More on BMI, etc..

[Guest guest]

Hi folks:

Letter to the editor of the London Times:

" Sir, Hutt and Dr N. P. Hudd (letters, May 15 and May16)

argue the relative " scientific " merits of body mass indices defined

as either the ratio of weight to height squared or height cubed.

The former, known as the Quetelet index, and the latter, known as the

Khosla-Lowe index, are both special cases of the Benn index, which is

defined as the ratio of weight to height raised to the power p. Since

the use of such an index is to provide an accurate indicator of the

percentage body fat of an individual, which is much harder to measure

accurately, the usual statistical approach to deciding which index

is " best " is to assess which is the most highly correlated with body

fat in a sample of individuals for which per cent body fat, height

and weight measurements are available.

An arguably better approach is to use regression techniques to

estimate the " optimal " value of p. Using a large data set kindly

provided by the New York Obesity Research Centre, I have done just

this for individuals categorised by gender and race (Caucasian,

Hispanic, African-American and Asian). Controlling for age, the

optimal estimate of p for males is around 1.5, except for Caucasians,

for which it is nearer to 1. For females, the optimal estimate of p

is around 1 for all races: ie, for this data set, the weight-height

ratio (p=1) is the most appropriate index to use for females, while

for males both the Quetelet (p=2) and Khosla-Lowe (p=3) indices are

inferior to indices that use a p value of between 1 and 1.5.

These results certainly show that the traditional body mass index

defined using height squared should not be used uncritically for all

individuals irrespective of gender, race and age.

TERENCE C. MILLS

Professor of Applied Statistics and Econometrics

Department of Economics

Loughborough University "

Rodney.