III. METHODOLOGY
Sample
This study examined electronic government-based second-hand data gathered from the Youth Risk Behavior Surveillance System (YRBSS) in 2011, 2013, 2015, and 2017 (Eaton et al., 2012; Kann et al., 2014; Kann et al., 2016; Kann et al., 2018). The data were collected by the Centers for Disease Control and Prevention, which is devoted to the public’s safety and health. A threestage cluster sample design produced a nationally representative sample of American students in grades 9–12 who attended public and private schools. The standard questionnaire in 2011 and 2013 included 86 questions; the standard questionnaire in 2015 and 2017 included 89 questions.
Statistical Analysis Because the observations for the four questionnaires used in 2011, 2013, 2015, and 2017 were from the same states, a certain amount of correlation/dependence was expected (Su, 2020). Indeed, a prior study that used data from the same surveys attempted to use Poisson regression, a parametric statistic, but there was a very large overdisperson problem, likely due to a lack of independence among the data values (Davis, 2020; Su, 2020). In the Davis (2020) study, the Poisson distribution assumed that the mean and variance were equal (i.e., a ratio of one). However, the variance was actually 5,036 times greater than the mean, which is a very large overdispersion problem. Thus, in order to address this overdispersion problem, generalized estimating equations (GEE), a nonparametric statistic, was used in the current study to assess the relationship between the number of females who were cyberbullied and the number of females who seriously considered suicide. Although GEE avoids the distributional assumptions of independent observations, the use of a nonparametric statistic would usually result in some loss of efficiency for estimation of the coefficients relative to the optimal likelihood-based estimates when distributional assumptions are satisfied (Fitzmaurice, Laird, & Ware, 2004).