Obesity in children and adolescents: development of predictive equations and their application to epidemiologic studies.

Epidemiological studies of body fatness cannot be based on laboratory measures. One alternative is to analyze values related to body composition such as skinfold thicknesses, but it is better to apply an equation that accurately predicts a body composition variable. If an equation is to be developed, the choice of the dependent variable affects the selection of the independent variables to be tested for inclusion. The validation group should have a wide distribution of values for the dependent and independent variables. To develop an equation that works well on validation and with other groups, all possible sub-set regression should be applied, multicollinearity examined, and ridge and robust regression used if these are needed. An equation chosen from the literature should have been developed and tested by appropriate methods and it should be validated against data from a sub-group of the study sample.