All of a sudden there is some attention to the misuse of statistics in science, particularly in the medical and climate disciplines. William Briggs has just had his latest book published, (Uncertainty: The Soul of Modeling, Probability and Statistics, and a review is here. Another interesting blogpost by John Reid on the Blackjay blog focuses on the problem of autocorrelation and the specious regressions it can produce.
Autocorrelation and its calculation reminds me of geostatistics, an area I studied for my post grad dissertation. The core calculation performed in geostatistics is the variogram which is derived by calculating the differences in observations at some lag value. Lag 1 means that the difference between the first and second in a series is computed. Then the difference between the 3rd and second is computed, 4th and 3rd and so on. Lag 2 is similar except now we calculate the difference between the 1st and 3rd variable in the series, or in geology, sample location, then the difference between the 2nd and 4th etc etc to get a variation value for Lag 2. Lag 3 is computed by calculating the difference between the 1st and 4th values in the series, 2nd and 5th, and so on.
What the variogram ends up telling you (in a very broad and non-technical way) is that at a certain Lag value and above the variation becomes random or uncorrelated, while at lags less than the critical lag, the values are correlated, or ‘autocorrelated’. In geostatistics this is known as the ‘range’ of the variogram.
It seems a similar phenomemon has been documented in climate temperature data by John Reid, though Steve McIntyre and I quite independently and unknowingly came to the a similar conclusion when studying Mann’s Hockey Stick graph a decade ago; McIntyre and I worked in the mining industry and from our long experiences developed finely tuned BS radars sensitive to mining company IPO’s, leading to the question of ” how did they get that result”. The rest is history, as they say.
Then we have Pat Frank’s excellent presentation on similar data here. Again the phenomenon of autocorrelation is explained plus the identification of a few statistics 101 ideas that climate scientists seem not to fully understand.
Basically the computed global warming or climate change forecasts are specious and derived from an incomplete understanding of basic physics and statistics.