Methods
Data
The data used in the current study are from the National Crime Victimization Survey (NCVS) School Crime Supplement (SCS). Every year, NCVS interviews each household member who is aged 12 years and older. SCS, as a supplement to the annual NCVS, started collecting data in 1989, then again in 1995. Beginning 1999, NCVS-SCS has been collected every two years. SCS interviews each household member aged 12 to 18 who attends a primary or secondary education program (Bureau of Justice Statistics, 2015). The NCVS-SCS dataset first introduced questions regarding cyberbullying in 2011. Starting in 2015, the NCVSSCS removed eight cyberbullying-related questions and included one question asking the place of bullying to capture whether the bullying occurred online or by text. Thus, we used the 2011 and 2013 data only for the current study purpose. To include students who experienced cyberbullying, the current study included respondents who indicated experiencing cyberbullying using the cyberbullying-related questions. The questions include posting hurtful information about the victim, purposely sharing private information, photos, or videos on the internet or mobile phones, threatened or insulted victims through email, instant messaging, chat, text message, online gaming, or excluding victims from online communications. We also cross-referenced victims using one question asking the frequency of all these questions. This procedure classified 498 cyberbullying victims from 2011 and 325 cyberbullying victims from 2013 (N=823).
Dependent Variables
Social harm. In the current study, social harm was measured through students’ answers on whether or not they were staying away from seven different places in school (e.g., school entrance, hallways and stairs, cafeteria, restrooms, parking lot, other school building, or other school ground) or route to school, as well as avoiding any online activities. Dichotomized response options for the nine items were summed across the items to reflect the severity of social harm as a count variable. An index reliability of Cronbach’s alpha is 0.78, and the higher numbers represent the higher severity of social harm.
Independent Variables
Adult support. Adult support was measured with six 5-point Likert scale questions that asked the support from adult figures in school, including teachers who cared, noticed, listened, told positive stuff, wished the best, and believed the students (e.g., “There is an adult at school who believes that you will be …” or “Teachers care about students.”; Strongly Negative = 0, Negative = 1, Nuetral = 2, Positive=3 Strongly Positive = 4). All Likert scale questions were summed across six items then divdied by six to reflect the severity of adult supports from strongly negative (= 0) to strongly positive (= 4). An index reliability of Cronbach’s alpha is 0.83, and the higher number represents more support from adult figures in school.
Peer support. Peer support was measured with one 4-point Likert scale question that asked the presence of friends whom the respondent can talk to, cares about feelings, and what happened to the students (i.e., “Would you agree, at school, you have a friend you can talk to, who cares about your feelings and what happens to you.”; Strongly Disagree = 0, Disagree = 1, Agree = 2, Strongly Agree = 3).
Control Variables
School experience and safety features. To capture the impact of negative experiences in school, the current study utilized victimization report questions regarding hate-related words on race, religion, ethnicity, disability, gender, and sexual orientation (e.g., “Were any of the hate-related words related to you race?”). Those who reported any one of the above victimizations were coded as 1 and 0 otherwise. The physical atmosphere of the school can also influence students’ social and psychological harm. Thus, the study utilized a dichotomized question regarding the presence of hate symbols in school (0 = No, 1 = Yes). In a similar vein, the study also included the nine dichotomized safety features at school (e.g., the presence of security guards or assigned police officers, metal detectors, visitor sign-in process, security cameras, and the code of student conduct) to represent none (0) to high (9) physical safety levels.
Demographic features. The current study included three demographic variables as covariates: gender (0 =Male, 1 = Female), age (12 – 18), and ethnicity (0= Others, 1= Caucasian).
Analytical Strategies
The two dependent variables for the current study, social and psychological harm, are the count variables. Although the Linear Regression Model (LRM) has often been applied to count outcomes, this can result in inefficient, inconsistent, and biased estimates. Even though there are situations in which the LRM provides reliable results, it is much safer to use models designed explicitly for count variable outcomes.
By utilizing count models, the current study assumed that every bullying victim has a positive probability of experiencing any given level of social and/or psychological harm. Depends on individual characteristics, the probability of being a victim may differ across victims, but all victims have some probability of experiencing harm. . To run count models, the study began with the Poisson Regression Model (PRM). Both social harm and psychological harm variable, however, contained strong evidence of over-dispersion (Social harm: G²= 1275.79, p = 0.001; X²= 2121.42 p = 0.001 and Psychological harm: G²= 959.02, p = 0.001; X²= 1555.81, p = 0.001). Thus, instead of applying the PRM, we ran the Negative Binomial Regression Model (NBRM) for both models. The NBRM improves upon the underprediction of zeros in the PRM by increasing the conditional variance without changing the conditional mean. The NBRM allows examining the over-dispersed portion of the count variable, which indicates that the variance exceeds the mean, and the distribution of outcomes is determined by both random and non-random (i.e., risk heterogeneity and/or event dependence) processes (Park & Eck, 2013; Winkelmann, 2008). The physical atmosphere of the school can also influence students’ social and psychological harm.
Table of Contents
- Introduction
- Literature Review
- Methods
- Results
- Discussions
- Conclusion
- Declaration of Interest Statement
- References