“They say that science is self-correcting, but it isn´t. Scientists correct it”, says Associate Professor Richard Telford of the UiB Department of Biology. Telford and colleagues recently published a paper in Nature criticizing findings that suggested a drastic impact from human agriculture on the rules that govern how species assemblages are structured as early as 6000 years ago.
Earlier this year, Lyons et al. published a paper that suggested that the proportion of species pairs that are aggregated (co-occurring in the same area) rather than segregated (not co-occurring in the same area) had been stable for three hundred million years before changing around 6000 years ago in response to human activities. The paper got its fair share of media attention and fits into an ongoing ecological debate on what drives the assembly of biotic communities (the gathering of species in an area).
While prehistoric societies might be expected to have had low impacts on their environment, Lyons et al. suggest otherwise, implying that biotic communities were so fragile that a minimal level of human interaction fundamentally changed their assembly rules.
These findings proved so surprising that Telford and colleagues immediately requested access to the data behind the study:
“We took a look at the paper and said: we do not believe this. The paper relies heavily on species presence – absence data, which is insufficiently detailed to substantiate claims of this kind. Their result just didn´t seem credible to us, and after accessing the raw data we immediately started finding problems.”, Telford says.
Datasets had been duplicated, some island and terrestrial datasets had accidentally been mixed, and one dataset had been turned around so that sites became species and species became sites. Some modern datasets included were inappropriate as they were not comparable with the older datasets.
More issues revealed themselves when Telford and colleagues began replicating the analyses:
“There were issues with the breakpoint analysis used to find when the change in the proportion of aggregated species. Their breakpoint dates to almost exactly 6000 years ago, but the confidence interval spans several hundred thousand years. When the modern island data (about a third of the total) are deleted, the breakpoint remains exactly the same, but the confidence interval increases to several million years. We repeated the analysis and found that the breakpoint could have occurred almost any time in the last three hundred million years, if there was one at all. There is simply no evidence anything happened 6000 years ago” Telford continues.
Telford and colleagues also found that the analysis Lyons et al used to identify aggregated and segregated species pairs is very sensitive to dataset size and that this bias can explain some of their results.
“The authors should really have asked themselves what drove their data. They believe their results, and while they ran some tests, the results were surprising enough that they should have been checked very thoroughly. They simply didn´t pay enough attention to their work. Thankfully the findings are open to debate”, Telford finishes.
Telford R. J., Chipperfield J. D., Birks H. H. & Birks J. H. How foreign is the past? Nature 538, http://dx.doi.org/10.1038/nature20096 (2016).