The presentation I gave at the seminar in honour of John Birks can be found here. I discussed how reconstructions from transfer functions can be evaluated, focusing on the significance tests of transfer function reconstructions that I developed in Telford and Birks (2011).
I show that significant results can arise because the variable of interest is an ecologically important determinant of variability in the fossil species data, or that it is correlated with one in the modern calibration set. This means that the significance tests do not circumvent Steve Juggins’ “sick science” problem.
Non-significant results can arise because the variability in the environmental variable of interest is small relative to the transfer function model uncertainty, other environmental variables are varying, or in some circumstances, because the tests have low power.
Whether the reconstruction is significant or not, low amplitude variability should be interpreted with caution as it may represent noise.
In questions, Cajo ter Braak suggested an alternative strategy for testing reconstruction significance. Initial tests (only two lines of code needed changing) suggest this alternative might be more powerful.
Exceedingly brief summaries of most of the other presentations can be found on twitter under the tag #HJBB2015.