*2.3. Statistical Analyses*

Descriptive statistics, including frequencies and percentages, are used to characterize categorical demographic subject characteristics (Table 1). Behavioral outcomes measured on a continuum are described using means and standard deviations, and compared by low zinc and iron level groups. Multivariate modeling of behavioral outcomes regressed on low zinc and iron group, with adjustment for important sociodemographic variables (family size, gender, house size, mother's education) is accomplished using general linear modeling. The father's education was not included in the analysis to avoid multicolinearity because the variable was found to be highly correlated with the mother's education. Models also accounted for clustering at the school level. Levine's tests are used to test for homogeneity of variance across all cells. In an attempt to identify multicollinearity, bivariate associations between independent variables are examined using chi-square tests. Statistical significance was taken at the two-sided *p* < 0.05 level. All the analyses were performed using STATA version 11.0 [45].
