Research Methods

A psychologist predicts that participants who revise in silence will recall more words than participants who revise with music playing. Which type of hypothesis is this, and why?

• AA non-directional (two-tailed) hypothesis, because two conditions are compared
• BA null hypothesis, because it predicts no effect of music
• CA directional (one-tailed) hypothesis, because the direction of the difference is stated✓
• DA correlational hypothesis, because two variables are measured

Answer: The prediction states the direction of the expected difference (the silence group will recall MORE), which makes it directional (one-tailed). A non-directional hypothesis would simply predict a difference without specifying which group does better.

A researcher conducts an unrelated-design experiment comparing reaction times (measured in milliseconds) between a caffeine group and a placebo group. The data are normally distributed. Which inferential statistical test is most appropriate?

• AWilcoxon signed-ranks test
• BSpearman's rho
• CRelated t-test
• DUnrelated t-test✓

Answer: The design is unrelated (independent groups), the data are interval (milliseconds) and normally distributed (parametric), so the unrelated t-test is correct. The related t-test is for repeated measures, and Wilcoxon is the non-parametric equivalent for related ordinal data.

Which statistical test should be used for a test of difference using a repeated measures design with data at an ordinal level of measurement?

• AWilcoxon signed-ranks test✓
• BMann-Whitney U test
• CChi-squared test
• DSpearman's rho

Answer: Wilcoxon is used for tests of difference with related (repeated measures/matched pairs) designs at ordinal level. Mann-Whitney is the unrelated equivalent; chi-squared is for nominal data and association; Spearman's is for correlation.

A researcher tests for a difference in ordinal-level scores between two separate, independent groups of participants. Which inferential test is appropriate?

• AWilcoxon signed-ranks test
• BMann-Whitney U test✓
• CRelated t-test
• DSign test

Answer: Mann-Whitney is the test of difference for an unrelated (independent groups) design with ordinal data. Wilcoxon and the sign test are for related designs; the related t-test requires a related design and interval, normally distributed data.

A study records whether participants are 'left-handed' or 'right-handed' and whether they prefer 'tea' or 'coffee', then tests for an association between the two. What test and level of measurement apply?

• ASpearman's rho; ordinal data
• BUnrelated t-test; interval data
• CSign test; nominal data
• DChi-squared test; nominal data✓

Answer: Both variables are categories (nominal) and the study tests for an association between them with independent data, so chi-squared is correct. The sign test is for differences in a related design, not associations between two nominal variables.

In hypothesis testing, what does a significance level of p ≤ 0.05 represent?

• AThere is a 5% probability that the alternative hypothesis is true and a 95% probability that the researcher should reject it outright
• BThe researcher accepts up to a 5% probability that the result occurred due to chance if the null hypothesis is true✓
• CThe result the researcher obtained will turn out to be correct on 95 occasions out of every 100 that the study is repeated
• DThere is a 95% probability that the null hypothesis is false, so it can be rejected with complete confidence by the researcher

Answer: p ≤ 0.05 means the researcher accepts up to a 5% (1 in 20) probability that the observed results occurred by chance (sampling error) when the null hypothesis is actually true. It is not the probability that a hypothesis is true or false.

A researcher rejects the null hypothesis and accepts the alternative hypothesis, but in reality the null hypothesis was true. What type of error has been made?

• AA Type II error (a false negative)
• BA counterbalancing error
• CA Type I error (a false positive)✓
• DA sampling error

Answer: Rejecting a true null hypothesis (claiming an effect exists when it does not) is a Type I error, or false positive. A Type II error is the opposite: retaining a false null hypothesis (missing a real effect).

If a researcher uses a more lenient significance level, such as p ≤ 0.10 instead of p ≤ 0.05, what is the most likely consequence?

• AThe risk of a Type I error increases and the risk of a Type II error decreases✓
• BThe risk of a Type I error decreases and the risk of a Type II error increases
• CBoth Type I and Type II error risks decrease
• DNeither error risk is affected by the significance level

Answer: A more lenient level (e.g. 0.10) makes it easier to reject the null, increasing the chance of a false positive (Type I error) but reducing the chance of missing a genuine effect (Type II error). The two error types trade off against each other.

