Study design and participants
This is a cross-sectional study of 7th to 9fh graders (approximately 13–15 years old) from public junior high schools in Taiwan, 2010. A multistage, random cluster sampling procedure was used. First, five schools were systematically sampled from one of the four geographical areas in Taiwan (North, Central, South, and East). Second, within each of the 20 selected schools, two classes from each grade were randomly selected. A total of 3721 students from 120 classes were invited to participate. Of these, 3441 students completed a self-report questionnaire in their classroom supervised by 22 trained college students, giving a completion rate of 92.50 %.
As those experiencing bystander reactions must have been a bystander in a bullying episode, we excluded all participants who reported that they had not witnessed any of the 13 types of bullying episodes in the bystander experience scale. We excluded 453 students (13.23 %) who reported that they had never observed or noticed any kind of bullying and 116 participants with incomplete data. The final analytic sample consists of 2872 students. There were no statistically significant differences between participants with complete (n = 2872) or incomplete (n = 116) data with regards to bullying experiences, defending behaviors, social anxiety score, depressive symptoms score and other related variables. There was, however, a significant gender difference as boys were more likely to have incomplete questionnaires than girls.
The questionnaire contained three sections: personal information, family background and bulling experiences. The section on bullying experiences included perpetration, victimization, and observation, and was developed based on the literature [1, 6, 8, 28]. The instrument was modified according to the results of content analysis of 10 focus group discussion sessions (5 groups of 6 boys each and 5 groups of 6 girls each) conducted in five junior high schools. The instrument was validated by five experts specialized in adolescent health and development, and pretested in a pilot study of 715 students.
Bullying roles: participant roles in the bullying process
A description of bullying was provided to students to ensure consistency: “A student is being bullied when another student, or a group of students, say or do nasty and unpleasant things to him/her. It is also bullying when a student is teased or socially isolated in a way he or she doesn’t like. But it is not bulling when two students of about the same strength quarrel or fight.” [1, 6, 28]. The following single question was used to measure bullying experiences (bullying others or being bullied) both in school and away from school: “Have you ever encountered the following situations in the past year?” Response categories were “never”, “seldom: less than once a month”, “sometimes: more than once a month but less than once a week”, “often: once or twice a week” and “always: more than three times a week” (Appendix 1). In this study, we combined the three responses, sometimes, often and always, into one category which represented those who had had experiences of bullying [6, 28]. The other two responses were combined into one category, which represented no bullying experiences. Bullying experiences (in or outside of school) were then categorized into four groups according to bullying roles: (1) bullies, students whose bullying experiences involved them bullying others; (2) victims, students whose bullying experiences involved them being bullied; (3) bully/victims, students whose bullying experiences involved both bullying others and being bullied; (4) bystanders, students who were not directly involved in bullying experiences but were observers of bullying episodes as measured by the bystander experience scale. We defined a bystander as someone who was aware of, or saw an episode of bullying but was not either the bully or the victim [29–31].
Students were asked to report how often they engaged in the following behaviors when they witnessed a bullying episode in the past year: reported it to the teacher, reported it to the disciplinary office of the school, stopped the perpetrator from hitting the victim, stopped the perpetrator from making fun of the victim, told others not to join in the bullying, comforted the victim, tried to help the victim, tried to be a peacemaker, and fetched people to come and help . Response categories ranged from 1 (never) to 5 (always) (Appendix 2). The standardized Cronbach’s alpha for defending behaviors was 0.9. The score for defending behaviors was calculated by adding the scores of each item divided by the number of the items the students responded to. However, if a student responded to fewer than four items, then the student’s score was recorded as missing. A higher score indicated a higher frequency of defending behaviors.
Mental health indicators
Two mental health indicators, social anxiety and depressive symptoms, were included in this study.
Social anxiety score
There were seven items related to social anxiety in which the students were asked if they had had any of the following experiences in the past two weeks: feeling afraid to make new friends, worrying about being laughed at, feeling afraid to talk to strangers, feeling someone is laughing at you, feeling afraid someone is talking behind your back, feeling afraid others dislike you, and feeling afraid to perform or answer questions in public. This scale was developed based on the Social Anxiety Scale for Children (SASC-R)  and the Social Phobia and Anxiety Inventory for Children (SPAI-C) . All the items were measured using a three-point scale with the possible responses of never (1), once or twice (2), or many times (3). The standardized Cronbach’s alpha was 0.82. We treated the score in the same way as the defending behaviors score.
Depressive symptoms score
Regarding depressive symptoms, students rated how often in the past two weeks they had experienced the following emotions: didn’t feel like eating their favorite foods, felt very sad, cried for no reason, found it hard to carry out tasks, felt frightened, didn’t sleep well, lacked motivation, and felt gloomy and unhappy. This scale was designed based on the Center for Epidemiological Studies Depression Scale for Children (CES-DC)  and related literature [35, 36], and has been used in previous studies [37–39]. All items were measured using a three-point scale with the possible responses of never (1), once or twice (2), or many times (3). The standardized Cronbach’s alpha was 0.76. We treated the score in the same way as the defending behaviors score.
Based on past literature [1–3], factors associated with both bullying and mental health indicators were controlled for including student characteristics (sex, grade, aggressive tendency and communication skills) and family indicators (parents’ marital status, parents’ highest education level, family activity, family support, family conflict and level of punishment).
Descriptive analysis was conducted to examine the distributions of the variables and followed by analysis of variance (ANOVA). Finally, a multivariate linear regression was performed to investigate the relationship between defending behaviors and mental health indicators whilst controlling for other variables. SAS 9.3 Survey Analysis procedures was used to conduct the above statistical analyses. Sampling weights were calculated from the inverse of the inclusion probabilities in consideration of sampling design effects.
Hypothesis 1 was tested using defending behaviors as the outcome variable and bullying roles as the predictor with adjustments made for several control variables. Hypothesis 2 was tested using mental health status as the outcome variable and defending behaviors as the predictor with or without bullying roles as a covariate along with other control variables. Hypothesis 3 was tested by stratifying the sample by bullying roles, then using mental health as the outcome variable and defending behaviors as the predictor, with adjustments made for other control variables.