We used data from the baseline survey of the INCLUSIVE study, a cluster randomized controlled trial of an intervention aimed at reducing bullying and aggressive behaviours in 11 to 16 year old students in secondary schools. Baseline data were collected before randomization in May–June 2014 from all Year 7 classes (age 11–12 years) in 40 participating secondary schools within the state education system across south-east England. Full details of the sampling methodology are available in the study protocol [18]. Schools exclusively for those with learning disabilities, behaviour problems (e.g. student referral units) and very poorly performing schools with an Ofsted rating of “Inadequate” were not included in the sample [18]. Data were collected through questionnaires completed in school in confidential sessions supervised by the research team. A total of 6667 students provided baseline data. Students provided demographic details by self-report. Other student-level outcome measures were also assessed by self-report as follows.
Bullying measures
Bullying victimization was assessed using the Gatehouse Bullying Scale (GBS), a 12-item short and reliable instrument previously used in school-based surveys and shown to be related to other measures of social attachments, school engagement, and anxiety and depressive symptoms [19]. The GBS enquires about four categories of bullying, i.e. being the subject of recent name calling, rumours, being left out of things, and physical threats or actual violence from other students in the last three months. In each of these, questions ask about the recent experience of that type of bullying (yes or no), how often it occurred (most days, about once a week, less than once a week), and how upset the student was by each type of bullying (from “I was quite upset,” “a bit upset” to “not at all”). We combined frequency and distress responses to calculate GBS scores as follows: 0 = Not bullied; 1 = Bullied, but not frequently, and not distressed by it; 2 = Bullied, either frequently or distressed, but not both; and 3 = Bullied frequently and distressed. We used these to define two categories of bullying: bullying victimization (GBS score of 2 or 3 collapsed together indicating either frequent or distressing bullying or both) or severe bullying (GBS score of 3 indicating frequent distressing bullying).
The GBS does not specifically include or exclude bullying through social media or other online activities. Cyberbullying was specifically assessed using two items adapted from Smith and colleagues’ DAPHNE II questionnaire [15] asking whether the participant was bullied (victim) and/or bullied someone else (perpetrator) through mobile phone use or the internet over the past three months. Responses were on a five-point Likert scale for each question, from 1 = No, I have not; 2 = Yes, once or twice; 3 = Yes, two or three times a month; 4 = Yes, about once a week; to 5 = Yes, several times a week or more. We dichotomised responses for these analyses into “not/rarely bullied” and “bullied/frequently bullied” for victims and “not/rarely bullied others” and “bullied/frequently bullied others” for perpetration (by collapsing responses 1 and 2 together and 3, 4 and 5 together to obtained these two categories for both cyberbullying victimization and perpetration).
Other student-level characteristics
Young people provided data on SES and family composition. Socioeconomic status was assessed using the Family Affluence Scale (FAS), developed specifically for reporting of SES by young adolescents [20]. Four questions assess car ownership, children having their own bedroom, the number of computers at home, and the number of holidays taken in the past 12 months. A composite FAS score is calculated for each student based on his or her responses to these four items. For our analyses, scores were collapsed to give FAS tertiles of low, medium, and high family affluence, where FAS low (score = 0, 1 and 2) indicates low affluence, FAS medium (score = 3, 4 and 5) indicates middle affluence, and FAS high (score = 6, 7, 8 and 9) indicates high affluence.
Family composition was assessed based on student reports of who lived in their house with them. To create a dichotomous variable (two parents vs lone parent), students were classified as having two parents if they reported living with any two of the following: mother, father, step-mother, step-father, foster mother, and foster father. Students were classified as having a lone parent if they reported living with only one of these parents. In our sample, 73.91% of students reported living with two parents.
School characteristics
Data were available on the following school-level characteristics:
-
1.
School-level deprivation:
-
a.
Proportions of students eligible for free school meal (FSM): this is a widely used proxy measure for economic deprivation in the UK [21, 22]. In England and Wales, local education authority-maintained schools must provide a free midday meal to students if they or their parents receive specific benefits. We used the percentage of students eligible for FSMs at any time during the past six years, obtained from publicly accessible data from Department of Education school performance Tables [23]. The proportion of students eligible for FSM in our sample schools ranged from 3.0% to 79.2% (mean = 36.4%, SD = 19.6).
-
b.
The Income Deprivation Affecting Children Index (IDACI) score of the schools’ postal address: the IDACI scores deprivation that measures the proportion of children in a small area under the age of 16 who live in low income households. It is supplementary to the Indices of Multiple Deprivation and is used for calculation of the educational contextual value added score, measuring children’s educational progress [24].
-
2.
School type: Our sample includes of five different types of schools: community (n = 5), where premises and funding are provided by local authorities; foundation (n = 6), where the school owns its own premises but funding comes from the local authority; voluntary-aided (n = 4), where the premises are owned by a charity but funding is at least partly from the local authority; sponsor-led academy (n = 6) which are usually created from an underperforming school which obtained an independent business or charitable sponsor and where funding comes directly from central government; and converter academy mainstream (n = 18), which are successful schools which have opted to gain more autonomy and have funding directly from central government [25]. Voluntary-aided, community and foundation schools follow the National Curriculum and are supervised by the local authority. In our sample, all voluntary-aided schools were faith schools. Academies do not have to follow the National Curriculum except for core subjects. In addition, they have more freedom in setting their own term times and changing the length of school days.
-
3.
School size: the total number of students in the school [26].
-
4.
Sex composition: mixed sex or single sex [26].
-
5.
School quality:
-
a.
Most recent overall Ofsted rating: in England, schools are inspected by a statutory body, the Office for Standards in Education, Children’s Services and Skills [26]. Ofsted inspections are carried out every 2–5 years, depending on inspection outcomes [27] and all schools had data from 2011 to 14. Schools were classified as 1 = “Outstanding”, 2 = “Good”, 3 = “Requires improvement” or 4 = “Inadequate” based on the quality of teaching, leadership and management, achievement of students, and behaviour and safety of students at the school. Our sample included no schools with a rating of “Inadequate.”
-
b.
Value added (VA) score: a second school quality rating was the VA score, an official measure of the progress students make between different stages of education. To calculate this, a median line approach is used whereby the VA score for each student is the difference (positive or negative) between their own output point score (end of Key Stage 4) and the median output point score achieved by others with the same or similar starting point (Key Stage 2 or 3), or input point score [23]. Scores for VA were given, with schools that neither added nor subtracted value being given a score of 1000.
Statistical analysis
We first described the frequency of bullying (collapsing GBS scores in two different ways to obtain both “significant” and “severe bullying” with only significant bullying being used for subsequent analyses), and cyberbullying perpetration and victimization by sex and ethnicity among students and the distribution of school-level factors across schools.
Previous research has shown that individual-level factors considered here (gender, ethnicity, SES and family composition) have been consistently associated with mental health and bullying outcomes. Therefore, all the models in our study included these factors as covariates.
In step one we examined the association of each school-level factor with bullying and cyberbullying outcomes in separate models, adjusted for all individual-level factors and took account of clustering at the school level. In step two we used multilevel mixed effects logistic regression to examine the associations between bullying and cyberbullying outcomes and individual- and school-level factors, with a random effect for school. This final model included all school-level variables that were found significant at the p < 0.1 in step two, together with all individual-level factors.
Interactions were tested between all individual- and school-level factors that were significant at the 10% level in the final multivariable model. Data were analysed using STATA 13.0 (College Station, TX).