A researcher measures the relationship between hours of revision and exam marks for 20 students, with both variables recorded at the interval level but not normally distributed. Which test is most suitable?

• ASpearman's rho✓
• BPearson's r
• CWilcoxon signed-ranks test
• DRelated t-test

Answer: This is a test of correlation. Because the data are not normally distributed, the non-parametric Spearman's rho (which ranks the data) is appropriate rather than the parametric Pearson's r. Wilcoxon and the related t-test are tests of difference, not correlation.

An experimenter selects participants by listing everyone in the target population and choosing every 5th person on the list. What sampling method is this?

• AOpportunity sampling
• BStratified sampling
• CSystematic sampling✓
• DVolunteer sampling

Answer: Selecting every nth member from an ordered list is systematic sampling. Stratified sampling selects proportionally from subgroups; opportunity uses whoever is available; volunteer relies on self-selection through advertising.

Which of the following is the clearest example of the median being a more appropriate measure of central tendency than the mean?

• AThe data are normally distributed, and so free of odd or extreme values
• BThe data set contains an extreme outlier that would distort the mean✓
• CThe data are nominal categories with no inherent order at all (e.g. eye colour)
• DThe researcher wants to report the most frequently occurring value in the set

Answer: The median is preferred over the mean when the data are skewed by an extreme outlier, because the median is not distorted by anomalous values. Nominal data require the mode, and the most frequent value is by definition the mode.

Why is a repeated measures design particularly vulnerable to order effects, and how are they typically controlled?

• AParticipants differ from one another between the two conditions, so individual differences (participant variables) can build up; controlled by random allocation to groups
• BThe same participants take part in all conditions, so practice or fatigue can affect later conditions; controlled by counterbalancing✓
• CDemand characteristics cause the order effects, because participants work out the aim and deliberately alter their answers; controlled by a single-blind procedure
• DSample sizes are always smaller in this design; controlled by switching to matched pairs or independent groups

Answer: In repeated measures the same people do every condition, so performance can improve (practice) or worsen (fatigue/boredom) across conditions, an order effect. Counterbalancing (e.g. ABBA) varies the order so these effects are spread evenly across conditions.

Which of the following best describes the difference between internal validity and external validity?

• AInternal validity concerns whether the results generalise to other settings and populations; external validity concerns whether the IV, rather than a confounding variable, caused the change in the DV
• BBoth terms refer to the consistency of a measurement when it is repeated over time (reliability)
• CInternal validity refers to ecological validity; external validity refers to temporal validity
• DInternal validity concerns whether the IV (not a confounding variable) caused the change in the DV; external validity concerns whether results generalise beyond the study✓

Answer: Internal validity asks whether the change in the DV was genuinely caused by the IV rather than by confounding variables. External validity (which includes ecological and temporal validity) concerns whether findings generalise to other settings, people and times.

According to the BPS Code of Ethics, which procedure most directly addresses the ethical issue of deception in a study where the true aim could not be revealed beforehand?

• AObtaining fully informed consent from each participant before the study begins, including the true aim
• BMaintaining the confidentiality and anonymity of all participant data
• CDebriefing participants fully after the study and offering the right to withdraw their data✓
• DConducting a cost-benefit analysis through an ethics committee (a research panel)

Answer: When deception is unavoidable, a full debrief afterwards is the main way to deal with it: participants are told the true aim and given the right to withdraw their data. Informed consent cannot fully address deception because the aim was hidden at the outset.

What is an independent variable?

• AThe variable which the researcher measures as the study outcome
• BA variable which varies systematically alongside the independent one
• CThe variable the researcher manipulates to see its effect✓
• DA variable held constant throughout the investigation

Answer: The IV is manipulated and the DV measured. A variable that changes systematically alongside the IV is a confounding variable, and one that simply adds noise is extraneous.

What is the difference between a Type I and a Type II error?

• AType I rejects a true null, Type II retains a false null✓
• BType I retains a false null, Type II rejects a true null
• CType I occurs in one-tailed tests, Type II in two-tailed tests
• DType I applies to parametric tests, Type II to non-parametric

Answer: A Type I error is a false positive and becomes more likely with a lenient significance level such as 0.10. A Type II error is a false negative and becomes more likely with a stringent level such as 0.01.

