ABSTRACT
Researchers in the behavioral and social sciences often deal with complex models involving interaction and nonlinear effects in addition to main effects of variables. For hypotheses about interaction and nonlinear effects between continuous variables, analyses are frequently conducted using multiple regression techniques (e.g., Aiken & West, 1991; Jaccard, Turrisi, & Wan, 1990). However, concerns with these techniques have been raised regarding measurement error that often introduces bias in regression coefficients and lowers the power of statistical tests for these nonlinear effects (Busemeyer & Jones, 1983; Jaccard & Wan, 1995). Therefore, methods such as structural equation modeling (SEM) with latent variables, which can account for errors in measurement, have generally been recommended for modeling interaction and quadratic effects (Jaccard & Wan, 1996).
