Research Methods in Psychology

Research methods in psychology are the systematic procedures psychologists use to collect and analyze evidence about behavior, thought and emotion. The main methods are experiments (which test cause and effect), correlational studies (which measure how variables are related), and descriptive methods such as observation, surveys and case studies, supported by qualitative approaches and meta-analysis. Each method answers a different kind of question, and each has trade-offs between control, realism and ethics.

Understanding Psychological Research

Psychology is a science because its claims are tested against evidence rather than accepted on intuition or authority. Research methods are the tools that make this possible. They let psychologists move from a hunch ("people seem to work harder when someone is watching") to a precise, testable question, and then to an answer that other researchers can check for themselves.

The same methodological toolkit is used across the discipline. It applies whether the goal is mapping memory and attention in cognitive psychology, tracing change across the lifespan in developmental psychology, or studying group behavior in social psychology. If you are new to the field, our psychology basics guide explains the core ideas these methods build on.

Studying methods also helps you read research critically. Many famous findings, from the Milgram obedience experiment to the Bobo doll experiment, are still debated today largely because of questions about how they were designed, run and reported. Knowing the strengths and weaknesses of each method is what lets you judge how much weight a headline finding deserves.

The Scientific Method in Psychology

The scientific method is a cycle rather than a straight line. Each study answers one question and usually raises several more. In psychology the cycle typically runs through these steps:

1. Observe and question. A researcher notices a pattern (for example, that students who test themselves seem to remember more) and turns it into a focused question.
2. Review the literature. They check what earlier studies have already found, so the new study builds on existing knowledge rather than repeating it.
3. Form a hypothesis. A hypothesis is a specific, testable prediction, such as "students who take practice quizzes will score higher on a final test than students who reread their notes."
4. Operationalize the variables. Abstract ideas such as "memory" or "stress" are defined in measurable terms, for instance "number of words correctly recalled after one week."
5. Design the study and collect data. The researcher chooses a method, recruits a sample, gains ethical approval and gathers data in a standardized way.
6. Analyze and interpret. Statistics show whether the results are likely to reflect a real effect and how large it is.
7. Report and replicate. Findings are written up for peer review, and other researchers attempt to repeat the study. Findings that survive repeated testing are integrated into theory.

Key scientific principles

Several principles run through every stage. Empiricism means knowledge comes from systematic observation, not assumption. Objectivity means procedures are designed to limit the researcher's own expectations and biases. Replicability means a study is described in enough detail that others can repeat it. Falsifiability, a principle associated with the philosopher Karl Popper, means a hypothesis must be stated so that some possible result could show it to be wrong. Parsimony means preferring the simplest explanation that fits the evidence.

Types of research questions

The question a researcher asks largely determines the method. Descriptive questions ("How common is insomnia among students?") call for surveys or observation. Relational questions ("Is screen time related to sleep quality?") call for correlational designs. Causal questions ("Does caffeine improve reaction time?") call for experiments. Developmental questions ("How does working memory change from childhood to old age?") call for longitudinal or cross-sectional designs.

Experimental Methods

An experiment is the only method that can directly establish cause and effect. The researcher deliberately changes one factor, the independent variable (IV), and measures its effect on an outcome, the dependent variable (DV), while holding other conditions constant.

True experiments

A true experiment has two defining features: manipulation of the IV and random assignment of participants to conditions. Random assignment means every participant has an equal chance of ending up in any group, so pre-existing differences (motivation, intelligence, mood) are spread evenly. If the groups then differ on the DV, the most plausible explanation is the manipulation itself.

Variables and control

A confounding variable is any factor other than the IV that varies between conditions and could explain the results. Researchers reduce confounds through several techniques:

• Control groups provide a baseline for comparison, such as a group that receives no intervention or a standard one.
• Placebos control for expectations, which can be powerful in their own right (see the placebo effect). Simply knowing you are being studied can also change behavior, a pattern often discussed under the Hawthorne effect.
• Blinding keeps participants (single-blind) or both participants and researchers (double-blind) unaware of who is in which condition.
• Standardization means every participant receives the same instructions, timing and setting.

