Late last year, the debate over climate change was stirred up when an environmental economist, Ross McKitrick and a mining executive, Steven McIntyre, published a piece claiming to refute climatological research crucial to the claim that the last few decades have seen unparalleled global warming (the ‘hockey-stick’ paper of Mann, Bradley and Hughes). According to McKitrick and McIntyre, the work of Mann et al was riddled with errors, The paper was loudly publicised by the American Enterprise Institute (home of John Lott) and, as you would expect, Flack Central Station. Mann et al produced an immediate rebuttal, and despite many promises of a rejoinder, McKitrick and McIntyre have never responded on the substantive issues.
This would be par for the course, except that McKitrick somehow managed to attract the attention of Aussie computer scientist Tim Lambert, famous for his demolition of Lott’s shonky research, which purported to show that guns reduce crime. The result: McKitrick’s work is even shoddier than Lott’s.
Lambert has mainly focused not on the McKitrick and McIntyre paper but on a subsequent piece by McKitrick and Pat Michaels, which contains a regression purporting to show that it is GDP growth that causes (measured) climate change. McKitrick and Michaels take this as support for the generally-discredited ‘urban heat islands’ hypothesis, that measured warming is an artifact produced by weather stations in or near big cities.
In previous rounds of the debate, Lambert has shown that McKitrick messed up an analysis of the number of weather stations, showed he knew almost nothing about climate, flunked basic thermodynamics, couldn’t handle missing values correctly and invented his own temperature scale.
But Tim’s latest discovery really takes the cake. It’s well-known that the rate of warming varies with latitude, but McKitrick and Michaels find no such effect for their variable, which is the cosine of absolute latitude. Lambert checked and, amazingly enough, found that the data set used by McKitrick and Michaels had latitude in degrees, but the cosine function in the SHAZAM econometric package, they used expected input in radians (which is what any mathematically literate person would expect). If you apply this function to angles measured in degrees you get nonsense.
Once Lambert did the correct analysis, latitude was highly significant and the economic variables became much less important. The results reported by McKitrick and Michaels can be explained by an obvious confounding effect. Rich countries tend to be at high latitudes, and so GDP acts as a proxy for latitude.
Although Tim is almost invariably right in such matters, it was hard to believe that such a gross error could go undetected – it would show up immediately if you looked at descriptive stats on the variables, for example. So I checked myself. The descriptive statistics in the McKitrick and Michaels paper (available here) include the latitude, which is clearly in degrees, but not the cosine variable. The SHAZAM documentation, here, indicates that input to the sine function is in radians ( McKitrick and Michaels derive cosine using a transformation of this).
Bear in mind that McKitrick’s main claim to fame is his assertion to have done a painstakingly careful check of the work of others and to have found numerous errors. Looking at Lambert’s demolition of this paper, I’m reminded of what Julian Sanchez had to say about Lott – if he told you the time was 12 o’clock, you’d check your watch before you believed him.
And Michaels was a reputable climate scientist before he sold out to the fossil fuel lobby. It looks as though, as long as he says what his employers want to hear, they don’t feel the need for quality control.
fn1. There was a secondary dispute about the provision and labelling of data, as a result of which Mann et al published a very short corrigendum in Nature, noting that they had incorrectly described some parts of the data set, but that this had no implications for the results.
fn2. Interestingly, Sanchez, like Michaels, has worked at the Cato Institute, which shows that it’s not safe to make generalizations about institutions based on a few, or even a lott, of bad apples. Even TCS publishes some work by reputable people, to add cachet to its real output of lobby-fodder.