Internal replication of computational workflows in scientific research.
Internal replication of computational workflows in scientific research.
复制标题
DOI：
10.12688/gatesopenres.13108.1
复制
发表时间：
2020-01-01
影响因子：
--
通讯作者：
Arnold, Benjamin F
中科院分区：
文献类型：
--
作者：
Benjamin-Chung, Jade;Colford, John M Jr;Arnold, Benjamin F
近年来，从心理学到物理学等科学学科都未能复制研究成果，这引起了越来越多的关注。外部调查人员对已发表的研究结果进行外部复制，已成为发现已发表文献中的错误和偏见的一种方法。然而，在外部复制努力能够确认或挑战原始贡献之前，一些研究会影响政策和实践。在发表之前发现和解决错误将通过提高发表的证据的准确性来提高科学进程的效率。在这里，我们总结了内部复制的基本原理和最佳实践，内部复制是多个独立的数据分析人员复制分析并在发布之前纠正错误的过程。我们解释了内部复制应该如何减少数据分析过程中出现的错误和偏差，并论证了当与预先指定的假设和分析计划相结合并与掩蔽到实验组分配的数据分析师一起执行时，内部复制将是最有效的。通过提高已发表证据的再现性，内部复制应该有助于更快的科学进步。
Failures to reproduce research findings across scientific disciplines from psychology to physics have garnered increasing attention in recent years. External replication of published findings by outside investigators has emerged as a method to detect errors and bias in the published literature. However, some studies influence policy and practice before external replication efforts can confirm or challenge the original contributions. Uncovering and resolving errors before publication would increase the efficiency of the scientific process by increasing the accuracy of published evidence. Here we summarize the rationale and best practices for internal replication, a process in which multiple independent data analysts replicate an analysis and correct errors prior to publication. We explain how internal replication should reduce errors and bias that arise during data analyses and argue that it will be most effective when coupled with pre-specified hypotheses and analysis plans and performed with data analysts masked to experimental group assignments. By improving the reproducibility of published evidence, internal replication should contribute to more rapid scientific advances.