Testate amoeba live in bogs, their shell or test is preserved in the peat after death. Some species seem to prefer drier conditions, others prefer wetter conditions. And thus we have the prerequisites for constructing a transfer function to reconstruct bog wetness from fossil testate amoeba. Many such reconstructions have been made: are they any good?
A paper published today in Quaternary Science Reviews tries to address this question. Richard Payne and co-workers compiled 30 fossil testate amoeba stratigraphies and related modern calibration data sets and tested whether the wetness reconstructions are statistically significant using one of the tests in my palaeoSig package (I helped design the study). Perhaps surprisingly, only five of the stratigraphies yielded statistically significant reconstructions at the p = 0.05 level.
I’m not sure why so many of the reconstructions failed the significance test. Perhaps wetness did not change very much. Perhaps the tests I have developed have low power. In the paper that first presented these tests, I discuss several circumstances in which they might have low power, including low diversity, few fossil samples, and poor analogues. None of these appear to explain which reconstructions are significant. As yet unpublished experiments with simulated data show that the significance tests can also have low power if multiple environmental variables change simultaneously.
Since I last worked with testate amoebae, I’ve become aware that it is more or less standard practice to use a strict outlier removal protocol on the calibration set. I do not think this represents good practice: it will artificially improve transfer function performance statistics. Obviously the same protocol cannot be applied to fossil stratigraphies so errant fossil samples cannot be removed but will give erroneous reconstructions. This protocol might partially explain why fewer than expected reconstructions are statistically significant.
A similar manuscript is being prepared for fossil chironomid datasets. To date, the outcome is similarly gloomy (I’m just not yet sure if the gloom is for my tests or the chironomid reconstructions).
Here is the abstract from Payne et al.
Transfer functions are valuable tools in palaeoecology, but their output may not always be meaningful. A recently-developed statistical test (‘randomTF’) offers the potential to distinguish among reconstructions which are more likely to be useful, and those less so. We applied this test to a large number of reconstructions of peatland water table depth based on testate amoebae. Contrary to our expectations, a substantial majority (25 of 30) of these reconstructions gave non-significant results (P > 0.05). The underlying reasons for this outcome are unclear. We found no significant correlation between randomTF P-value and transfer function performance, the properties of the training set and reconstruction, or measures of transfer function fit. These results give cause for concern but we believe it would be extremely premature to discount the results of non-significant reconstructions. We stress the need for more critical assessment of transfer function output, replication of results and ecologically-informed interpretation of palaeoecological data.
Payne, R.J. et al (2016) Significance testing testate amoeba water table reconstructions. Quaternary Science Reviews doi:10.1016/j.quascirev.2016.01.030