Two other types of variable help explain more complex patterns. A moderator changes the strength or direction of a relationship (an intervention might work for adolescents but not adults). A mediator explains how an effect happens (exercise might improve mood partly by improving sleep).

Quasi-experiments

Sometimes random assignment is impossible or unethical. Researchers cannot randomly assign people to grow up in poverty, experience a natural disaster or be a certain age. A quasi-experiment compares pre-existing groups or tracks outcomes before and after an event, for example comparing test scores in schools that did and did not adopt a new reading program. Because the groups may differ in other ways, causal conclusions are more tentative.

Field experiments

Field experiments manipulate a variable in a real-world setting, such as a train carriage, a workplace or a website. They trade some control for realism. Many classic studies of helping behavior, including research inspired by the bystander effect, were run as field experiments. They raise particular ethical questions because participants often do not know they are in a study.

Research Designs

Within experiments, researchers must decide how participants are allocated to conditions. The choice affects how many participants are needed and which problems must be controlled.

Counterbalancing means varying the order of conditions across participants (half do A then B, half do B then A) so that practice or fatigue doesn't systematically favor one condition.

Factorial designs

A factorial design studies two or more independent variables at once. A "2×2" design has two IVs, each with two levels, giving four conditions. Factorial designs reveal main effects (the effect of each IV on its own) and interactions (when the effect of one IV depends on the level of another). For example, background music might impair reading comprehension for introverts more than for extraverts, an interaction between music and personality.

Descriptive and Correlational Research Methods

Not every question can or should be answered with an experiment. Descriptive methods aim to describe behavior accurately, and correlational methods measure relationships. Together they make up a large share of published psychology.

Observational studies

In naturalistic observation, researchers record behavior in its natural setting without intervening, such as coding children's play on a school playground or Jane Goodall's long-term observation of chimpanzees. It offers high ecological validity, but people may behave differently if they know they are watched, and observers may see what they expect to see. Researchers counter this with clear coding schemes and by checking agreement between independent observers.

In structured (controlled) observation, researchers set up a standard situation and record responses. Mary Ainsworth's Strange Situation, a sequence of brief separations and reunions between infant and caregiver, is a famous example used to classify attachment styles. In participant observation, the researcher joins the group being studied, which can give insider insight but makes objectivity harder to maintain.

Survey research

Surveys collect self-reported data from many people using questionnaires or interviews, often with rating scales such as Likert items ("strongly disagree" to "strongly agree"). They are efficient and can reach large, diverse samples. Their weaknesses are well known: people may misremember, misunderstand questions, or give answers that make them look good (social desirability bias). Question wording and order can also shift responses, which is why careful piloting matters.

Case studies

A case study is an intensive investigation of a single person, group or event, usually drawing on interviews, records, tests and observation. Case studies are invaluable for rare conditions. Phineas Gage, a railroad worker whose personality reportedly changed after an iron rod passed through his frontal lobe in 1848, and Henry Molaison ("H.M."), whose surgery in 1953 left him unable to form new long-term declarative memories, shaped neuropsychology and memory research. The limitation is generalization: one unusual case may not represent anyone else, and accounts of famous cases can drift over time.

Correlational studies

A correlational study measures two or more variables and assesses how they move together. The correlation coefficient (r) ranges from −1 to +1. A positive correlation means both variables rise together; a negative correlation means one rises as the other falls; a value near zero means little linear relationship.

Correlational research is still essential. Many important variables, such as childhood adversity, personality traits or brain injury, cannot ethically be manipulated, so measuring their associations is the only option.

Longitudinal and cross-sectional studies

Longitudinal studies follow the same people over time, sometimes for decades. The Harvard Study of Adult Development (which began in 1938 as the Grant Study) and the Dunedin Multidisciplinary Health and Development Study in New Zealand (following people born in 1972–73) are well-known examples. They show how individuals change, but they are costly and lose participants over time (attrition), which can bias results if those who drop out differ from those who stay.

Cross-sectional studies compare different age groups at a single point in time. They are quicker and cheaper, but differences may reflect cohort effects, the distinct experiences of people born in different eras, rather than aging itself. Cross-sequential designs combine both approaches to separate age from cohort.

Data Collection Methods

Whatever the design, researchers need a way to measure their variables. Most studies combine several types of measure, because each has blind spots.

Self-report measures

Questionnaires and interviews ask people directly about their thoughts, feelings and behavior. Closed-ended questions (multiple choice, rating scales) are easy to score and compare; open-ended questions capture richer detail. Interviews range from structured (fixed questions in a fixed order) through semi-structured (a guide with flexible follow-ups) to unstructured (an open conversation). Self-report is the only direct window into private experience, but it depends on people's insight and honesty.

Behavioral measures

Behavioral measures record what people actually do: how often a child shares a toy, how long someone persists at an unsolvable puzzle, or how quickly they respond to a stimulus. Reaction time tasks are a mainstay of cognitive psychology. Behavior is usually coded with a predefined scheme, and reliability is checked by having two coders rate the same material independently.

Physiological and neuroscience measures

Researchers can also measure the body and brain. Heart rate and skin conductance index arousal; hormones such as cortisol relate to stress; eye tracking shows where attention goes. Brain imaging methods include EEG (excellent timing, poor spatial detail), fMRI (good spatial detail, slower timing, based on blood-oxygen changes) and PET. These methods are powerful but expensive, and brain activity alone rarely tells you what someone is thinking; see our guide to neuroscience.

Psychological tests

Standardized tests measure abilities and traits under consistent conditions, compared against norms from a reference group. Examples include intelligence tests such as the WAIS and Stanford-Binet (see IQ testing), personality inventories based on the Big Five, and clinical rating scales used by professionals. Projective tests such as the Rorschach inkblots remain controversial because their reliability and validity are debated.

Archival research

Archival research analyzes data that already exist, such as census records, court files, historical letters, social media posts or large public datasets. It places no burden on participants and can cover long time spans, but researchers have no control over how the data were originally collected or what is missing.

Sampling Techniques

Researchers almost never study an entire population (everyone they want to draw conclusions about). Instead they study a sample, and how that sample is chosen determines how far the findings can be generalized.

Probability sampling

In probability sampling, every member of the population has a known chance of selection, which supports generalization.

• Simple random sampling: every member has an equal chance, like drawing names from a hat. Unbiased in principle, but small subgroups may be missed by chance.
• Stratified random sampling: the population is divided into strata (such as age bands) and people are randomly selected from each in proportion, guaranteeing subgroup representation.
• Cluster sampling: whole groups (such as schools) are randomly chosen and everyone in them is studied. Efficient for spread-out populations.
• Systematic sampling: every nth person on a list is selected after a random start. Simple, but problematic if the list has a hidden repeating pattern.

Non-probability sampling

Non-probability methods are faster and cheaper, and sometimes the only practical option, but they limit generalization.

• Convenience sampling: whoever is easiest to reach, such as university students. Very common, and a major reason for criticism that psychology over-relies on WEIRD samples (Western, Educated, Industrialized, Rich and Democratic), a term popularized by Henrich, Heine and Norenzayan in 2010. See cross-cultural psychology.
• Purposive sampling: deliberately choosing people with specific characteristics relevant to the question.
• Snowball sampling: participants recruit others they know; useful for hard-to-reach groups, though the sample reflects social networks.
• Quota sampling: filling set quotas for demographic groups, but without random selection within each group.

How big should a sample be?

