Is Attribution Science Really Science?

 

A recent issue of Physics Today described a new field called "attribution science" which endeavors to attribute certain extreme weather events to human-caused climate change.  I'm not going to get into the specifics of the particular examples it cites, because they are many and varied.  But I will note that this kind of activity is becoming quite popular, not only in the physics community, but in geosciences as well.  I have read several papers in which the authors make statements like "This Japanese heat wave is virtually certain to have been caused by human-induced climate change."

 

What I would like to examine is the philosophical bona fides of this type of activity.  That is, what is the logical chain, assuming there is one, from the starting premises of such a statement's argument that leads to such disturbing and definite conclusions?

 

To say A is caused by B with some degree of confidence, we can do one of a number of things.  One way to verify such a statement is to take B away and see if A still happens.  Unfortunately, we can't just take another identical Earth, remove most of the carbon dioxide from the atmosphere, and then see how it runs after that.  The experiment we are running with this planet is unique, as far as we know, and so our ability to fiddle with the variables in different experimental runs is nil. 

 

Another way to verify causality is to show that when A happens, B happens shortly afterwards and not when A doesn't happen.  The Physics Today article calls this kind of causality "Granger causality" and says there are statistical methods to predict B happening based on statistics concerning A and B, whatever they might be.  But, as the article points out, this kind of test is subject to the post hoc, propter hoc fallacy (meaning roughly "afterwards, therefore because of").  That is, just because the New York City blackout of 1965 happened right after a kid hit a light pole with a stick doesn't mean that the kid caused the blackout—despite what the terror-stricken kid confessed to his mother when he got home.

 

What the attributionists typically do is to gin up some atmospheric models that produce probabilities of this or that event that they are trying to blame on people.  Then they tinker with the atmosphere's CO₂ levels, or concentration of aerosols, or something that is pretty clearly due to human activity.  And then they run their models again to see if they get the same disastrous weather that happened before, or whether leaving the human activity out makes it less likely.

 

The logical problem with this game is that no climate model I know of includes absolutely everything that affects the climate.  They all approximate or ignore certain factors.  So anything claimed for these models can be answered by asking, "Sez who?  And how do you know that something you ignored wouldn't give you different results than you got?"  Logically speaking, there is no defensible reply to that question.

 

I have done some very simple modeling of physical systems, and let me tell you:  it's very easy to get almost anything you want, and the more variables there are, the easier it is to do that.  And I'm not necessarily accusing the parties involved of fraud.  They may be honestly trying to make their models work better, not make them produce the results they want.  But the dividing line between "work better" and "get the results we want" is a thin and permeable one. 

 

After reading several such articles, I start to get the feeling that all pretense of what used to be called objective science has been abandoned.  Everybody knows that if you write a paper using a model that shows Hurricane X was conclusively not caused by HICC, the chances of getting it published are small, to say the least.  All the media, all the funding, and all the scientists, nearly, are predisposed to hear the bad news that we humans are the problem that causes every single bad thing that happens weather-wise. 

 

Was it Mark Twain who said there are lies, damned lies, and statistics?  If so, he was writing at a time when the discipline of statistics was just beginning to be recognized as a new and powerful way of understanding the world.  And don't get me wrong—statistical study is a vitally necessary part of doing big-data science.  It can't be done any other way. 

 

But when sophisticated mathematics, not comprehensible to anyone except a few experts, is brought to bear on the question of whether HICC "caused" a particular adverse weather event—a flood, a fire, a drought, a tornado, a hurricane, you name it—the average layperson is placed at an unfair disadvantage.  If he denies that this is the case, then the scientist will say the layman doesn't know what he's talking about.  And in a sense, that's true.  But even laypeople can understand that the kind of certainty we can have about the sun coming up tomorrow morning is not obtainable when it's a question of complicated and incomplete climate models and highly localized and specific weather events.

 

The Physics Today article said as much, if one bothered to read between the lines, or just the lines themselves.  They addressed several types of attribution science and showed there were serious flaws in each one, although not so severe as to render the enterprise completely worthless.  But I wonder.

 

 Everybody agrees that even if we shut off all fossil-fuel burning tomorrow, we would be stuck with whatever effects our present CO₂ levels have for the better part of a century.  It seems to me that scientists would be better employed by looking for ways to mitigate the climate changes we are going to have, rather than disrupting or destroying the world's economy in a fruitless attempt to shut the barn door after the CO₂ horse has escaped.

 

Sources:  The September 2023 issue of Physics Today carried the article "Connecting Extreme Weather Events to Climate Change" by Michael Wehner on pp. 40-46.  I also referred to the Wikipedia article on Granger causality.