Biometrical Journal and Reproducible Research

Abstract

Biometrical Journal serves the scientific community for over 50 years by publishing biostatistical innovations (Bergholt et al., 2008;Victor et al., 2008).These articles make a difference in many fields of our discipline, for example, in clinical trials methodology or survival analysis.Even more important, the journal helps to transfer new theoretical insights and statistical methods to medicine and life sciences.Although being already very successful in this direction, there remains much to be done to further increase the quality of articles published by Biometrical Journal and to increase the scientific impact of this research.Almost all articles in Biometrical Journal present new methodology, and many authors provide mathematical proofs demonstrating correctness or optimality in some sense.These theoretical contributions are carefully checked by referees and Associate Editors.This thorough review process ensures the high quality of accepted articles.Theoretical results are often accomplished by empirical investigations.The majority of authors choose to analyse data previously published elsewhere to demonstrate the practical relevance of their contribution.The performance of new statistical methodology is often studied empirically through simulation experiments.However, empirical investigations rely heavily on data and computer programs.Correct and efficient implementation in a computer program is crucial for simulation experiments, for summarising the results in figures and tables, and for the statistical analysis of real data.Although referees try to do their best to detect inconsistencies in the theoretical parts of an article, it is hard to comment on the validity of simulation experiments or the correctness of a specific data analysis.Indeed, due to the increasing complexity of statistical methodology with many control or tuning parameters, it is often impossible to reproduce published results.For example, in a recent letter to the Editor, Rubio and Pe´rez-Elizalde (2009) have identified numerical problems of a Markov Chain Monte Carlo (MCMC) analysis in Mendoza and Gutie´rrez-Pena (1999).The article is perfectly valid from a theoretical point of view and the MCMC algorithm used is also correct in principle.However, convergence problems and perhaps too small Monte Carlo sample size have led to rather misleading results.Moreover, for many practitioners, it may be hard or even impossible to apply a new method if no software is readily available.Under these circumstances, it is unlikely that a new method will make its way into applications.Thus, an article might not get the credit (usually measured by the number of citations) it deserves.Why?The reason is simple and best described by what de Leeuw (2001) coined Claerbout's principle after the geophysicist Jon Claerbout:An article about computational science in a scientific publication is not the scholarship itself, it is merely advertising of the scholarship.The actual scholarship is the complete software development environment and the complete set of instructions which generate the figures.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 19 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.