Research methods in psychology: practice questions with explanations

1Why psychology needs research methods

Psychology makes claims about people: that sleep improves memory, or that a therapy reduces anxiety. Research methods are the rules for testing such claims fairly, so that the answer does not depend on opinion.

A study starts with a hypothesis, a testable prediction. The null hypothesis says there is no effect or no relationship; the alternative hypothesis says there is one. A directional (one-tailed) hypothesis also states the direction of the effect; a non-directional (two-tailed) one does not.

In an experiment, the researcher changes one variable and measures another:

• The independent variable (IV) is the one the researcher deliberately changes, such as the amount of sleep.
• The dependent variable (DV) is the one measured to see if it is affected, such as a memory score.
• Extraneous variables could also affect the DV. If one of them varies with the IV, it becomes a confounding variable and the result is unclear.

Operationalisation means defining a variable so it can be measured, for example “memory” as “the number of words recalled from a list of 20”.

2Designs, samples and ethics

Experimental designs differ in who takes part in each condition:

• Independent groups: different people in each condition. There are no order effects, but the groups may differ.
• Repeated measures: the same people do every condition. Individual differences are controlled, but order effects (practice, fatigue) appear; counterbalancing (half do A then B, half do B then A) controls them.
• Matched pairs: different people, matched on key traits such as age or IQ.

In a true experiment participants are randomly assigned to conditions. In a quasi-experiment the IV is a naturally existing difference, such as gender or age, so it cannot be assigned. A field experiment manipulates the IV in a natural setting.

Non-experimental methods include correlational studies, observations, questionnaires and interviews, case studies, and longitudinal studies that follow the same people over time.

Sampling: in a random sample everyone has an equal chance; systematic takes every nth person from a list; stratified keeps the population’s proportions of subgroups; opportunity uses whoever is available; volunteer uses people who respond to an advert.

Bias: participants may change behaviour because they guess the aim (demand characteristics) or want to look good (social desirability). A double-blind design, where neither participants nor experimenters know the condition, reduces this and experimenter bias.

Ethics: informed consent, the right to withdraw, protection from harm, confidentiality, and a debrief, especially when deception was used.

3Reliability, validity and statistics

Reliability is consistency: a measure gives the same result each time. It is checked by test–retest, by inter-rater agreement between observers, or by split-half and internal consistency.

Validity is accuracy: the study measures what it claims to. Internal validity means the IV really caused the change in the DV. External validity means the results generalise to other people and settings.

A measure can also show concurrent validity (it agrees with an established test) and predictive validity (it predicts a later outcome). A measure can be reliable without being valid.

Levels of measurement: nominal (categories), ordinal (ranks), interval (equal steps) and ratio (equal steps and a true zero, such as reaction time).

Descriptive statistics: mean, median and mode describe the centre; the standard deviation describes the spread. In a normal distribution about 68% of scores lie within 1 SD of the mean and about 95% within 2 SD.

Inferential statistics test the null hypothesis. If p < 0.05, the result is called significant. Two errors are possible:

• Type I error: rejecting a null hypothesis that is true, a false positive.
• Type II error: failing to reject a null hypothesis that is false, a false negative. Small samples make it more likely.

Choose the test by design and data: an independent t-test for two separate groups, a paired t-test for repeated measures, chi-square for categories, Pearson’s r for a linear relationship and Spearman’s rho for ranks.

A p-value says nothing about how big an effect is. Effect sizes, such as Cohen’s d or r, and confidence intervals show that.

4How to evaluate a study

Many questions describe a study and ask what can be concluded. Work through it in this order:

1. Identify the variables: the IV and DV, or the two measured variables in a correlation.
2. Name the design: true experiment, quasi-experiment, correlational, observation or case study.
3. Look at the sample: how it was chosen, how large it is, and who the results apply to.
4. Look for threats: confounding variables, demand characteristics, missing control groups, unreliable measures.
5. Check the statistics: the right test, the p-value and the effect size.
6. State the conclusion the design allows. Only a well-controlled experiment supports a causal claim.

5A worked example

Evaluate the study one step at a time. Try to say the next step before you open it.

6Common mistakes

• Treating correlation as causation. A third variable or reversed direction may explain the link.
• Swapping the IV (changed) and the DV (measured).
• Swapping Type I (false positive) and Type II (false negative) errors.
• Reading p as the probability that the null hypothesis is true. It is the probability of data this extreme if the null hypothesis were true.
• Concluding “no effect” from a non-significant result in a small sample. The study may lack power.
• Confusing reliability (consistent) with validity (accurate).

7How exam questions look

• Easy (key terms and facts): definitions. What an IV, a null hypothesis, reliability or informed consent is.
• Medium (applying the concepts): a short study to classify. You name the design, the sampling method, the type of validity or the right statistical test.
• Hard (cases and analysis): a study with a problem. You spot a confound, p-hacking, a restricted range or a WEIRD sample, interpret R² or a confidence interval, or apply a Bonferroni correction.

For hard questions, ask what else could explain the result and whether the design allows the conclusion that is drawn.