Who's Doing Bad Statistics?

Phil Plait, at Discover Magazine, objected to an article in the Wall Street Journal where a number of climate change skeptics claimed, among other things, that the last decade has seen no global warming. Plait asserted that this amounts to "blatantly misinterpreting long term trends, instead wearing blinders and only looking at year-to-year variations in temperature" and reproduced a shot from an animation in which the folks at skepticalscience.com demonstrate how there can be many periods in an upward trending time series over which the trend is not apparent.

William M. Briggs then accused Plait of doing "Bad Statistics" and in particular of ignoring "how the starting point made a big difference in the slope of the line, and how not accounting for uncertainty in the starting date translates into over-certainty in the results". Briggs also questioned the provenance of the data in the graph and objected to the lack of error bars on the data points, but I am in no position to comment on those aspects except to note that the data apparently come from the Berkeley Earth Surface Temperature  project lead by former skeptic Richard Muller (see this WSJ article by Muller)who was critical of the processes used in previous data compilations but now believes that the month-by-month global average land temperature anomalies (differences from some kind of long term multi-year average for the given month - presumably designed to eliminate the effect of seasonal variation throughout each year) that they have put together do have value as measurements with known margins of error (which appear to be small compared to the month-to-month variations)

What I want to focus on is just Briggs apparent claim that the data do justify an assertion that the increase (in whatever it may be that they represent) has stopped.

This is the issue addressed in the animation and blog post from which Plait got his version of the Berkeley data and what is remarkable is that Briggs own post about cherry picking intervals from a time series contains the germ of why his claim is false.

The fact is that a time series involving a substantial amount of random variation around a trend may often (and, if the scale of the random fluctuations exceeds  the trend increase for one step, will always) include intervals where the trend appears to be reversed and the shorter the interval the more extreme the variation and reversal of apparent trend.

What is most remarkable though is the fact that Briggs (who does apparently have an academic background in statistics) has followed up his attack on Plait with a series of five posts purporting to give an explanation of Time Series Analysis which is supposed to be in support of the claim but so far as I can tell is either vacuous or nonsense.

And further that having claimed in an article that “a civilian needs little or no maths to understand what ‘the probability that A is better than B is 80%’ means” he was either unable or unwilling to say when asked (in a comment to his blog post advertising it) what he meant by that other than the glib and vacuous response "It means the evidence is such that the probability 'A is better than B' is 80%."

I have not had quite such a sense of disconnect with a professional in the mathematical sciences before so I would like to check whether or not I am missing something


Leave a Reply