Scientific Replication
Advanced · Cần chú ý qualifier, quan điểm và giới hạn của bằng chứng.
A For the present account of scientific replication, a replication repeats a study's essential test to see whether a finding appears with new data. In metascience, the term statistical power refers to the probability that a study detects an effect of a given size when it exists. The definition gives researchers a common starting point for discussing scientific replication, but it does not identify a cause by itself. Two observations of scientific replication can share the label statistical power while differing in scale, timing or origin. B Researchers rely chiefly on registered replications to investigate scientific replication. Research on scientific replication has found that methods and analysis are reviewed before results are known, and several sites may use a shared protocol. They decide their comparison, exclusions and outcome measures for scientific replication in advance. A result about scientific replication is treated as stronger when it survives more than one source of evidence, not simply when one instrument measuring scientific replication reports many decimal places. C The evidence about scientific replication is informative but conditional. One point relevant to scientific replication is that many effects replicate but some shrink or vary, revealing both false starts and important context dependence. Researchers test sampling variation as an explanation. Evidence reviewed for scientific replication shows that results differ by chance because each study observes only part of a wider population. Confidence in sampling variation rises when independent measures of scientific replication agree and rival explanations fail, rather than when a single comparison happens to be statistically precise. D Knowledge of scientific replication accumulated unevenly across laboratory and field research. A striking report could establish that a pattern existed, yet it could not show whether sampling variation operated elsewhere. Researchers examining scientific replication therefore moved toward shared definitions and planned comparisons based on registered replications rather than discarding the earlier record. E Practical programmes translate evidence about scientific replication into action. The discussion of scientific replication notes that researchers share methods, run adequately powered tests and investigate variation instead of treating one failure as final. Their stated focus is reliable evidence. Teams working on scientific replication compare later outcomes with conditions before implementation and record unintended effects. This evaluation of scientific replication determines whether the original explanation involving sampling variation remains useful outside the research setting. F Interpretation of scientific replication must stop short of a universal claim. For the present account of scientific replication, an exact procedure may reproduce an old limitation, while a changed context can make disagreement hard to interpret. Future work on scientific replication is organised around multi-site replication. For future research on scientific replication, multi-site replication will estimate both an average effect and meaningful variation among settings. This use of multi-site replication targets a specific uncertainty about scientific replication rather than merely increasing the volume of data.