Climategate and three lessons of using models for persuasion
Blog: Bridgeland and Zahavi on Business Modeling
Last week, the Climatatic Research Unit—one of the world’s leading institutions devoted to the study of climate change—suffered a leak. Someone anonymously released thousands of previously secret CRU emails and documents to the internet. Since then, critics of the CRU—including global warming skeptics—have found embarrassing materials among the leaked files. Some emails appear to advocate changing measured temperature data to better fit global warming arguments. Other emails discuss pressuring academic journals to reject papers skeptical of global warming. Still other emails recommend deleting models and data files rather than providing them to a Freedom of Information Act request. The Daily Telegraph has (inevitably) dubbed this scandal Climategate.
What does Climategate mean for global warming? Is the scandal a “smoking gun” as some have suggesting, showing that man-made global warming is not occurring, or perhaps only occurring slowly? Or is it a “tempest in a teapot”, as others claim, certainly embarrassing for the scientists involved but ultimately providing no evidence against man-made global warming? Frankly I don’t understand the science deeply enough to have a useful opinion. But no matter what it means for the science and politics of climate change, Climategate offers lessons for us, for those of us who use models to persuade.
As we describe in our book, business models are used for eight purposes. One of those purposes is analysis, using a business model to gain insight, insight about customers, regulations, productivity improvements, or whatever. Another purpose is persuasion, using a business model to change someone’s mind. Models are used to persuade customers to buy, to persuade stakeholders to invest, and to persuade employees to commit.
CRU created models, models of climate of course, not models of business situations, but models nonetheless. They used these models for both analysis and persuasion. They used their climate models to analyze temperature data, as part of the scientific process, ultimately publishing academic papers based on their analysis. And they used their climate models for persuasion, to convince other scientists, politicians, the media, and the general public that the world is warming, and that fossil fuel emissions are causing that warming.
The CRU kept their models secret. They did not provide them to global warming critics in the scientific community, and in fact they did not provide them to supporters in that community. No one outside CRU had access to the models.
Secret models are fine for analysis. But secret models fail to persuade. Skeptics ask why they should believe the results of your model, if you won’t share it with them. How do I know you haven’t cheated in some way, created a model that simply supports what you want rather than one that is independent of your desires?
So the first lesson of Climategate is share your models. Make the models available to anyone who wants them. And in particular share your models with skeptics. The skeptics are the people who most need to be convinced, and they are the people who are the least likely to be convinced by secret models.
Among the leaked CRU documents were model files, software code that implements the CRU climate models. As outsiders slowly pore over the code, they have found many issues of professionalism: poor programming practices, apparent bugs, and comments that indicate that a later model developer does not understand the code written by an earlier one. These professionalism issues with the CRU models have provided skeptics with more reasons to be skeptical. If the CRU follows poor programming practices, how can we know that the results are accurate? If the CRU personnel who are responsible for the model code do not themselves understand it, how can we outsiders trust it?
The second lesson of Climategate is ensure that your models are professional. Models meant to persuade should be easy to understand, exhibit best practices, and of course be free of bugs. Reasonable people may disagree with the assumptions contained in the models, but those same reasonable people should not have cause to question the professionalism of the models themselves. Only professional models persuade.
The CRU relies on temperature data from many different sources all over the world, collected over many years. Some of the data was collected more than 100 years ago. Altogether the CRU assembled millions of individual temperature measurements. But this raw data was collected by hundreds of different organizations, with varying equipment and varying disciplines. Before the data could be analyzed, it had to be statistically adjusted and normalized, so it could be properly compared.
Since the leak, the adjustments themselves have become controversial. No one is denying the need to do statistical adjustments, but some climatologists have accused the CRU of always adjusting in ways that increase the amount of apparent global warming. They claim that the raw data shows much less warming than the adjusted data, that the global warming is an artifact of the statistical adjustment process not the climate.
There is no easy way to compare the adjustments that the CRU made to other potential adjustments that could have been made to normalize the data. The CRU made assumptions about the adjustments, and perhaps those adjustments were reasonable, or perhaps the assumptions themselves introduced bias. There is no easy way to know.
So the third lesson of Climategate is make it easy for others to change the assumptions. When I create models to persuade, I provide a user interface of assumptions, giving skeptics the ability to change the assumptions I made and understand the effect of each assumption on the outcome. Allowing skeptics to provide their own assumptions is critical to convincing them.
What if the CRU had learned and applied these three lessons? First, they would have given their models to everyone who wanted them. The models could not have been leaked if they were not secret in the first place. Second, the CRU models would have been professional, free of bugs and easy to understand. Doubts about the results of the models would have been averted. Third, the models would have included assumption user interfaces, allowing critics to understand how different assumptions would have changed the outcome.
I suspect that global warming would have been contentious no matter what actions the CRU had taken. The stakes are large—trillions of dollars and thousands of lives—too large to avoid controversy. But certainly if the Climatic Research Unit had learned these three lessons of using models for persuasion, they could have avoided much of the Climategate mess they are suffering today.