Reciprocal relations in categorical variables

The analysis of reciprocal relations in categorical variables poses methodological challenges. Effects that go in opposite causal directions must be integrated into the same model, and parameters must be interpretable. In this article, we propose taking an event-based perspective and present a new approach to the analysis of reciprocal relations in manifest categorical variables. Instead of asking questions about associations of categorical variables, the event-based perspective asks whether the occurrence of one event (the cause) leads to the occurrence of another event (the effect), and vice versa. Event-based reciprocal log-linear models are described. The presented approach enables one to estimate separate unidirectional causal effects in the same log-linear model. The Schuster transformation is applied to obtain interpretable parameter estimates when design matrices are nonorthogonal. A simulation study illustrates the viability and power of the proposed approach. Data examples illustrate the applicability of the proposed method, and that analysis of reciprocal relation hypotheses without Schuster transformation can lead to incorrect conclusions. Extensions of the proposed models are discussed. (PsycInfo Database Record (c) 2020 APA, all rights reserved).