Early longitudinal research on school bullying suggested that it predicted later offending. For example, in his follow-up study of over 700 Stockholm boys, Olweus (1991) reported that 36% of bullies at age 13-16 were convicted three or more times between ages 16 and 24, compared with 10% of the remainder. However, there have been surprisingly few more recently published longitudinal studies of the relationship between school bullying and later offending. Therefore, the main aim of this special issue is to present such results from major longitudinal studies, and also fi ndings from a systematic review and meta-analysis, in order to advance knowledge about the link between school bullying and later offending.
Bullying may predict offending because both are behavioural manifestations of the same underlying construct, such as an antisocial personality. Alternatively, bullying may predict offending because bullying is an earlier stage in a developmental or causal sequence that leads to offending. If bullying causes an increase in the probability of later offending, then interventions that reduce bullying should be followed by a reduction in offending. On the other hand, if bullying and offending are merely symptoms of the same underlying disorder, it is less clear that eliminating an earlier symptom would reduce the probability of a later symptom.
In order to establish whether bullying per se is followed by an increase in offending, it is important to investigate whether bullying predicts offending after controlling for earlier risk factors that predict both bullying and offending . All contributors to this special issue were asked to investigate: (1) the strength of the relationship between bullying and later offending, and (2) the strength of this relationship after controlling for earlier risk factors (e.g. child, parental, child-rearing, peer, school, socioeconomic, and neighbourhood). They were not asked to study whether changes in bullying predicted changes in offending, partly because this would have required more data waves, and partly because such change variables are likely to have great variability.