Several readers asked me to look at Ross McKitrick’s paper “HAC-Robust Measurement of the Duration of a Trendless Subsample in a Global Climate Time Series”, which is receiving the usual internet peer-reviewing (here, here, and here).
Before we begin, it is absolutely crucial that you understand the following point: both the IPCC (you know I mean the people and groups which contribute to it) and McKitrick have produced time series models.
Many people and groups have created time series models of the temperature, including the rank amateurs who attended the People’s Climate March. The latter model is, in essence, “The End Is Nigh”. This is simplistic, yes, and stupid certainly, but it is still a time series model.
Now we know, without error, that the IPCC’s time series model stinks. That it should not be trusted. That decisions should not be made based on its forecasts. That it is, somewhere, in error.
How do we know this? Because it has consistently and for many, many years said temperatures would be high when in reality they were low (relative to the predictions). People who refuse to see this are reality deniers.
Because the IPCC’s model said temperatures would be high these past eighteen or so years, when in reality the temperature bounced around but did nothing special, the IPCC has taken to calling reality a “pause” or “hiatus”. Everybody must understand that this “hiatus” is model-relative. It has nothing to do with reality. Reality doesn’t know squat about the IPCC’s model. The reality versus the model-relative “hiatus” is how we know the IPCC’s model stinks.
If the IPCC’s model did not stink, it would have predicted the reality we saw. It did not predict it. Therefore the model stinks. The debate really is over.
Now where the IPCC’s model goes wrong is a mystery. Could be it represents deep ocean circulation badly; could be that cloud parameterizations are poor. Could be a combination of things. It’s not our job to figure that out. The burden is solely on the IPCC to identify and fix what’s busted.
Enter McKitrick, who has his own model (or models; but for shorthand, I’ll speak of one). McKitrick’s model is a standard econometric model, which uses the Dickey-Fuller test (economists are always using the Dickey-Fuller test; I just like to say, “Dickey-Fuller test”; try it).
Is McKitrick’s model any good? There is no reason to think so. (Sorry, Ross.) It’s just a simplistic set of equations which is scarcely likely to capture the complexity of the atmosphere. If McKitrick’s model should be trusted, there is one test it could take to prove it. The same test the IPCC took—and failed.
McKitrick needs to use his creation to predict data he has never before seen. He hasn’t done that; and in fairness, he hasn’t had time. We need to wait a decade or so to see whether his model’s predictions have skill. But in a decade, I predict nobody will care.
The objection will be raised: but McKitrick’s model was built not to make predictions but to measure how long the “hiatus” was.
We needed a model for that? No, sir. We did not. We could just use our eyes. We need no model of any kind. We just take reality as she comes. To show you how easy it is to fool yourself with time series, here’s Figure 1 from McKitrick’s paper:
It shows “Globally-averaged HadCRUT4 surface temperature anomalies, January 1850 to April 2014. Dark line is lowess smoothed with bandwidth parameter = 0.09.” Let’s don’t argue about the dots, i.e. the temperature, a.k.a. reality, which really should have accompanying error bounds. Let’s just assume that the dots were the reality, full stop.
The black line is a chimera, a distraction, put there to fool the eye into believing the author has discovered some underlying “signal” in the reality. Well, he might have done. But if he has, he should be able pass the reality test mentioned above. Unfortunately, you can’t make forecasts with that kind of black line. The black line is not what happened! To say it is is to commit the Deadly Sin of Reification.
We must take reality as she is. All we need is a working definition of trend. Easy, right? No, sir. Not really. See this post. But skip all that and call a trend, “Over any ten year period, the temperature increased more than it decreased.” That’s one possible definition of trend.
Accepting that definition (but feel free to make up your own, using the post as a guide), there is no trend in the last two decades. But then there are many other periods sine 1850 without trends. So maybe bump up the time window to 20 years. Still no trend in the latter years.
And so on. No model is needed. None. We just look. There is no need for “statistical significance”, or any other pseudo-quantification.
Listen: make sure you get this. It doesn’t even matter if the IPCC or McKitrick perfectly predicted reality. We still do not need their models to see whether there was a trend. A trend only depends on (1) its definition, and (2) reality.
Update Ken below discovered this gem, which shows Richard Feynman destroying the IPCC’s global warming models.