Correlates of these groupings during childhood were analyzed using the full dataset of 6674 observations age 9-16 observations in which each individual contributed up to 7 observations. Adult outcomes were analyzed in a dataset of 2132 observations collected at ages 19 and 21 years, with each individual contributing up to two observations. The survey models of Stata 10.1 were used for these analyses. These models account for the non-independence of observations from the same participant by identifying the individual as the primary sampling unit and allow inclusion of the GSMS sampling weights (StataCorp, 2007). Robust variance estimation is used to produce unbiased parameter estimates and appropriate standard errors generalizable to the original population. The majority of analyses used logistic regression to predict binary dependent variables. Ordinal logistic regression models were used for analysis of symptom counts with skewed distributions.
