Commenter rhhardin over at Tim Blair's blog adds his two bits to a thread on the fanciful Lancet data on excess deaths in Iraq (comment No. 18)
It's a great metaphor for sampling error, and a handy generic put-down for dodgy sampling-based statistical research. I will use it often.
Sampling is used to find out what percentage of a cat is teeth. You poke a cat at random with a pin, and record the number of times you hit teeth, and the number of times you hit something else. Most people find it’s about 100% teeth, after a few samples.
This is then reported in journals.