Ross McKitrick writes:
I give a demonstration of why the Parker and BEST analyses don’t disprove the evidence of contamination of temperature data, and outline what it would likely take to settle the issue properly.
ENCOMPASSING TESTS OF SOCIOECONOMIC SIGNALS IN SURFACE CLIMATE DATA: I have a new paper out in Climatic Change on the question of whether surface climate data are biased by non-climatic factors relating to socioeconomic development:
Rather than try to settle the debate once and for all, I focus on why the various attempts to show the data are not contaminated do not disprove the results showing that they are. The problem has been that authors use non-overlapping data sets and different methods, and end up talking past each other. The way to settle matters, I argue, is to adopt an encompassing framework in which both types of results can be demonstrated on the same data set, where one arises in a restricted subset of another model, and the restrictions can formally be tested.
I give two examples, one replicating a Parker-style equivalence between nighttime minimum trends in calm and windy conditions, then showing that this persists in a temperature data set that can be shown to be correlated with population growth. I also replicate the BEST-type results that rural trends are slightly greater than those of urban areas, and show that this result appears in a restricted subset of a larger model in which socioeconomic growth is significantly correlated with temperature trends. In both cases the restrictions necessary to yield the model that supposedly shows no data contamination are rejected. Data/code archive here.
Posted at http://www.rossmckitrick.com/.
McKitrick, Ross R. (2013) Encompassing Tests of Socioeconomic Signals in Surface Climate Data
Climatic Change doi 10.1007/s10584-013-0793-5
The debate over whether urbanization and related socioeconomic developments affect large-scale surface climate trends is stalemated with incommensurable arguments. Each side can appeal to supporting evidence based on statistical models that do not overlap, yielding inferences that merely conflict but do not refute one another. I argue that such debates are only be resolved in an encompassing framework, in which both types of results can be demonstrated as restricted forms of the same statistical model, and the restrictions can be tested. The issues under debate make such data sets challenging to construct, but I give two illustrative examples. First, insignificant differences in warming trends in urban temperature data during windy and calm conditions are shown in a restricted model whose general form shows temperature data to be strongly affected by local population growth. Second, an apparent equivalence between trends in a data set stratified by a static measure of urbanization is shown to be a restricted finding in a model whose general form indicates significant influence of local socioeconomic development on temperatures.