Friday, September 21, 2007

Forecasting Global Warming

 Scott Armstrong is a professor of marketing at Wharton who I remember as the geeky Decision Sciences instructor. He's an expert on forecasting, an important business discipline for market research of course, but also relevant for evaluating any long-term predictions of the future. His team runs a web site on forecasting principles, and publishes a book which is the standard textbook on the systematic study of decision-making about the future. His team has studied more than 82,000 forecasts to see which techniques work and which don't and one of their conclusions is that, without rigorous attention to some key details, "expert" long-range forecasts are almost always wrong.

This week he published a draft of a new report where he analyzes the IPCC's report on climate change (you know, the big UN-sponsored one that got all the attention earlier this year). Using his forecasting principles he concludes that the IPCC methodology is riddled with classic errors that essentially render it useless. Here are some biggies:

  1. Forecasts made by experts who collaborate on their projections are weakly related to accuracy. The IPCC would have had better results if the various contributors reached their conclusions independently.
  2. Even given modest uncertainty in models, the degree of long-range variability among predictions is so huge that you're unlikely to get valuable policy recommendations.
  3. Models that "fit" to historical data, like much of the complex math behind the IPCC models, rarely produce accurate results without rigorous attention to details. All it takes it one or two unexpected events (volcanoes? El Nino?) and the entire projection goes out the window. He offers ways to control for this, but notes that IPCC didn't do that and probably would have had different results if they had.

Ever notice how discussions of global warming rarely (if ever) offer serious policy suggestions to fix the problem? Most recommendations I've seen on how to "save the planet" are worth doing for other short-term and very practical reasons: lowering your carbon footprint will lower your electric bill, cutting America's "addiction to oil" lessens our dependence on Middle East suppliers, a Prius is a fun high-tech car anyway, etc.  Nothing about that advice changes, regardless of the accuracy of these long-term (and probably inaccurate) projections. 

But don't be fooled when you see a company or politician use Global Warming as marketing spin to make it look like they have your long-term interests at heart.  If Scott Armstrong is right, the long-range forecasts are meaningless and may in fact be serving the short-term interests of people who just want you to buy their stuff or vote them into office.

No comments: