Growth mixture modeling was then implemented to determine whether a single quadratic trajectory was sufficient to represent change in externalizing behavior across the entire sample. Consideration of several selection criteria suggested that a 3-class solution fit the externalizing data best. Table 3 shows decreases in both BIC (21955.83 to 20751.22) and AIC (21906.71 to 20685.72) between a single class and the 2-class solution. Likewise, the likelihood ratio test comparing the fit of one versus two trajectories favored the 2-class solution (p < .001). The addition of a third trajectory class was also marked by improved fit as indicated by BIC (20751.22 to 20439.87) and AIC (20685.72 to 20358.00), though there was some decrement in classification entropy (from .90 to .86). However, a formal comparison of differences in the likelihood ratios of the two- and three-class solutions implied that an additional longitudinal pattern of externalizing behavior could be differentiated (p < .01). Regarding the addition of a fourth distinct pattern of externalizing problems, BIC and AIC once again decreased (20275.21 and 20176.97, respectively), while the overall quality of classification was further
