Replication in science is the process of repeating a scientific investigation, experiment or analysis to determine whether a finding can be obtained again under the same or appropriately comparable conditions. Replication is central to empirical science because a scientific observation that occurs only once may result from chance, measurement problems, uncontrolled variables, methodological artefacts or other circumstances specific to the original investigation.
A replication can be close to the original study, using substantially the same methods and conditions, or it can deliberately alter particular conditions to test whether a finding generalises. Replication therefore provides evidence about the stability, reliability and scope of a scientific result. It does not necessarily require obtaining exactly the same numerical value because measurements naturally contain variability and uncertainty. The relevant question is usually whether the replicated observations are sufficiently consistent with the original claim given the expected variation, statistical uncertainty and experimental design.
Direct replication attempts to reproduce the essential methods and conditions of an earlier study as closely as practical. Conceptual replication tests the same underlying hypothesis using different methods, measurements, populations or experimental procedures. Direct replication is useful for determining whether a reported effect can be obtained under closely similar conditions, while conceptual replication can provide evidence that the finding does not depend entirely on one particular operationalisation. Neither form automatically establishes truth.
A failed direct replication can occur because the original finding was false, because the replication was methodologically different, because the phenomenon is context dependent, because statistical power was inadequate or because the effect is genuinely variable. Consequently, interpreting replication requires examination of design, sample size, measurement quality, effect size, uncertainty, protocol fidelity and contextual differences rather than treating replication as a simple success or failure binary.
Replication is distinct from reproducibility. Reproducibility often involves obtaining the same result from the same underlying data, methods or computational workflow, whereas replication normally involves obtaining new observations or independently conducting the investigation. Reanalysis of an existing dataset can be reproducible without being a replication because no new empirical observations have been collected. Conversely, a new experiment can be a replication even if the exact original computational environment is unavailable.
The terminology is not perfectly standardised across disciplines, so researchers should specify what was repeated. Other concepts such as repeatability, reproducibility, robustness and generalisability also overlap. Repeatability often concerns consistency under the same conditions and within the same laboratory or operator, while reproducibility can concern agreement across changed conditions or independent laboratories. These distinctions are particularly important when evaluating claims about scientific reliability.
Statistical power is important when interpreting replication studies. A study with insufficient sample size may fail to detect an effect that genuinely exists, producing a false impression that the original finding did not replicate. Power depends on factors including effect size, sample size, variability, significance criteria and study design. Replication studies should therefore be planned using an appropriate statistical framework rather than simply repeating the original sample size without considering the intended inference.
Effect estimates and confidence intervals can provide more informative evidence than relying solely on whether a significance threshold was crossed. If an original study reports a small effect with substantial uncertainty, a replication may produce a somewhat different estimate without contradicting the underlying phenomenon. Scientific replication is therefore about evaluating evidence quantitatively and contextually rather than demanding identical numerical outcomes.
Replication has particular importance in fields where results can be affected by laboratory conditions, biological variability, investigator decisions or complex analytical pipelines. In experimental sciences, independent laboratories may use different equipment, reagents, environmental conditions and personnel. In psychology, medicine and social science, participant populations and cultural contexts can influence results. In ecology, seasonal and geographical variation can be substantial. In computational science, differences in datasets, software versions and hardware can influence results.
These realities mean that replication studies need sufficiently detailed protocols and appropriate controls. Independent replication can be especially informative because it tests whether a finding depends on the original researcher's specific procedures or expectations. Multi laboratory replication can provide stronger evidence about robustness across settings than repeated experiments conducted by the same team under identical circumstances.
Replication can also reveal boundary conditions. Suppose an effect occurs in one population but not another. That does not necessarily mean one study is correct and the other is wrong. The difference may indicate that the effect depends on age, environment, genotype, dosage, socioeconomic conditions, temperature or another moderating variable. Replication can therefore lead to refinement rather than simple confirmation or rejection of a hypothesis. Researchers can compare effect sizes across studies using meta analysis and investigate heterogeneity among results. A heterogeneous set of findings may indicate differences in underlying conditions rather than random inconsistency. Replication thus contributes not only to verification but also to discovering when and where scientific relationships operate.
Replication has historical significance because the reliability of scientific knowledge depends on independent observations rather than authority alone. Modern research practices increasingly encourage preregistration, transparent protocols, data sharing, independent analysis and registered reports to strengthen replication efforts. Registered reports can establish research questions and methods before results are known, reducing some forms of selective reporting. Meta analysis can combine evidence from multiple studies, although its conclusions depend on study quality and the handling of heterogeneity and publication bias. Replication initiatives have also highlighted the importance of distinguishing statistical significance from practical importance. A replicated effect may be statistically detectable but scientifically trivial, while an important effect may be difficult to estimate precisely.
A replication should therefore be judged against the claim actually being tested. Exact duplication is neither always possible nor always scientifically desirable. Good replication preserves the critical causal or observational structure of the original study while documenting justified changes. Researchers should report deviations from the original protocol, differences in population or equipment and analytical decisions. A non replication is informative when it is carefully designed and transparently interpreted, just as a successful replication strengthens confidence without proving universal truth.
Replication is best viewed as a process for accumulating independent evidence about the reliability, magnitude, conditions and limitations of scientific claims. It is one of the principal mechanisms through which scientific communities distinguish robust phenomena from findings that depend on chance, artefact or narrowly specific circumstances.
