To examine the predictive accuracy of antisocial behaviours among 4- and 5-year-old children for problem behaviours 4 years later (ages 8 and 9 years).
Data from the Ontario Child Health Study (1983) and Follow-Up (1987) are used. Predictive accuracy is conceptualized using positive predictive value (PPV) and sensitivity. The predictive accuracy of early antisocial behaviours for the 1987 outcomes is examined overall, by gender, by variable thresholds of predictor and outcome by gender, and by using contextual variables alone or in combination with antisocial behaviour recorded in 1983.
The predictive accuracy of 1983 antisocial behaviour for 1987 outcome is generally modest and differs by gender (better for boys for externalizing disorder [PPV = 41%, sensitivity = 57%]; better for girls for internalizing disorder [PPV = 13%, sensitivity = 80%]; better for boys for conduct problems [PPV = 54%, sensitivity = 21%]). Using either gender-specific thresholds or gender-neutral thresholds does not alter predictive accuracy in a consistent way, nor does the use of a single contextual variable. Use of a cumulative risk index increases PPV but decreases sensitivity.
The predictive accuracy of antisocial behaviour in 4-and 5-year-old children over 4 years in a nonclinical community population is limited. The clinical, research, and policy implications of this work are discussed.