Measuring violence perpetration: Stability of teachers' self-reports before and after an anti-violence training in Cote d'Ivoire.

BACKGROUND: Epidemiological studies of interpersonal violence commonly use self-reported violence perpetration as an outcome measure, but few studies have investigated the stability of and influences on self-reports. OBJECTIVE: To assess changes in teachers' self-reported use of physical violence against students before and after a one-day violence prevention training, and factors associated with changed reports in Cote d'Ivoire. METHODS: Before and after the training, 157 teachers completed surveys containing 32 questions adapted from the ICAST-CI. Changes in physical violence usage were summarized over lifetime, past school term, and past-week timeframes, and the consistency in responses assessed via intraclass correlation coefficients (3,k), percent agreement, and kappa statistics. Factors associated with changed reports were assessed using robust multiple linear regression with 1,000 bootstrapped replications. RESULTS: Although reports before and after the training should have remained constant, the proportion of teachers reporting 1+ act of violence dropped substantially (lifetime: 73% to 47%). Most teachers (73%) changed 1+ response. Kappa for individual items showed ranging disagreement (lifetime: 0.275-0.795). Variables significantly associated with greater numbers of changed reports included: greater mental health distress (lifetime: beta = 1.061, 95% CI = 0.229, 2.404), older age (past school term: beta = 0.067, 95% CI = 0.018, 0.113); and variables targeted during training, including increasing awareness of consequences of violence (past week: beta = 0.241, 95% CI = 0.046, 0.435) and decreasing acceptance of physical discipline practices in schools (past school term: beta= -0.169, 95% CI= -0.338, -0.045). CONCLUSIONS: Interpreting self-reports of violence perpetration requires caution. Formal investigations into reliability and validity of self-reported violence perpetration and victimization are needed.