Understanding the bullying structure on a complex national data set and sample size requirement for non-normal complex multilevel data for the multilevel structural equation model
Powell, Adeya Shontelle
MetadataShow full item record
This study consisted of two parts. The first study sought to explore the relationship between bullying and depression/suicide using the most frequently used survey given to American youths, the Youth Risk Behavior Survey (YRBS), and to employ one of the least used statistical tools for this survey, Structural Equation Modeling. Using a mediated structural model with two possible mediators, school violence and teen alcohol abuse and use, school violence was shown to mediate the relationship between the dichotomous bullying variable and the depression/suicidality factor, while teen alcohol abuse and use was not a mediator. The second part of the study analyzed sample size requirements for a multilevel structural equation model when non-normality was present. This study had two subparts: non-normal continuous and non-normal non-continuous/categorical sample size requirement. Robust Maximum Likelihood Estimator (MLR) was the only estimator to perform well at both the between and within level on non-normal continuous data. Weighted Least Square Estimators did not perform well on categorical data when the sample size was 100 or less.