Replication is an important tool to promote high quality research and ensure policy makers can rely on studies in making guidelines or funding programs. By ensuring influential studies are replicable we provide assurance that the policies based on these studies are wellfounded and the conclusions and recommendations are robust-to different estimation models or different choices. In this paper, we argue that replication is not only useful but necessary to ensure that an author's choice in how to analyse data is not the only factor that determines whether an intervention is effective or not. We also show that while most research is done well and provides robust results, small differences can lead to different interpretations and these differences need to be acknowledged. This special issue highlights 5 such replication studies, which are replications of influential studies on biomedical, social, behavioural and structural interventions for HIV prevention and treatment. We reflect on their findings. Four out of five studies, which conduct push button replication and pure replication, were able to reproduce the results of the original studies with minor differences, mainly due to minor typographical errors or rounding differences. The analysis of the measurement and estimation analyses conducted in these five studies reveals that the original results are not very robust to alternative analytical approaches, especially when these results rely on a small number of observations. In these cases, the original results are weakened. Furthermore, in contrast to the original papers, two of the five included replication studies conducted a theory of change analysis-to explore how or why the interventions work (or do not) not just whether the intervention works or not. These two analyses indicate that the estimated impacts of the interventions are drawn from few mediators. In addition, they demonstrate that, in some cases, a lack of effect may be related to lack of adequate exposure to the intervention rather than inefficacy of the intervention per se. However, overall, the included replication studies show that the results presented in the original papers are trustworthy and robust, especially when based on larger sample sizes. Replication studies can not only verify the results of a study, they can also provide additional insights on the published results, such as how and why an intervention was effective or less effective than expected. They can thus be a tool to inform the research community and/ or policymakers about whether and how interventions could be adopted, which need to be tested further, and which should be discontinued because of their ineffectiveness. Thus, publishing these replication studies in peer-reviewed journals makes the work public and publicized. The work advances knowledge, and publication should be encouraged, as it is for other types of research. Copyright:
