Description of the data set and design of the study
The data reported here are coming from the Epirus School Project . This was a cross-sectional survey carried out in selected upper secondary schools in Greece with the aim to investigate the prevalence and associations of common mental disorders in late adolescence.
Sampling of Schools and Pupils
Upper secondary schools in Greece are either Senior High Schools (Lycea) or Technical Vocational Schools but 75% of students attend the first. In the current study only Senior High Schools were selected while Technical Vocational Schools will be included in a separate future survey. Approximately 75000 students attended 1193 Senior High Schools at the time of the design of the study. Schools were selected according to the following rules: a) all senior high schools of the major cities in the North-Western Part of Greece (Regions of Epirus and Aetoloakarnania) due to the proximity with the University of Ioannina, b) all senior high schools in one randomly selected district of the Athens Greater Area (the district of Kallithea was selected), c) all senior high schools of one island in the Aegean Sea (the island of Paros was conveniently selected).
All students in the selected schools were invited to participate in the study. Written consent for participation was actively obtained from both the students and their parents. Ethical approval for the study was also obtained by the Ministry of Education.
Design of the study and data collection procedure
The study used a two-phase design . In the first phase, all consenting students (N = 5614) were administered a brief screening instrument (see next section) in the classroom and then students were invited for the second phase using a stratified random sampling procedure according to the scores on the screening questionnaire: 100% of those scoring high on the screening instrument (>75th percentile), 30% of those scoring in the middle and 10% of those scoring low (<25th percentile). The second phase (N = 2431) consisted of the computerized version of a fully-structured psychiatric interview (see next section) and was carried out in the computer laboratories of the schools. The main fieldwork took place between January 2007 and April 2008.
Assessment of Psychiatric Morbidity
Psychiatric symptoms were assessed with the revised clinical interview schedule (CIS-R), a fully structured psychiatric interview designed to be used by trained lay interviewers . The CIS-R was the main instrument used in the national psychiatric morbidity surveys in the UK [34, 35] and has been used in several other similar surveys around the world [36, 37]. A computerized version has also been developed and found to be comparable with the regular interview . The CIS-R was originally designed to assess symptoms in participants above 16 years old but has been previously used in teenagers above 14 years old in Australia [8, 39].
The CIS-R assesses the presence and severity of 14 different common psychological symptoms (somatic symptoms, fatigue, concentration/memory problems, sleep problems, irritability, worry about physical health, depression, depressive ideas, worry, free-floating anxiety, phobias, panic, compulsions and obsessions). Two screening questions in each section ask about the presence of the symptom during the past month and then there is a more detailed assessment of the presence, frequency, duration, and severity of the symptom during the past seven days. Each symptom section is scored from 0 to 4 (except depressive ideas from 0 to 5) and a score of 2 or more denotes a clinically significant symptom and a total score of 18 or more indicates a clinical significant case . Additional questions enable the application of the ICD-10 research diagnostic criteria using specially developed computerized algorithms .
For screening purposes in the first phase of the study we used the screening questions of the several symptom sections of the CIS-R. The full interview was given to those selected for the second phase (N = 2431)
The Greek version of the CIS-R has been validated and its psychometric properties have been published elsewhere . The Cronbach's alpha for each symptom dimension ranged from 0.84 to 0.87 with an overall alpha for CIS-R of 0.86. A test-retest reliability of the CIS-R has been calculated in a subset of the present data set (two schools of the city of Ioannina with an interval between assessments of two weeks) and was found to be 0.84 . For the purposes of the present study psychiatric morbidity can be assessed either in a dimensional way, using the total score on the CIS-R (by adding-up all 14 symptom dimensions), or in a categorical form using diagnostic categories. We have selected to use the total score in our analyses because in that way we are able to adjust for the full spectrum of psychiatric morbidity including sub-threshold forms of illness.
Assessment of Suicidal Ideation
Suicidal ideation is commonly assessed by a set of questions of increasing severity that aim to investigate the full spectrum of suicidal thoughts and/or behaviour. The CIS-R starts by asking the following question about "tiredness of life": "in the past week have you felt that life isn't worth living?". Participants who reply positively are then asked the subsequent questions about death wishes ("have you wished that you were dead?") and actual suicide thoughts ("have you thought of taking your life even if you would not really do it?"). In the context of the present study we selected to ask all participants the first question only, without investigating further the more severe spectrum of suicidal ideation. We did this for the following reasons: a) our sample was not clinical and consisted of generally healthy adolescents attending secondary schools. We anticipated that the more severe spectrum of suicidal ideation would be relatively rare in this population and the statistical analysis would have been underpowered; b) there seems to be a continuum between less severe forms of suicidal ideation such as "tiredness of life" (as assessed by the "life isn't worth living" question) and the more severe forms of death wishes or actual suicidal ideas and there is not any clear cut-off to distinguish between these three groups . Previous studies have shown that the pattern of associations with sociodemographic factors and psychiatric morbidity is the same between these groups and any observed differences are of a quantitative rather than a qualitative nature [28, 41]; c) inclusion of the more severe forms of suicidal ideation in our survey would make necessary the implementation of an intervention for those pupils that would admit actual ideas of harming themselves. Such an intervention was not feasible for half of the schools that we planned to include in the study, therefore we preferred to exclude these questions in order to include a larger sample of schools and pupils. For all of the above reasons we opted for excluding these questions.
Participants could select three possible answers to the question of whether they were thinking that life was not worth living in the past week: "no", "yes sometimes", "yes all the time". As this question is the least severe form of the spectrum of suicidal ideation, we classified students as having suicidal ideation if they selected the third answer "all the time". All other students were classified into the "no/uncertain" category.
Assessment of Bulling Behavior
Involvement in bullying either as a perpetrator (bully others) or as a victim (being bullied by others) was investigated in the second phase of the study using two questions taken from the revised Olweus Bully/Victim Questionnaire  which was also used in a WHO youth health study . An introductory sentence defined bullying as follows:
''The next questions are about bullying. We say a pupil is being bullied when another pupil, or a group of pupils, says or does nasty and unpleasant things to him or her. It is also bullying when a pupil is teased repeatedly in a way he or she doesn't like. But it is not bullying when two pupils of about the same strength quarrel or fight.''
Thereafter the respondents were asked how frequently they had been bullied or they had bullied others, during the last 2 months in school. The possible answers were: "many times a week", "about once a week", "2 or 3 times per month", "1 or 2 times during the last 2 months" and "not at all". Based on these responses we classified participants into the following groups: a) Being a perpetrator ("bullying others") versus not being a perpetrator (reference category); b) being a victim versus not being a victim (reference category). We should like to note that this grouping allows the comorbidity between the two states, i.e. a perpetrator may also be a victim or vice versa. Other studies have used pure states ("pure" victims, "pure" perpetrators and both victim and perpetrators) but in our study we allowed comorbidity to investigate more formally whether there is statistical interaction between victims and perpetrators.
If the participant had been involved in this behavior at least once a week, this was classified as "frequent" bullying or victimization respectively, whereas all other instances were classified as " less frequent" bullying or victimization. Although this categorization is a bit arbitrary, it has been used in the past in other papers [9, 10]. Bullying is considered to be a continuous process and including in the "bullying category" those pupils who had been involved once or twice during the past two months may not be universally accepted. However, we included those pupils in our definition, first to increase the statistical power of our study and second because empirically those pupils wee more similar regarding their association with psychiatric morbidity to the pupils with higher frequency bullying.
Information about several sociodemographic variables were obtained from the students in the first phase of the study (own age, parent's age, gender, parent's marital status, number of brothers and sisters, mother's educational status, father's educational status, mother's employment status, father's employment status). Students were also asked to subjectively rate their academic performance in school on a 4-point scale (excellent, very good, good, fair) and their relationship with mother and father (excellent, very good, good, fair, bad). In addition we asked students to subjectively assess their family's financial condition by asking them whether their family was having any financial difficulties (measured on a 4-point scale: no, few, some, a lot).
All analyses were performed with STATA/SE 9.2 (StataCorp, College Station, Texas). To take into account the potential effect of clustering of our data (since adolescents were nested into 25 schools) we first carried out a two-level logistic model (level 1: individuals, level 2: schools) in Stata using the gllamm command . We also performed the models with the survey commands of Stata (svylogit) using school as the stratum. Results were very similar with both models and therefore in the paper we present the results using the survey commands because their use is more widespread in the literature. It should be noted that the effect of schools was negligible with an intraclass correlation coefficient close to zero (<0.08). In all analyses we have used probability weights to take into account the stratified random sampling procedure.
Adjusted population-attributable risk fractions (PAFs) and their 95% CIs were calculated from the final multivariable logistic regression model by using the aflogit procedure in Stata .