Monday, June 13, 2011

The worst in statistical reasoning, from Climategate scientist Phil Jones

According to the BBC News, Phil Jones declared global warming since 1995 statistically insignificant last year, and statistically significant this year. Why? Last year, the probability that “ordinary” variation of global temperatures yielded warming as extreme as that observed was .1, and this year it is .05.

I clearly recall the day my statistics professor railed against categorical response to the 95% confidence level as “statistically significant.” It is hard to believe that Jones, who is younger than I, did not hear something similar in school. Suppose that when next year’s data are processed, the confidence level goes down to 94%. Will Jones revert to saying that the observed warming is not statistically significant?

It is impossible to decide absolutely whether human activity is causing global warming. Scientists in general, and climate scientists in particular, should explain that their research leads to degrees of belief.

Furthermore, what policymakers need is Bayesian processing of climate data, not the Fisherian silliness of Phil Jones. That is, the possibility of anthropogenic global warming calls for sophisticated risk management rather than a true-false decision. We are gambling on the future of the world, and the only reasonable response, given present information, is to hedge our bets.

[Thanks to jivlain for his comment. I'm presently having to use an outdated browser, and cannot enter a comment of my own. So I'll respond here. From the infamous BBC Q&A with Jones in February 2010:
B - Do you agree that from 1995 to the present there has been no statistically-significant global warming

Yes, but only just. I also calculated the trend for the period 1995 to 2009. This trend (0.12C per decade) is positive, but not significant at the 95% significance level. The positive trend is quite close to the significance level. Achieving statistical significance in scientific terms is much more likely for longer periods, and much less likely for shorter periods.
Jones should have known to respond that the trend was statistically significant at the p% confidence level.]

[Bob O'H, it's great to see a statistician call for such a straightforward interpretation. But you're apparently looking at different numbers than Jones is:
C - Do you agree that from January 2002 to the present there has been statistically significant global cooling?

No. This period is even shorter than 1995-2009. The trend this time is negative (-0.12C per decade), but this trend is not statistically significant.
I'll hazard a guess that your data have been adjusted to take radiative forcing into account — as they should be. Total solar irradiance (TSI, plotted here) contributes substantially to global temperatures, and cycles up and down with a quasi-period of about 11 years. The most recent downswing in TSI began in 2002, was more pronounced than usual, and lasted longer than usual (until 2009). The BBC question is utterly ill-posed, because it calls for analysis of cherry-picked data.]


  1. The origin of this is that last year, he answered a bunch of questions put to him by "sceptics", one of which concerned the direct question "has there been any statistically significant warming in the last 14 years". He answered this accurately, with a large addendum to the effect that this was a silly question.

    Obviously, the "sceptics" then dropped the addendum, and it magically turned into "there has been no global warming in the last 14 years". This is an update on that story.

    But nonetheless, I completely agree. In my opinion (and with the benefit of hindsight), he should have answered the silly question with a Bayesian response. Yes, the crazies would have ranted about how he was avoiding the question, but it was a dishonest question and there was no way of avoiding them ranting about something.

  2. Even without Bayesian stuff, it's dumb. The data show an increase. All of the data. There is no uncertainty, except in the estimates for each year. So you can just fit the line, say "it's increasing" and be done.