Curvilinear relationships between the continuously assessed index of pubertal timing and later substance use and parent-child relationship quality were tested using the curve fit procedure and examined using plots. For boys and girls, a linear relationship characterised the pattern of associations with the exception of the relationship between boys’ pubertal timing and alcohol use at Time 2 where a quadratic trend was significant after controlling for a linear effect (linear, β = .18, p<.01; quadratic, β = -.11, p<.01). Analysis followed guidance by Ping (1998) for examining interaction and quadratic variables using stacked procedures in structural equation modelling (SEM). Models were estimated using maximum likelihood estimation in LISREL 8.50 (Jöreskog & Sörbom, 1996). The sample was split according to pubertal timing (early; on-time; late). Constraining the model coefficients to be equal between subgroups provides a chi-square statistic which, if significant, indicates a difference in the magnitude of the coefficient between the groups. Subgroup comparison tests using stacked modelling procedures were also employed to consider whether the magnitude of parameter estimates in each model were significantly different for boys and girls.
