Wednesday, June 11, 2003
The Numbers Game

This month's print editions of IEEE Spectrum features an article by Jonathan G. Koomey of Lawrence Berkeley National Laboratory, author of Turning Numbers into Knowledge: Mastering the Art of Problem Solving. His nit to pick: The way bogus numbers get published, then refuted, but the bogus numbers continue to get published.

His issue: It's not the number of civilian casualties in Afghanistan in the war with the Taliban; it's four assertions about the volume of knowledge on the net and the use of electric power by computers and electronic devices.
True or false:
  • The amount of data flowing over the Internet is doubling every 100 days.
  • The Internet accounts for 8 percent of all electricity used in the United States.
  • Computers and information technology equipment, including Internet usage fromthe previous question, use 13 percent of all U. S. electricity -- and this total will grow to half of all electricity use in 10 years.
  • A wireless Palm VII uses as much electricity as a refrigerator when you include all the "behind-the-wall" networking, sever, and switching equipment.
All four of these assertions are wildly wrong. But if you answered "true" even once, you've got plenty of company. Most major U.S. newspapers, several well-known CEOs, respected institutions, and politicians of both political parties have cited these assertions in the past few years. The 8 percent Internet energy figure even came up in a Doonesbury comic strip a couple of years ago.
The column goes on to point to published refutations of each of the above items, out also to show the trail of each already-refuted claim through continuing articles and columns. It demonstrates how the bogus numbers lead to bad decisions: For example, the information volume numbers were used to influence the building of excess capacity for the networking infrastructure of the USA.

The author points out six simple things that can be done to check whether numbers make sense. There are more details about these at the web site for the Numbers Into Knowledge book.
  • Don't believe everything you read....
  • Do your homework.... Published documentation, cited sources, and back-of-the-envelope calculations should be sufficient to reproduce any number. If they aren't, be wary. [Emphasis mine.]
  • When in doubt, talk to experts in the field.... Your interest will make them happy, and you just might learn something significant.
  • Rely on peer-reviewed research.... Be skeptical of results that are not peer-reviewed....
  • Dig into the footnotes....
  • Follow the money.... [Or, as the history of the number of dead Afghan civilians in the Taliban war suggests: Follow the motivation, whether money, politics, religion, etc.]
Good stuff. Unfortunately, it's only available online to IEEE members, but the dead-tree version is carried by many university libraries and is available for purchase at the larger chain bookstores.

p.s. Koomey's study on energy usage by computers and the net is available here.