What are demand characteristics?

• ACues leading researchers to record their results in a biased way
• BCues leading participants to guess the aim and change behaviour✓
• CFeatures of the sample which make it unrepresentative of the population
• DRequirements a study must meet to be ethically approved

Answer: Participants may show the please-U or screw-U effect once they think they know the aim. Single-blind procedures and deception reduce it, at the cost of raising ethical concerns.

Why is a matched pairs design used?

• AIt controls participant variables without order effects✓
• BIt controls order effects but not participant variables
• CIt requires the smallest number of participants possible
• DIt removes the need for random allocation altogether

Answer: Participants are paired on relevant characteristics and split across conditions, so it combines a strength of repeated measures with one of independent groups. The cost is time-consuming matching and a larger sample.

What is counterbalancing used for?

• AControlling participant variables in an independent design
• BEnsuring the sample represents the target population
• CEnsuring the researcher remains blind to the condition
• DControlling order effects in a repeated measures design✓

Answer: Half the participants do condition A first and half do B first, so practice and fatigue effects are spread evenly rather than favouring one condition. It does not remove the effects, only balances them.

What does a correlation coefficient of −0.85 indicate?

• AA strong negative relationship between the two variables✓
• BA weak negative relationship between the two variables
• CA strong positive relationship once the sign is ignored
• DA relationship that is not meaningful between the two variables

Answer: Magnitude gives strength and sign gives direction, so −0.85 is strong and inverse. Correlation never establishes causation, and a curvilinear relationship can produce a coefficient near zero despite a clear pattern.

What is the purpose of a pilot study?

• ATo gather the main findings by using a much smaller sample size
• BTo replicate an earlier study and thereby confirm its conclusions
• CTo obtain ethical approval from the review committee
• DTo test procedures and identify problems before the main study✓

Answer: A small-scale trial run reveals ambiguous instructions, poorly calibrated materials or unworkable timings while they are still cheap to fix. It is about refining method, not collecting data.

When is a sign test used?

• AOrdinal data, independent groups, testing for a difference
• BInterval data, repeated measures, testing for correlation
• CNominal data, independent groups, testing for association
• DNominal data, repeated measures, testing for a difference✓

Answer: The three questions are always level of measurement, design, and difference or association. Chi-square handles nominal data with independent groups and tests association, which is the 'Nominal data, independent groups, testing for association' option.

What does it mean for a result to be significant at p ≤ 0.05?

• AThere is a 95% probability that the hypothesis is definitely true
• BThe result will replicate in 95% of all future investigations
• CThere is a 5% or lower probability the result is due to chance✓
• DThe effect found is large enough to matter in practice

Answer: Significance is about probability, not proof, and says nothing about effect size or practical importance. A very large sample can make a trivial difference statistically significant.

A researcher compares the number of words recalled by a group who learned using mnemonic images (n = 10) and a group who learned by repetition (n = 10). Her hypothesis is non-directional, and she calculates U = 18. Critical values of U for n1 = 10, n2 = 10 (two-tailed): p ≤ 0.05: 23 p ≤ 0.01: 16 For the Mann-Whitney test, U must be equal to or less than the critical value to be significant.

Which conclusion is correct?

• ASignificant at both p ≤ 0.05 and p ≤ 0.01, as 18 is larger than the value of 16
• BSignificant at p ≤ 0.05 but not at p ≤ 0.01, as 18 is below 23 but above 16✓
• CNot significant at p ≤ 0.05, as 18 is smaller than the critical value of 23
• DSignificant at p ≤ 0.05 provided she had predicted the direction beforehand

Answer: For the Mann-Whitney test the calculated U must be equal to or less than the critical value. 18 is below 23, so the result is significant at p ≤ 0.05 (two-tailed), but 18 is above 16, so it does not reach the stricter p ≤ 0.01 level. Treating a larger U as 'more significant' reverses the rule for this test; that 'bigger is better' rule belongs to tests such as chi-squared and Spearman's rho. No directional prediction is needed, because the result is already significant in the two-tailed column.

A researcher predicts a positive correlation between hours spent on social media each week and self-rated loneliness. For 10 participants she calculates Spearman's rho = −0.61. Critical value for N = 10, one-tailed, p ≤ 0.05: 0.564 The calculated value must be equal to or greater than the critical value to be significant.

What should she conclude?

• AHer hypothesis is supported at p ≤ 0.05, as 0.61 is greater than the critical value of 0.564
• BHer hypothesis is not supported, as the correlation is in the opposite direction to her prediction✓
• CHer hypothesis is supported, as a negative value shows loneliness rising as social media use rises
• DHer hypothesis cannot be tested, as Spearman's rho is a test of difference rather than correlation

Answer: A one-tailed (directional) hypothesis is supported only by a result in the predicted direction. A coefficient of −0.61 is a moderately strong NEGATIVE relationship: more social media use went with lower loneliness. Its size passes 0.564, but the sign contradicts the predicted positive correlation, so she must retain the null hypothesis rather than claim support. A negative sign means one variable falls as the other rises, and Spearman's rho is a test of correlation, so it was the right test.

In a pilot study, the mean number of spelling errors per pupil fell from 16 before a phonics programme to 12 after it.

What is the percentage decrease in the mean number of errors?

• A33.3%
• B25%✓
• C4%
• D75%

Answer: Percentage change = change ÷ original value × 100 = (16 − 12) ÷ 16 × 100 = 25%. Dividing the change by the new value of 12 gives 33.3%, the most common slip. 4 is the raw drop in errors, not a percentage, and 75% is the 'after' score expressed as a percentage of the 'before' score.

Reaction times (ms) for five participants: 300, 890, 295, 310, 305.

Which measure of central tendency best represents a typical score in this data set, and what is its value?

• AThe mean, 420 ms, as it takes account of every score in the data set
• BThe median, 295 ms, as it is the middle value of the five scores listed
• CThe mode, 300 ms, as it is the score that occurs most often in the set
• DThe median, 305 ms, as the 890 ms score would pull the mean upwards✓

Answer: In order the scores are 295, 300, 305, 310, 890, so the median is 305 ms. The mean is 2100 ÷ 5 = 420 ms, which is higher than four of the five scores because of the 890 ms outlier, so it misrepresents a typical score. The median must be found after ordering the data (295 is simply the middle of the unordered list), and there is no mode because every score appears once.

Two groups completed a 20-item memory test. Group A: mean 14.2, standard deviation 1.1 Group B: mean 14.0, standard deviation 5.6

Which interpretation of these descriptive statistics is correct?

• AScores in Group B were far more spread out, so its mean is a less representative summary✓
• BGroup B did better overall, as a larger standard deviation shows higher average scores
• CGroup A contained more extreme scores, as its standard deviation is much smaller than B's
• DThe two groups differ significantly, as their standard deviations differ so much in size

Answer: Standard deviation measures how far scores spread around the mean. The means are almost identical, but Group B's standard deviation is about five times larger, so B's scores ranged widely and its mean describes individual scores less well. A larger standard deviation says nothing about which group scored higher, a smaller one means FEWER extreme scores, and a significant difference can only be established with an inferential test.

In a puzzle study, a researcher timed each participant in seconds. She then discarded the times and kept only each participant's finishing position (1st, 2nd, 3rd and so on).

What level of measurement are the finishing positions, and what information has been lost?

• AOrdinal; the size of the gaps between the participants' times✓
• BNominal; the categories into which the participants were placed
• CInterval; the order in which the participants finished the task
• DOrdinal; the order in which the participants finished the task

Answer: Finishing positions are ranks, which is ordinal data. Ranking keeps the order of finishing but throws away how far apart the participants were: 1st and 2nd might be separated by one second or by a minute. The original times in seconds had equal units and were interval-level data. The order is exactly what ranks preserve, so it has not been lost.

To test whether a revision app improves test scores, a researcher pairs students on prior attainment and gives one member of each pair the app. Scores are percentages and are normally distributed.

Which inferential test should be used?

• AAn unrelated t-test, as each condition contains different people
• BA Mann-Whitney U test, as the two groups are independent samples
• CA chi-squared test, as scores are being compared across two groups
• DA related t-test, as matched pairs are treated as related data✓

Answer: This is a matched pairs design: each score in one condition is compared with the score of its matched partner, so the data are treated as related, just as in repeated measures. The data are interval and normally distributed and a difference is being tested, so the parametric related t-test fits. The unrelated t-test is tempting because the conditions contain different people, but it ignores the matching. Mann-Whitney is for unrelated ordinal data and chi-squared needs nominal data.

Participants are given a drink and told either that it contains alcohol or that it is alcohol-free. The researcher who hands out the drinks also scores each participant's performance on a driving simulator, and she knows which drink each participant received.

Which change would most directly reduce the main threat to validity in this procedure?

• AA single-blind procedure, so the participants are not told which drink they received
• BA repeated measures design, so participant variables are held constant across conditions
• CA larger random sample, so the findings generalise to the wider population of drivers
• DA blind scorer, so the person scoring performance does not know the condition✓

Answer: Because the scorer knows each participant's condition, her expectations could bias how she scores or interacts with them: an investigator effect. A blind scorer, such as a second researcher who does not know the conditions scoring the simulator, removes this. A single-blind procedure deals only with the participants' expectations and leaves the scorer's bias in place; repeated measures and sample size address participant variables and generalisability, not investigator effects.

A school compares two reading schemes. Class 1 uses Scheme X with Mr Patel in morning lessons; Class 2 uses Scheme Y with Ms Ade in afternoon lessons. Class 1's reading ages improve more over the term.

Which is the most serious flaw in concluding that Scheme X is more effective?

• AThe reading scheme is the dependent variable, so it should be measured, not changed
• BThe study uses a repeated measures design, so order effects distort the comparison
• CTeacher and time of day vary with the scheme, so both could be confounding variables✓
• DGains in reading age are qualitative data, so they cannot be analysed statistically

Answer: The IV is the reading scheme, but the teacher and the time of day change along with it, so any of the three could explain Class 1's greater gain: these are confounding variables. A fairer test would use the same teacher and timing, or randomly allocate pupils to schemes. The scheme is the IV, not the DV; the design is independent groups, so order effects do not arise; and reading ages are quantitative scores.

Two observers watch the same recordings of children's play and tally 'aggressive acts'. Their tallies correlate at only +0.42.

Which change would most improve the reliability of the observations?

• AAsk a single observer to record the sessions so no comparison has to be made
• BSwitch to covert observation so the children do not change how they behave
• COperationalise the behavioural categories and train both observers to use them✓
• DIncrease the number of children observed so tallies come from a larger sample

Answer: Inter-observer reliability is usually judged acceptable at a correlation of about +0.80; +0.42 shows the observers are applying 'aggressive act' differently. Defining precise, observable categories (for example 'pushes another child') and training the observers together, then re-checking agreement, tackles the cause. Using one observer hides the problem rather than fixing it, covert observation addresses validity rather than consistency (and the sessions are already recorded), and a bigger sample does not make two observers agree.

A questionnaire asks sixth-formers, 'How many units of alcohol do you drink in a typical week?' Students write their name at the top and hand the form to their tutor.

Which change would most improve the validity of the answers?

• AMaking the responses anonymous to reduce socially desirable answering✓
• BUsing closed questions so the data are easier to analyse statistically
• CGiving the same questionnaire a week later to compare the two answers
• DAsking the tutor to read each question aloud to the whole tutor group

Answer: With names attached and a tutor collecting the forms, students have a strong reason to under-report drinking, so answers may not reflect real behaviour: a social desirability bias that lowers validity. Anonymity removes that pressure. Closed questions help analysis but do not make answers more truthful, repeating the questionnaire checks test-retest reliability rather than validity, and reading questions aloud to the group would increase the pressure.

A researcher plans to give a questionnaire on self-esteem to 14- and 15-year-olds during a school lesson. The head teacher has agreed, and the pupils will be asked whether they wish to take part.

Which further step is needed to meet the BPS guidelines on consent for this sample?

• AObtaining informed consent from the pupils' parents or carers as well✓
• BObtaining presumptive consent by asking a different class of similar age
• CObtaining prior general consent from the pupils for future studies too
• DObtaining retrospective consent from the pupils after data are collected

Answer: Under BPS guidance, participants under 16 cannot give informed consent on their own, so parental (or carer) consent is needed in addition to the pupils' agreement and the school's permission. Presumptive, prior general and retrospective consent are ways of handling studies that involve deception; none replaces the consent of a parent for a child participant.

