Steve Juggins has a new paper in QSR with the provocative title “Quantitative reconstructions in palaeolimnology: new paradigm or sick science?”. It is based on a presentation that Steve gave at INQUA in Bern in 2011, and will interest a lot of people.
Steve’s strategy is to take a fresh look at the assumptions of transfer functions, and test what happens when they are violated. The short answer is bad things.
These are the assumptions:
1. The taxa in the modern training-set are systematically related to the environment in which they live.
2. The environmental variable(s) to be reconstructed is, or is linearly related to, an ecologically important determinant in the system of interest.
3. The taxa in the training-set are the same biological entities as in the fossil data and their ecological responses to the environmental variable(s) of interest have not changed over the time represented by the fossil assemblage.
4. The mathematical methods adequately model the biological responses to the environmental variable(s) of interest and yield numerical models that allow accurate and unbiased reconstructions.
5. Environmental variables other than the one of interest have negligible inﬂuence, or their joint distribution with the environmental variable does not change with time.
He focuses on assumptions 2 and 5 which are rarely critically challenged, using a variety of tests to show how bad things can get if these assumptions are violated.
Starting with a training set from Danish coastal waters where depth is the single most important predictor of diatom assemblages, he makes a depth transfer function which yields nonsense reconstructions when applied to a core from Roskilde Fjord. The moral being that the diatoms do not directly respond to depth, but instead respond to several variables, some unknown and unmeasured, that correlate with depth. Only if these correlations are constant through time, which is unlikely when nutrient loadings vary, will depth reconstructions be valid with this transfer function.
Next, he shows that the diatom-pH optima from one training set are well correlated with their optima in other training sets. The same is not true for diatom-DOC and diatom-temperature optima, suggesting that these may not yield robust transfer functions.
Finally, using simulated species data, Steve shows that if one environmental variable changes over time, substantial changes may be reconstructed in a second variable that was actually constant. Although this effect is more pronounced when the two environmental variables are highly correlated in the training set, it occurred even when the variables are orthogonal.
While much of this is alarming, none of this should be controversial, but does it justify the title: sick science?
John Birks thinks not, and has sent out an email supporting the text but bemoaning the title.
The title comes from a paper by Dan Simberloff (1980), “The Sick Science of Ecology: Symptoms, Diagnosis, and Prescription”, published in an obscure Finnish journal, which complains about the reliance on untested models “… as remote from biological reality as are faith-healers” in ecology. Simberloff (1980) lists five symptoms of a sick science (from Birks 1997):
(1) Active censorship by the ‘ecological establishment’ of opposing ideas;
(2) A dominance in ecology of theoretical ideas unrelated to real-life field situations;
(3) An obsession of a holistic view of nature organised into integrated, high-level entities with emergent properties despite there being little or no evidence to support this model;
(4) An obsession to gain the approval of physicists, so-called ‘physics-envy’; and
(5) The input of large research grants into ecology with the creation of large research groups whose main concern becomes self-survival, namely to attract further large research funds rather than to do critical research.
While there are certainly examples of several of these, I’m not convinced that they pervade palaeoecology. But nor am I convinced that these are the only symptoms of sickness. Certainly, a science that publishes without challenge many of the dubious transfer functions listed by Steve is not a healthy science.
Steve cautions that “we should not throw out the baby with the bathwater”. We might need a sieve.
What amuses me most, is that Steve has been previously cautioning me not to be too negative about transfer functions, and then he produces a paper with a title far more negative than anything I have published.