5 ms·

Dude, thanks for linking to Dr. Keith Devlin's column. Probably the only thing I read religiously in the MAA. BMI is just one of several mistakes attributed to

by dxbydt 14y ago

Dude, thanks for linking to Dr. Keith Devlin's column. Probably the only thing I read religiously in the MAA. BMI is just one of several mistakes attributed to the great mathematician Quetelet. You can credibly implicate Quetelet for a variety of modern day statistical crimes such as profiling ( attempting to figure out if a person is a criminal based on statistical analysis of race/sex/height/etc...),"normal" age of marriage, statistical rates of divorce in different countries and their implications on fidelity, and such other sociological nonsense. Dr. Devlin labels these things "junk math, numerological nonsense". But you can take a more charitable view - given that Quetelet practically invented both statistics & sociology ( which he called "physics of society" ), he came up with a number of formulae that seemed true at that time ( mid 1800s ). On hindsight much of this stuff is just silly mathematical manipulation ( Instead of dividing weight by square of height & multiplying by the constant 703, you can square the weight & divide by the cube root of height and put in some other constant...) If you are interested in the kind of statistics practiced in Quetelet's timeframe, here's some wacky references. http://en.wikisource.org/wiki/Popular_Science_Monthly/Volume_1/May_1872/Quetelet_on_the_Science_of_Man http://en.wikisource.org/wiki/Popular_Science_Monthly/Volume... http://statistiks.net/wp-content/plugins/downloads-manager/upload/v0101021.pdf http://statistiks.net/wp-content/plugins/downloads-manager/u... btw, much of what we do today in quant finance is suspiciously similar to what Quetelet suggests :) We build statistical "factor models" that credibly relate earnings of a company in california to the price of chicken in China :) We take a rigged number called the LIBOR & price interest rate derivatives on the spead between this rigged fixed rate & the floating rate. Then the public loses money & we scream - hey who the fuck rigged the LIBOR in the first place ?:) Talk about bootstrapping... honestly, we are no different from Dr. Quetelet. Might as well base Google's quarterly earnings on Page's BMI.

dredmorbius 14y ago
That's Dr. Dude to you. Quetelet's work was reasonably appropriate for his time. One source I recall (cannot find it now) stated that he was looking for a rapid paper-and-pencil method for estimating aggregate troop strength. As a basic, good-enough first approximation, BMI works reasonably well. The problem is that it's now been grossly misapplied. Quants (and economists) spend, I suspect, a lot of time chasing what turn out to be transitory relationships between various measurements. The relationships may very well be statistically valid, when first measured, but either changing real-world conditions, or, more likely in the financial world, adoption of the relationships by others seeking profits, rapidly erodes their validity. Thanks for the links, I've got some reading for later :) An enterprise I had some association with in the past spent some years and considerable capital trying to establish a strongly predictive relationship between a physical property and a health measurement. It turns out that there was a relationship, but not one which could be reliably used to determine an accurate quantification of the health measurement. This didn't keep them from torturing the data extensively. And Page's health is in the news today: http://www.washingtonpost.com/google-ceo-larry-pages-health-is-much-better/2011/04/07/gJQA6ohBiW_page.html http://www.washingtonpost.com/google-ceo-larry-pages-health-...

by dxbydt 14y ago

dredmorbius 14y ago

That's Dr. Dude to you. Quetelet's work was reasonably appropriate for his time. One source I recall (cannot find it now) stated that he was looking for a rapid paper-and-pencil method for estimating aggregate troop strength. As a basic, good-enough first approximation, BMI works reasonably well. The problem is that it's now been grossly misapplied. Quants (and economists) spend, I suspect, a lot of time chasing what turn out to be transitory relationships between various measurements. The relationships may very well be statistically valid, when first measured, but either changing real-world conditions, or, more likely in the financial world, adoption of the relationships by others seeking profits, rapidly erodes their validity. Thanks for the links, I've got some reading for later :) An enterprise I had some association with in the past spent some years and considerable capital trying to establish a strongly predictive relationship between a physical property and a health measurement. It turns out that there was a relationship, but not one which could be reliably used to determine an accurate quantification of the health measurement. This didn't keep them from torturing the data extensively. And Page's health is in the news today: http://www.washingtonpost.com/google-ceo-larry-pages-health-is-much-better/2011/04/07/gJQA6ohBiW_page.html http://www.washingtonpost.com/google-ceo-larry-pages-health-...