rating scale as measuring agreement with a statement on a scale (e.g., strongly agree to strongly disagree), producing ordinal data.
- Define a semantic differential rating scale as measuring the connotation of a concept by placing opposing adjectives at the ends of a scale and asking for a position between them.
- Explain how to reduce social desirability bias using anonymous completion or filler items, and distinguish this from reversing item direction, which detects acquiescence or response bias.
- Use evidence: Chaney et al. (2004) collected parental self-reports on the Funhaler, showing self-report can detect real behavioural changes despite relying on truthful responses.
Examiner Tips
- 💡Design questions are marked in levels, so plan one paragraph for method and design, one for sample and procedure, and one for controls and ethics before you start writing.
- 💡The word justify signals that the marks sit in the reasons rather than the decision, and most lost marks are decisions stated with no reason attached.
- 💡Keep one study of your own design ready to adapt to a new topic; you will be quicker than candidates inventing a procedure from nothing under time pressure.
- 💡When writing an alternative hypothesis from an unseen scenario, ensure you include both conditions and the operationalised dependent variable, adding a direction only if a one-tailed hypothesis is required.
- 💡If the stem says previous research found an effect, write a one-tailed (directional) hypothesis; if findings are mixed or absent, write a two-tailed (non-directional) hypothesis.
- 💡When asked to identify the sampling technique used in an unseen study and justify that from the stem, quote the specific recruiting detail provided in the text.
- 💡If asked to suggest another way of obtaining the sample and one advantage of it, give a method plus a benefit specific to this target population.
- 💡Questions regularly ask you to identify the design and explain one weakness of using it here, so read the procedure to see if the same people appear in more than one condition.
- 💡The design decides the statistical test, so link the two: related data comes from repeated measures and matched participants designs, unrelated data from independent measures designs.
- 💡In Section B or C of the research methods paper, you may be asked to identify or operationalise variables from a scenario; ensure your wording is specific to the context provided.
- 💡When asked to explain how a variable was controlled, use the pattern: name the variable, name the control method, and explain what would have happened without it.
- 💡The design question frequently asks you to suggest two behavioural categories, and each must be written as an observable act rather than as a label for a state of mind.
- 💡Where you have to justify a sampling method, write about the frequency of the behaviour you chose, because frequency is what decides between the two.
- 💡Strengths and weaknesses of observation are usually asked against a named core study, so keep Bandura and Piliavin ready as the worked examples.
- 💡When asked to design open and closed questions, ensure both are directly answerable by the specific participants described in the scenario.
- 💡If evaluating self-reports, explicitly name a bias (e.g., social desirability), explain why it occurs in the specific context, and describe its impact on validity.
Common Mistakes
- Describing an aim where a hypothesis was asked for, so the answer contains no prediction and no measurable variables.
- Listing controls generically, such as promising to control all extraneous variables, instead of naming the one variable that actually threatens this study.
- Offering informed consent as the fix in a study where the deception is the method, without considering presumptive consent, debriefing or the right to withdraw data afterwards.
- Writing the research aim in the form of an alternative hypothesis, or writing a hypothesis that is really a description of the procedure. Correction: Ensure aims state intentions (e.g., 'To investigate...') and hypotheses state testable predictions.
- Treating the null hypothesis as a real-world forecast that nothing will happen. Correction: The null hypothesis is a statistical baseline stating there is no difference or relationship in the population, and any observed difference is due to chance.
- Leaving the measure out, so a claim that boys will be more aggressive has no stated way of being scored. Correction: Always operationalise the dependent variable in the hypothesis.
- Using the word random for opportunity sampling, so stopping whoever walks past in a corridor gets described as random selection. Correction: Use 'opportunity sampling' for selecting available people; reserve 'random' for mathematical probability methods.
- Saying a sample is too small with no reference to how variable the measure is or how wide the target population is. Correction: Link sample size criticisms to the specific target population and the risk of anomalies skewing the data.
- Confusing self-selected sampling with opportunity sampling. Correction: The difference is who made the first approach; self-selected means the participant volunteered, whereas opportunity means the researcher approached them.
- Confusing counterbalancing with randomising the order, or claiming counterbalancing removes order effects when it actually spreads them evenly across conditions.
- Saying a repeated measures design needs fewer participants and stopping there, without explaining that each person supplies data in every condition.
- Describing a matched participants design as matching people to conditions, or matching on a variable with no bearing on the dependent variable.
- Writing the DV as a broad concept like 'memory' or 'aggression' instead of the specific measurement taken. Correction: Always include the unit or method of measurement, such as 'speed estimates in mph'.
- Stating that quasi-experiments do not have an IV. Correction: Quasi-experiments do have an IV, but it is naturally occurring or pre-existing (a quasi-IV) rather than directly manipulated by the researcher.
- Listing controls without linking them to a specific extraneous variable. Correction: Always explain what the control removes or prevents, such as using random allocation to control for participant variables.
- Offering categories that are really inferences about intention, such as feeling anxious or trying to be helpful, which two observers will score differently.
- Describing time sampling as watching for a set time and event sampling as counting events, which fails to separate them, since both involve watching for a set time.
- Claiming a covert observation raises no ethical problem because nobody was upset, when consent, the right to withdraw and privacy are all compromised by it.
- Writing so many categories that no observer could hold them in mind, which destroys the reliability the frame was meant to create.
- Calling any numbered scale a Likert scale. Correction: A Likert scale specifically measures the level of agreement with a statement, not just any numerical rating.
- Conflating social desirability bias with response bias. Correction: Reversing the direction of statements detects acquiescence (response bias), whereas anonymity reduces social desirability bias.
- Writing leading or double-barrelled questions in design answers. Correction: Ensure questions ask only one thing and do not suggest a preferred answer.
- Treating rating scale totals as interval data. Correction: Rating scales produce ordinal data because the intervals between points are not guaranteed to be equal.