Previous
This paper shows that there is a wide variety of statistical tests which can be used to analyse cross-sectional data as illustrated by adult anthropometric variables, particularly BMI. The appropriate statistical test to use depends very much on the aims and objectives of the particular study.
Researchers should test whether or not BMI is significantly skewed, and if so, are encouraged to use a log transformation to normalize the data. Geometric rather than arithmetic means should also be used.
For prediction researchers are encouraged to move from the use of sensitivity and specificity to logistic regression analysis. For bivariate analyses multiple regression analysis seems the most obvious test to use.
Armitage P & Berry G (1987): Statistical methods in medical research. Oxford: Blackwell.
Ferro-Luzzi A, Sette S. Franklin M & James WPT (1992): A simplified approach of assessing adult chronic energy expenditure. Eur J. Clin. Nutr 46, 173-186.
Lemeshow S. Hosmer DW, Klar J & Lwanga SK (1990): Adequacy of sample size in health studies. Chichester: World Health Organization/Wiley.
Mascie-Taylor CGN (1993): Research designs and sampling strategies. In Research strategies in human biology, eds GW Lasker & CGN Mascie-Taylor. Cambridge: Cambridge University Press.
Shetty PS & James WPT (1994): Body mass index: a measure of chronic energy expenditure. Rome: FAO.
Youden WJ (1950): Index for rating diagnostic tests. Cancer 3, 32-35.
Ferro-Luzzi: Why should we use BMI, weight and height separately when they co-vary?
Mascie-Taylor: I was showing in one data set how weight alone was a better predictor than height; weight alone might be used instead of BMI.
Ferro-Luzzi: Would that hold for cross-country comparisons?
Mascie-Taylor: I think we need to do meta-analyses across countries to answer these questions. The empirical approach is always the best.
Norgan: You express doubt about the use of sensitivity and specificity analyses. Why do you think logistic regressions are not country specific?
Mascie-Taylor: Because they measure an odds ratio which is independent of country. You can bring in more terms to derive an odds ratio.
Naidu: Logistic regression gives good results, but we may be better with odds ratio rather than an R2 term.
Mascie-Taylor: Odds ratio only gives risk, but R2 gives probability of death at each point on the line.
Scrimshaw: In this population where you found weight was as good as BMI, I suspect it was one of these populations where there is very little variation in the height of women. Isn't this a misleading measure to take, when in other populations and between populations there is a greater range of height?
Mascie-Taylor: In a cross-sectional analysis, BMI is better, but in longitudinal analysis, when height isn't going to change very much, there seems little point in doing BMI analyses rather than weight alone.
James: In the National Academy of Sciences analysis of pregnancy, BMI and birth weight [National Academy of Sciences, Institute of Medicine et al. (1990) Nutrition during pregnancy. Washington, DC: National Academy Press] the two extremes of height are excluded. In the classic Aberdeen studies of Thomson [Thomson AM & Billewicz WZ (1963): Nutritional status, maternal physique and reproductive efficiency. Proc. Nutr. Soc. 22, 55-63], there was a component of height with an intergenerational effect which operated not through the effect of size on maternal weight but perhaps through pelvic size, so we ought to be a little cautious.
Waterlow: If you have a series of factors and a series of odds ratios, as a policy-maker, do you focus on the factor that gives the highest odds ratio?
Mascie-Taylor: You need to look at each factor alone and then build up a sequential story. The odds ratio tells us the risk of any particular factor being relevant, but we also need to take into account the variety of factors and the dominance of these in any particular society.
Funding for this l/D/E/C/G workshop was provided by the United Nations University (UNU), the Food and Agriculture Organization of the United Nations (FAO) and Canada's International Development Research Centre (IDRC). The publication of these proceedings was made possible by a grant from the Nestle Foundation.
This publication is available free of charge from the Secretariat of l/D/E/C/G c/o Nestle Foundation P.O. Box 581 1001 Lausanne Switzerland