Last year, I presented a paper at the AGU Fall Meeting (December 2004) which was the core of our GRL paper. I’ve just submitted an abstract for the 2005 Fall Meeting for session PP-19 co-chaired by Hans von Storch.
Last year’s abstract is here.
The session is entitled Climate Change in the Recent Past: Integrating Meteorological, Proxy, Borehole, and Modeled Climate Reconstructions and is described here.
Here is the abstract that I submitted. Suggestions are welcome as I can edit it a little for a week or so.
Methodological Issues in Multiproxy Reconstructions
We review some important and under-discussed methodological issues in multiproxy reconstructions, applying insights from recent statistical and econometric studies and considering examples in the recent multiproxy literature. Issues include:
1) “spurious” (in a statistical sense) relationships between highly autocorrelated series, considering both regression and scaling methodologies. We discuss problems arising from proxies potentially affected by nonclimatic trends or factors and the related question of the robustness of multiproxy methods.
2) series selection and the potential problem of data mining. We show examples where variance attenuation arises without regression. The effect can be demonstrated when series are selected from a menu of persistent red noise examples and then averaged together.
3) non-linear and especially non-monotonic relationships with temperature.
Each issue will be related to the problem of calculating confidence intervals.