The dependent variables (metal
concentration in lg L1
) were each regressed against potential predictors
(independent variables) such as life style habits and sociodemographic
characteristics. As recommended by Barr et al.
(2005), urine creatinine levels were included as an independent
covariate in the regression models rather than using creatinineadjusted
metal concentrations (creatinine standardization)
because this approach is less prone to produce biased effect
estimates.