Sample size is planned with a power analysis, which estimates how many participants are needed to detect an effect of a given size. Conventionally, researchers aim for 80% power (an 80% chance of detecting a true effect) with a significance threshold (alpha) of .05. Small effects need much larger samples. Underpowered studies were a major contributor to psychology's replication problems, because they produce unstable and often exaggerated estimates.

Statistical Analysis

Descriptive statistics

Descriptive statistics summarize data. Measures of central tendency describe a typical score: the mean (average), median (middle value) and mode (most frequent value). The median is often preferred for skewed data such as income, where a few extreme values pull the mean upward. Measures of variability describe spread: the range, the variance and the standard deviation (the typical distance of scores from the mean). Many psychological variables, such as IQ scores, are roughly normally distributed in a bell curve.

Inferential statistics and hypothesis testing

Inferential statistics help decide whether a pattern in a sample is likely to reflect something real in the population. Researchers start with a null hypothesis (no effect) and calculate a p-value: the probability of getting results at least this extreme if the null hypothesis were true. A p-value below .05 is conventionally called "statistically significant." A p-value is not the probability that the hypothesis is true, a common misunderstanding.

Two kinds of error are possible. A Type I error is a false positive (concluding there is an effect when there isn't). A Type II error is a false negative (missing a real effect). Common tests include t-tests (comparing two means), ANOVA (comparing three or more means), chi-square (associations between categories), and correlation and regression (relationships and prediction).

Effect sizes and confidence intervals

Statistical significance says little about how big or important an effect is. With a large enough sample, trivial effects become "significant." That is why researchers report effect sizes such as Cohen's d (a standardized difference between means; Cohen suggested roughly 0.2 as small, 0.5 as medium and 0.8 as large) and r. A 95% confidence interval gives a range of plausible values for the true effect; a narrow interval indicates a precise estimate.

Qualitative Research Methods

Qualitative research studies meaning, experience and context through words, images and observations rather than numbers. It is especially useful for new or poorly understood topics, for understanding how people make sense of events, and for giving voice to groups often overlooked in quantitative research.

Main approaches

• Phenomenology explores lived experience, such as what it is like to care for a parent with dementia, usually through in-depth interviews.
• Grounded theory, developed by Glaser and Strauss in 1967, builds theory directly from data by constantly comparing new material with emerging categories until no new themes appear (saturation).
• Ethnography studies a cultural group through extended fieldwork and participant observation, producing "thick description" of everyday life.
• Qualitative case studies examine a bounded system, such as one school or clinic, using multiple sources of data.

Analyzing qualitative data

Thematic analysis, set out in an influential 2006 paper by Braun and Clarke, involves coding data and developing patterns of meaning (themes) across it. Content analysis categorizes material systematically and can count how often categories appear. Narrative analysis examines how people structure stories about their lives and identities.

Because qualitative work does not rely on statistics, it uses different quality criteria: credibility (do the findings faithfully represent participants' views?), transferability, dependability, confirmability and reflexivity (the researcher's openness about how their own background might shape interpretation). Mixed-methods studies combine qualitative and quantitative data to get both breadth and depth.

Reliability and Validity

No research method is useful unless its measurements are both consistent and meaningful. These qualities are captured by reliability and validity. A measure can be highly reliable yet invalid: a bathroom scale that always reads five pounds too heavy is consistent but wrong.

Reliability refers to consistency. Test-retest reliability means the same people score similarly when retested. Internal consistency means items on a scale correlate with one another, often quantified with Cronbach's alpha. Inter-rater reliability means independent observers reach the same judgments. High reliability reduces measurement error and is a prerequisite for validity.

Validity refers to whether a study measures or tests what it intends to. Internal validity concerns whether effects are genuinely caused by the independent variable rather than confounds. External validity concerns how well findings generalize to other people, settings and times. Construct validity asks whether operational definitions truly capture the underlying concept, and ecological validity reflects how closely a study mirrors real-world conditions. Tight laboratory control often improves internal validity at the expense of ecological validity, so good research programs use several designs. The same standards apply to psychological testing, where a test must be reliable and valid before its scores can be meaningfully interpreted.

