Validation in Statistics and Machine Learning - Abstract
A recent trend in biostatistical journals encourages authors to make the raw data available for reanalysis. This trend has been viewed as unproblematic and useful, and it is indeed important that research is reproducible.
In this contribution I point out some problems in this connection. Meaningful biostatistics takes place in close collaboration between the biostatistician and the subject matter specialist (who understands the scientific questions and collected the data) and this is hard to imitate when another biostatistician attempts a reanalysis far removed from the substantive specialist and thereby (usually) from the substantive context. Unfortunately poorly informed reanalyses are already quite common in our journals.
The desire for arranging access to original data behind biostatistical papers will continue, and it is necessary to incorporate a proper place for the substantive context if reanalyses are to rise above mechanistic imitations.
- Keiding, N. (2010). Reproducible research and the substantive context (with discussion). Biostatistics 11, 376-396.