A researcher runs 20 separate significance tests on one data set, each at p ≤ 0.05. One of the 20 tests gives a significant result, and she reports that one alone.

What is the most serious problem with her conclusion?

• AAbout one result in 20 is expected to be genuine at this level, so it is a Type II error
• BA level of p ≤ 0.05 is too strict for a single test, so the result is likely a Type II error
• CAbout one result in 20 is expected by chance at this level, so it may well be a Type I error✓
• DRunning many tests lowers the chance of a false positive, so the finding is understated

Answer: At p ≤ 0.05, each test has up to a 5% chance of a significant result when the null hypothesis is true. Across 20 tests, 20 × 0.05 = 1 chance 'hit' is expected, so a single significant result may be a false positive, which is a Type I error. A Type II error is the opposite: missing a real effect. Running more tests raises, not lowers, the overall risk of a false positive, which is why reporting only the significant test is misleading.

To study exercise and mood, a researcher puts up a poster in a gym asking members to email her if they would like to take part.

Which sampling method is this, and what is its main weakness?

• AOpportunity sampling; the researcher picks whoever is in the gym that day
• BRandom sampling; every gym member has an equal chance of being selected
• CVolunteer sampling; those who respond may share traits that bias the sample✓
• DSystematic sampling; members are picked from a list at a fixed interval

Answer: Participants select themselves by replying to an advert, which is volunteer (self-selected) sampling. People who reply may be unusually keen, confident or interested in exercise, so the sample may not represent gym members in general: volunteer bias. In opportunity sampling the researcher approaches whoever is available; random and systematic sampling would need a list of members chosen by chance or at fixed intervals.

A researcher compares the memory test scores of people who already smoke with those of people who have never smoked.

Which type of experiment is this?

• AA field experiment, as the IV is manipulated in people's usual setting
• BA laboratory experiment, as people are randomly allocated to each group
• CA quasi-experiment, as the IV is an existing difference between people✓
• DA natural experiment, as the researcher changes the IV in a real setting

Answer: Smoking status is a characteristic the participants already have; the researcher cannot manipulate it or randomly allocate people to it, which makes this a quasi-experiment. In field and laboratory experiments the researcher manipulates the IV, and in a natural experiment the IV is changed by an event outside the researcher's control rather than by the researcher.

Participants learned a word list and were tested 24 hours later. Mean words recalled (out of 25): Caffeine at learning, caffeine at recall: 15 Caffeine at learning, placebo at recall: 10 Placebo at learning, placebo at recall: 14 Placebo at learning, caffeine at recall: 11

Which conclusion is best supported by these data?

• ARecall was better when the drug state at recall matched the state at learning✓
• BCaffeine improved recall, whether it was taken at learning or at recall
• CRecall depended mainly on whether caffeine was taken at the recall stage
• DRecall depended mainly on whether caffeine was taken at the learning stage

Answer: Matched states give a mean of (15 + 14) ÷ 2 = 14.5 words, mismatched states (10 + 11) ÷ 2 = 10.5, a gap of 4 words. Caffeine at learning averages (15 + 10) ÷ 2 = 12.5, the same as placebo at learning (14 + 11) ÷ 2 = 12.5; caffeine at recall averages 13 against 12 for placebo at recall, a gap of only 1. So matching, not the drug itself, drives recall, which is the pattern of state-dependent forgetting. Caffeine did not help on its own: taken at learning only, it gave 10 words against 14 for placebo throughout.

Participants witnessed a staged accident either alone or as part of a group, and the researcher recorded whether each one helped. Alone: helped 24, did not help 6 (row total 30) In a group: helped 16, did not help 34 (row total 50) Column totals: helped 40, did not help 40; grand total 80

For a chi-squared test, what is the expected frequency for the 'alone, helped' cell?

• A24
• B15✓
• C20
• D25

Answer: Expected frequency = row total × column total ÷ grand total = 30 × 40 ÷ 80 = 15. 24 is the observed frequency for that cell, which is compared with the expected value rather than being it. 20 assumes the 80 participants are spread evenly over the four cells, ignoring the unequal row totals, and 25 uses the 'in a group' row total of 50 instead of 30.