Abstract:
|
Replication of scientific research is a current hot topic in the media. Research studies have been under fire because their results have failed to replicate either in another setting possibly due to lack of standardization of procedures, lack of documentation, or misuse of statistical techniques. The latter point emphasizes that statistical replication is an important part of reproducibility of research. The International Initiative for Impact Evaluation (3ie) funds impact evaluations and systematic reviews to generate evidence for development programmes. As part of a grant from 3ie, two PhD biostatistics students worked to reproduce the results presented in high impact papers using the author's raw data and reported statistical methods. The outcomes of the paper were also subjected to robustness tests and sensitivity analysis, and extended with a theory of change analysis. The analysis plan for the replication was finalized prior to obtaining the data and posted on the 3ie website for transparency. The PhD students involved in the replication work obtained a clear understanding of the importance of communicating statistical methods in publications. The students gained experience in communicating statistical techniques in writing, coding in various platforms, multiple statistical techniques that could be applied as robustness checks, and the importance of documentation.
|