George Box (1919–2013) is the man behind the most quoted sentence in all of
statistics: "All models are wrong, but some are useful." It sounds like a shrug; it is
actually a whole philosophy. A jovial, deeply practical Englishman, Box spent his life building
models he cheerfully admitted were false — and getting enormous mileage out of them anyway,
including the
Box never set out to be a statistician. He trained as a chemist, and during the Second World War he was assigned to run experiments on defences against poison gas. The results were so noisy he couldn't make sense of them, so he asked for a statistician — and was told there wasn't one. So he taught himself, working straight through R. A. Fisher's books by candlelight (more or less) until he could analyse his own data. He turned out to be brilliant at it. Years later, in a plot twist worthy of a novel, he married Fisher's daughter, Joan.
With the Welsh engineer Gwilym Jenkins, Box wrote the 1970 book that gave time series
forecasting its modern shape. Their
Box's toolbox is enormous: the Box–Cox transformation for taming skewed data, response surface methodology for tuning industrial processes, and evolutionary operation for improving a factory while it keeps running. But he may be best remembered by his students for something warmer. At the University of Wisconsin he ran legendary Monday-night sessions — "beer and statistics" — where researchers gathered at his home to thrash out real problems over a drink. It captured the man perfectly: statistics wasn't a temple of theorems for him, it was a friendly, slightly rowdy conversation about how to get useful answers out of messy reality.
The fuller story is on Wikipedia: George E. P. Box — Wikipedia.