Suicidal Behavior in Adolescent Friendship Networks: Disentangling Peer Selection and Influence Effects in an Effectiveness Trial

Introduction: Approximately 1 million adolescents in the US attempt suicide each year. Adolescents who report that a friend attempted suicide are 2-3 times more likely to be suicidal themselves, and the relative risk of dying by suicide after a suicide death in one’s social sphere is 2-4 times higher among 15-19 year olds than other age groups. These findings have been widely interpreted to suggest that adolescents are susceptible to suicide imitation. Determining whether imitation occurs, however, requires disentangling influence from adolescents’ tendency to select friends with similar behaviors. Disentangling selection and influence has potentially significant implications for the design and testing of preventive interventions.

Method: In an effectiveness trial of a peer-led universal suicide prevention program, 40 high schools were randomly assigned to either immediate training (n=20) or two-year wait-list (n=20). All students were invited to complete a baseline survey, followed by surveys 6 months, 12 months, and 18 months post-baseline. At each wave, students named up to 7 closest friends at school and answered questions about whether they had considered or attempted suicide (suicide ideation) in the past 12 months. We recoded this variable to focus on new occurrences (student did not report suicide ideation at the immediately preceding wave). Students also answered demographic, attitude, and behavioral questions. After baseline in early intervention schools, key student peer leaders were selected, trained and conducted suicide prevention activities over 16 months.

Results: We used an innovative analytic approach (actor-based modeling in RSiena) to test whether selection and influence impacted likelihood of suicidal ideation and behavior. Initial models with one school provided evidence of selection: students were more likely to select friends whose suicide ideation was similar to their own, even after controlling for selection due to gender, race, and grade. These effects were reduced after controlling for selection due to behavioral characteristics (e.g., depression, delinquency, drug use intentions). Initial models also provided evidence of influence: Students whose friends had higher rates of suicide ideation were more likely to report suicide ideation. These effects were reduced after controlling for other predictors of suicide ideation, such as depression.  

Conclusion: Our preliminary findings provide support for both selection and influence with respect to suicide ideation. Final analyses will include results from all 40 schools and will test whether similar selection and influence processes apply for suicide attempts and whether these effects are moderated by the suicide prevention activities of key opinion leaders.