2.8. Analysis:

Variables of interest were described according to their mean and standard deviation for normal distributions or median and interquartile range for non-normal distributions. For the WNV assessment, raw test scores were age adjusted by converting them to a t-score using the standard scoring manual[19].. t-scores on the two sub-tests were combined for a composite WNV test score.

The sample was separated into “high performers,” (HP) who had attended school and had never been held back a grade, and “low performers,” (LP) who had either dropped out of school or been held back from at least one grade at least once. Logistic regression models were constructed with school performance as the primary outcome variable. Directed acyclic graphs were drawn to select potential confounders in each model. Four models were analyzed: one for each primary explanatory variable of interest (lifetime seizure estimate, WNV score, stigma score, and familial wealth status). A p-value of 0.05/4=0.0125 was used to correct for multiple comparisons in the method of Bonferroni. All statistical calculations were conducted using the programming language R (Vienna, Austria).