Using Early Childhood Behavior Problems to Predict Adult Convictions

Abstract

The current study examined whether teacher and parent ratings of externalizing behavior during kindergarten and 1st grade accurately predicted the presence of adult convictions by age 25. Data were collected as part of the Fast Track Project. Schools were identified based on poverty and crime rates in four locations: Durham, NC, Nashville, TN, Seattle, WA, and rural, central PA. Teacher and parent screening measures of externalizing behavior were collected at the end of kindergarten and 1st grade. ROC curves were used to visually depict the tradeoff between sensitivity and specificity and best model fit was determined. Five of the six combinations of screen scores across time points and raters met both the specificity and sensitivity cutoffs for a well-performing screening tool. When data were examined within each site separately, screen scores performed better in sites with high base rates and models including single teacher screens accurately predicted convictions. Similarly, screen scores performed better and could be used more parsimoniously for males, but not females (whose base rates were lower in this sample). Overall, results indicated that early elementary screens for conduct problems perform remarkably well when predicting criminal convictions 20 years later. However, because of variations in base rates, screens operated differently by gender and location. The results indicated that for populations with high base rates, convictions can be accurately predicted with as little as one teacher screen taken during kindergarten or 1st grade, increasing the cost-effectiveness of preventative interventions.

Keywords: Base rates; Convictions; Preventative intervention; Screening.

Fig. 1

Full Sample - Predicting Any Conviction by Age 25 (Prevalence = 39%). Note: T_K refers to model including kindergarten teacher screen score. T_1 refers to model including 1st grade teacher screen score. T_K_ 1 refers to model including kindergarten and 1st grade teacher screen scores. TP_K refers to model including kindergarten teacher and parent screen scores. TP_1 refers to model including 1st grade teacher and parent screen scores. TP_K_1 refers to model including kindergarten and 1st grade teacher and parent screen scores. All points above the dotted horizontal 0.5 sensitivity line represent cutoffs for which sensitivity was 0.5 or greater. All cutoffs to the left of the solid vertical line represent cutoffs for which PPV is 0.5 or greater (given the sample base rate). Therefore, the cutoffs represented by the ROC curve in the upper left quadrant of the figure reflect cutoffs that met both requirements of a good screening measure

Fig. 2

Predicting Any Conviction by Age 25, by Site. Note: Abbreviations refer to screen scores included in the model: T_K = kindergarten teacher screen score only, T_1 = 1st grade teacher screen score only, T_K_1 = kindergarten and 1st grade teacher screen scores, TP_K = kindergarten teacher and parent screen scores, TP_1 = 1st grade teacher and parent screen scores, and TP_K_1 = kindergarten and 1st grade teacher and parent screen scores. All points above the dotted horizontal 0.5 sensitivity line represent cutoffs for which sensitivity was 0.5 or greater. All cutoffs to the left of the solid vertical line represent cutoffs for which PPV is 0.5 or greater (given the sample base rate). Therefore, the cutoffs represented by the ROC curve in the upper left quadrant of the figure reflect cutoffs that met both requirements of a good screening measure

Fig. 3

Predicting Any Conviction by Age 25, by Gender. Note: Abbreviations refer to screen scores included in the model: T_K = kindergarten teacher screen score only, T_1 = 1st grade teacher screen score only, T_K_1 = kindergarten and 1st grade teacher screen scores, TP_ K = kindergarten teacher and parent screen scores, TP_1 = 1st grade teacher and parent screen scores, and TP_K_1 = kindergarten and 1st grade teacher and parent screen scores. All points above the dotted horizontal 0.5 sensitivity line represent cutoffs for which sensitivity was 0.5 or greater. All cutoffs to the left of the solid vertical line represent cutoffs for which PPV is 0.5 or greater (given the sample base rate). Therefore, the cutoffs represented by the ROC curve in the upper left quadrant of the figure reflect cutoffs that met both requirements of a good screening measure