A multilevel analysis of parental discipline and child antisocial behavior

We demonstrate graphical and analytical methods for multilevel (2- and 3-level) models using the analysis of observed microsocial interaction between parent-child dyads as an example. We also present multilevel path diagrams and argue that while not as compact as equations, path diagrams may communicate results better to a wider audience. The example data were drawn from an urban risk sample of kindergarten children and their parents observed in a 1-hour structured lab task and coded for behavioral process using the Family Peer Process (FPP) Code. Four repeated assessments of aversive parent discipline across the 1-hour observation session were nested within dyads, which were nested under coders, which created a 3-level model. The model also included the random effect of observed child aversive behavior as a time-varying predictor of parent aversive discipline at Level 1, child school antisocial behavior as a Level 2 predictor and coder gender as a Level 3 predictor. Our results indicated that coders accounted for about 13% of the variance, families accounted for about 20% of the variance, leaving 67% of the variance in parent aversive discipline due to random, time-specific influences. We discuss the psychometric implications of the small (20%) trait-like component of parent discipline in detail. We also found that observed child aversive behavior provoked parent aversive discipline more strongly in some families than others but child school antisocial behavior was not related to the magnitude of the child aversive effect on the parent.