ABSTRACT

Kenny and Judd (1984) formulated a model with interaction effects of two latent variables and suggested using product variables to estimate the model. Using one and four product variables and simulation techniques, Yang-Jonsson (1997) studied the estimation of this model with three methods: maximum likelihood (ML), weighted least squares (WLS), and weighted least squares based on the augmented moment matrix (WLSA). Because the model implies non-normality, one would expect WLS and WLSA to be better than ML at least in large samples, but Yang-Jonsson (1997) found that ML often works well over a range of sample sizes from 400 to 3,200, except that asymptotic standard errors and chi-squares of ML estimates are, in principle, incorrectly computed. In this chapter, we show that both asymptotic standard errors and chi-squares for ML can be corrected for non-normality using Satorra-Bentler type scaling corrections.