91ɫ��Ƭ

Replications are widely recognized as crucial for establishing the reliability of scientific findings. In fact, they are not very common. There are some good reasons why replications are relatively rare. One reason is that there is little consensus about what we can learn from a replication. The first part of this talk will discuss some of the difficulties with defining “replication success.” As a practical application, I will relate that discussion to the challenges facing DARPA’s SCORE (“Systematizing Confidence in Open Research and Evidence”) Project. I will then connect this to a wider discussion of what we can learn from any empirical analysis. This is where I will talk about significance testing, confidence intervals, and p-values. I will then suggest a way forward involving replication and meta-analysis. Given the brief time available, this talk is aimed at providing a broad overview of some key issues related to replication and scientific reliability.