Abstract

VALENTINI, Felipe; MOURAO, Luciana  and  FRANCO, Vithor Rosa. Latent models and random slopes for testing moderation and mediation. Aval. psicol. [online]. 2018, vol.17, n.4, pp. 439-450. ISSN 1677-0471.  http://dx.doi.org/10.15689/ap.2018.1704.4.04.

Complex models, such as mediation and moderation, are essential for comprehending psychological variables. Nevertheless, during the testing of such models, the structure and reliability of the instruments might affect the estimated relationship between variables. Thus, models of mediation and moderation that correct the parameters yielding error estimates are fundamental. The objective of this study is to compare the performance of three Structural Equation Modeling models that include mediation and moderation effects: Random slopes, latent interactions and path analysis. The three models were compared using both simulated and real data. The results showed that when the variance contained in the slopes is small, both the random slopes model and the latent interaction model are equivalently good. The path analysis model, on the other hand, performs well to recover the effects of moderation, but it underestimates the direct and indirect effects of mediation. Best practices for testing hypotheses with complex effects are discussed, and we suggest using latent models for testing mediation and moderation.

Keywords : slopes randômicos; path analyses; moderation; mediation; latent interactions; random slopes; path analysis; path analysis.

        · abstract in Portuguese | Spanish     · text in Portuguese     · Portuguese ( pdf )