Research Ethics

Modern research ethics grew out of serious abuses. The Nuremberg Code (1947) followed the trials of Nazi doctors, and the US Belmont Report (1979) followed revelations about the Tuskegee syphilis study. In psychology, studies such as the Little Albert experiment, Milgram's obedience research and the Stanford prison experiment sharpened debate about stress, deception and consent.

Core principles

The Belmont Report identified three principles: respect for persons (autonomy and informed consent), beneficence (minimizing harm and maximizing benefit) and justice (fairly distributing the burdens and benefits of research). The American Psychological Association's Ethical Principles of Psychologists and Code of Conduct and the British Psychological Society's Code of Human Research Ethics translate these into practical rules.

Informed consent, deception and debriefing

Participants should understand the study's purpose, procedures, risks and benefits before agreeing, and they must be free to withdraw at any time without penalty. Children give assent, with parents or guardians giving consent. Deception is permitted only when the research question cannot be answered otherwise, the deception is unlikely to cause significant distress, and participants are fully debriefed afterward. Data must be kept confidential and, where possible, anonymous.

Ethics committees

Before research begins, an independent committee, called an Institutional Review Board (IRB) in the United States or a research ethics committee elsewhere, reviews the protocol. Low-risk studies may be exempt or receive expedited review, while higher-risk studies get full review. Committees also monitor ongoing research and require researchers to report adverse events. Many studies that were acceptable in the 1960s would not be approved today.

Reporting Research

Psychological research is usually written up in APA style, set out in the Publication Manual of the American Psychological Association (7th edition, 2020). A standard empirical report has a title page; an abstract (a short summary, typically up to about 250 words); an introduction reviewing past research and stating hypotheses; a method section detailed enough for replication (participants, materials, procedure); a results section; a discussion interpreting findings, limitations and implications; and references.

Good reports are clear, precise and objective. Results include the test statistic, degrees of freedom, p-value and effect size, for example: F(2, 147) = 8.92, p < .001, η² = .11. Tables and figures should make sense on their own with clear captions. Before publication, most journal articles go through peer review, in which independent experts evaluate the work; peer review catches many problems, but not all.

Replication, Meta-Analysis and the Replication Crisis

A single study is rarely decisive. Confidence grows when findings are replicated, ideally by independent teams. In 2015 the Open Science Collaboration attempted to replicate 100 studies published in three leading psychology journals; while 97% of the originals had reported significant results, only about 36% of the replications did, and replication effects were on average about half the size of the originals. Other famous effects, including some forms of ego depletion and social priming, also proved hard to reproduce.

Researchers identified several causes: small samples, flexible analysis choices (sometimes called "p-hacking"), selective publication of positive results (publication bias), and incentives favoring novel findings. A meta-analysis statistically combines results from many studies to estimate an overall effect size and can test for publication bias, but it is only as good as the studies included.

Current Trends in Research Methods

The replication crisis triggered a broad reform movement often called open science. Key practices include:

• Preregistration: publicly recording hypotheses and analysis plans before collecting data, so they can't be changed after seeing the results.
• Registered Reports: journals agree to publish a study based on its question and methods, before results are known, which reduces publication bias.
• Open data and materials: sharing datasets and code so others can check and reuse them.
• Large collaborative studies: many labs running the same protocol (such as the Many Labs projects) to get big, diverse samples.

Technology is also changing methods. Online experiments reach participants worldwide; ecological momentary assessment uses smartphones to sample experiences in real time during daily life; wearable devices record sleep and activity; and machine learning helps find patterns in very large datasets, raising new questions about privacy and consent (see psychology and AI). Researchers are also working harder to include participants beyond WEIRD populations.