Traditionally, the scientific process has been based on prepublication screening through peer review and postpublication validation by independent reproduction. Of course all data published in reputable journals must be reproducible within the original lab as a condition of publication. However, before a finding can attain 'dogma' status, it must pass the independent reproducibility test.

When publication pressures were more manageable, and before the old adage 'publish or perish' emerged as a primary driving force in the molecular biosciences, reproducibility was still a key step in the scientific process. Scientists at the beginning of their academic career often learned their craft by reproducing published data — a bit like teaching art by copying the great masters.

These days, scientists request reagents from each other more than ever; however, the primary aim is not to reproduce, but to move to the next step. It is exceedingly difficult to convince a postdoc to spend months reproducing a complex set of experiments when the outcome is either an unpublishable confirmation, or a lack of confirmation, which would require much more work to ensure that the case made is watertight and often result in the publication of an abbreviated refutation (see editorials, May and November 2005). The PI will worry about the significant drain on resources that a rock solid refutation requires, and the drain on morale that may result from a protracted fight for acceptance of negative data by the original author and the broader community.

Consequently, competitive labs are not often motivated to reproduce data; more importantly, it is not something they are encouraged to do. One way to address this would be allocate a percentage of the time of each lab and researcher solely for independent data confirmation. Granting agencies should take these endeavours seriously and give credit for documented evidence of data reproduction. Initiating an online repository for this data would also be worthwhile and the confirmatory nature of the data may allow for curation without full-blown peer review.

Nowadays, somewhat ironically, most data reproduction occurs through related studies published in a similar timeframe. Increased competitiveness has meant that many researchers prefer the relatively safe option of working on predictable projects that are likely to result in publications, in favour of more esoteric research. The result is increased redundancy as everyone jumps on the next obvious question (occasionally primed by exposure to unpublished data). Copublication can seem frustratingly redundant and is certainly not the most efficient way to spend limited resources. However, in the current system, parallel research is increasingly the favoured means of data validation and in the absence of grass roots policy changes, we may have to live with it.