Replication is the process of repeating a scientific investigation, experiment, observation, analysis, or computational procedure to determine whether its findings can be obtained again under the same or appropriately comparable conditions. In science, replication is important because a single result may arise from genuine effects, random variation, measurement error, uncontrolled variables, statistical fluctuations, selective reporting, or methodological mistakes. A replicated finding provides additional evidence that an observed result is not merely an accidental feature of one particular investigation.
Replication therefore contributes to the reliability and cumulative nature of scientific knowledge, although successful replication does not by itself prove that an explanation is universally true. The precise meaning of replication depends on the field: laboratory sciences may repeat experimental procedures directly, psychology may repeat behavioural studies with new participants, astronomy may independently reanalyse observations, and computational research may reproduce analyses using the same or independently implemented methods.
Several forms of replication are distinguished in research methodology. A direct or exact replication attempts to reproduce the original procedures, materials, conditions, measurements, and analytical approach as closely as practicable, whereas a conceptual replication tests the same underlying hypothesis using different methods, populations, operational definitions, or experimental settings.
Reproduction is sometimes distinguished from replication: in computational science, reproduction may mean obtaining the same results from the same data and code, while replication may involve collecting new data to investigate whether the finding recurs. Terminology varies substantially between disciplines, so these distinctions should not be treated as universally standard. Systematic replication programmes can also vary the context deliberately, testing whether an effect survives changes in researchers, locations, samples, instruments, laboratories, or experimental conditions. Such variation can reveal whether a finding is robust, context-dependent, or dependent on particular methodological choices.
Replication is closely connected to reproducibility, reliability, validity, and generalisability, but these concepts are not interchangeable. Reproducibility commonly concerns whether an analysis or result can be obtained again using the same underlying data, computational procedures, or materials, whereas replication usually involves repeating the investigation or obtaining new observations. Reliability concerns consistency of measurement or results, while validity concerns whether a method actually measures or tests what it is intended to measure.
Generalisability concerns whether findings extend beyond the particular population, setting, conditions, or sample studied. A study can therefore be reproducible without being externally generalisable, or a measurement can be highly reliable while consistently measuring the wrong construct. Replication provides evidence relevant to these issues but cannot independently establish all of them.
A replication can fail for many different reasons, and a non-replication should not automatically be interpreted as proof that the original finding was false. Differences in participant populations, environmental conditions, instruments, statistical models, sample sizes, treatment implementation, researcher expertise, or random sampling can alter observed results. An original effect may also be genuine but conditional on a moderator that the replication did not reproduce.
Conversely, an apparent replication failure may expose weaknesses in the original research, such as uncontrolled confounding, publication bias, selective analysis, measurement problems, or an effect that was substantially overestimated. Researchers therefore examine effect sizes and confidence intervals, methodological differences, statistical power, and the distribution of results across studies rather than relying solely on whether an individual replication crosses a conventional significance threshold.
Replication has particular importance in discussions of the scientific research process and the so-called replication crisis. In several research fields, large collaborative projects have attempted to repeat published findings and have reported that some effects were smaller, less consistent, or absent under replication conditions. These observations have encouraged reforms including preregistration, registered reports, larger samples, improved statistical practices, transparent reporting, open data where appropriate, open materials, independent analysis, and greater attention to effect sizes and uncertainty.
The existence of replication difficulties does not mean that scientific research as a whole is unreliable; rather, replication studies provide information about which findings are robust and which require qualification or further investigation. Replication is one component of a broader system of methodological criticism, independent verification, and cumulative evidence.
Statistical considerations are central to interpreting replication. Every empirical result contains sampling variability, so two studies examining the same phenomenon will rarely produce exactly identical numerical estimates. Statistical power, sample size, effect magnitude, variance, measurement precision, and study design all affect the probability of detecting an effect. A replication with insufficient statistical power may fail to detect a real effect, producing a false negative result.
Conversely, repeated attempts with flexible analyses can produce apparently supportive findings by chance if researchers conduct many tests and selectively report successful outcomes. Modern replication assessment therefore often examines effect-size estimates and their uncertainty rather than asking only whether both studies produced statistically significant results. Meta-analysis can combine evidence from multiple studies, although its conclusions can themselves be distorted by publication bias, heterogeneity, selective reporting, or inappropriate statistical assumptions.
Replication is also important outside conventional laboratory experimentation. In computational science, researchers may attempt to reproduce published numerical results from source code, data, software versions, random seeds, and computational environments. In observational research, independent teams can analyse new datasets or conduct studies in different populations. In clinical research, replication may involve repeated trials, independent sites, different patient populations, or subsequent evidence synthesis. In engineering, replicated tests establish whether a component performs consistently across specimens and operating conditions.
In social science, replication can reveal how strongly findings depend on cultural or institutional context. In all of these applications, careful documentation is essential because replication requires sufficient information about procedures, data generation, instruments, analysis, and decision rules for another researcher to perform a meaningful independent test.
The principal value of replication is cumulative: confidence in a scientific claim generally increases when independent investigations repeatedly obtain compatible evidence across appropriate conditions, while inconsistent results identify questions requiring further explanation. Replication does not require numerical identity between studies, because legitimate sampling variation and contextual differences are expected. Nor does a single successful replication establish that a proposition is universally valid.
Scientific knowledge instead develops through multiple observations, experiments, analyses, theoretical explanations, methodological critiques, and independent attempts to test important claims. A robust research culture therefore treats replication as a normal component of scientific practice rather than as an exceptional challenge to an established result. Its deeper significance lies in making claims answerable to evidence obtained by investigators other than the people who originally proposed them.
