Replication
Test a claim with new observations
Replication gathers new data to assess whether a previous result or theoretically predicted pattern recurs. Direct replications preserve key procedures; conceptual replications change operationalisations while targeting a claim. Computational reproducibility reruns an analysis on existing data and is a related but different task. Replication evaluation needs effect estimates, uncertainty and contextual differences.
Use when an influential claim needs independent testing or its scope is uncertain. Obtain original materials where possible, plan adequate precision and specify which differences would alter the interpretation of success.
Strengths
- Tests whether evidence extends beyond one dataset
- Can reveal boundary conditions and overestimated effects
Limitations
- Protocol differences complicate interpretation
- A failed replication does not uniquely identify its cause
Know the boundary
One significant or nonsignificant p-value is not a complete replication verdict.