Linear and non-linear

Abstract

Objectives: Prior scholarship assessing the association between verbal IQ and antisocial behavior has generally assumed that the functional form of the association is linear. Nevertheless, if this assumption is not valid, the interpretations from prior scholarship could be biased. The current study re-examines the functional form of the association between verbal IQ and antisocial behavior to address three limitations in the prior scholarship and four research questions. Methods: The current study uses data derived from the restricted version of the Add Health (N = 9947–14,921). Forty-eight multivariate negative binomial regression models and twelve bivariate non-parametric kernel regression models were used to estimate the association between verbal IQ and antisocial behavior. Results: The findings suggested that verbal IQ had both a linear and curvilinear association with antisocial behavior depending upon the operationalization of antisocial behavior, the age of the participants, the sex of the participants, and the race of the participants. Conclusions: The findings demonstrated the need for a formalized process to ensure the proper specification of the direct association between verbal IQ and antisocial behavior. As outlined, the two-step process encourages future scholarship to be extremely meticulous when specifying the association between verbal IQ and antisocial behavior.