Abstract: Dyadic studies, in which two persons interacting with each other (called a dyad) are the fundamental unit of analysis, are widely used in psychological studies involving interpersonal phenomena. The integration of such studies with intensive longitudinal designs helps to further investigate the dynamics of both individual behaviors and interpersonal effects during social interactions. However, due to the data complexity, there is a lack of appropriate statistical approaches that can adequately answer the dyadic research questions of interest based on the characteristics of intensive longitudinal data.
Drawing on the Dynamic Structural Equation Modeling (DSEM) framework for intensive longitudinal data from independent individuals, this study focuses on the construction, extension, and application of statistical models for intensive longitudinal data with different dyadic designs, which can describe and explain the dynamics of individual behaviors and interpersonal relationships in different social interaction contexts, as well as their between-person differences. Specifically, the main contents of this study are as follows. Study 1 focuses on the standard dyadic design in which information is collected from two persons who are interdependent (e.g., couples, siblings, friends). It aims to improve the existing extension of the actor-partner interdependence model in DSEM to better characterize the relationships between two dyad members, and more importantly, to compare and select the appropriate detrending approach based on the improved model, as DSEM assumes stationarity of variables but time trends, a possible source of non-stationarity, are common in longitudinal studies. We conjecture that the detrending practice based on the residual DSEM (RDSEM) framework would work well. Study 2 focuses on the one-with-many (OWM) design for examining features of multiple dyadic relationships, in which one set of focal persons (e.g., therapists, physicians) has with others (e.g., multiple clients, multiple patients). For the intensive OWM design, Study 2 is intended to develop an appropriate statistical model, as well as the corresponding parameter estimation algorithm, to examine the dynamic processes of multiple interpersonal relationships of focal persons. Specifically, a model assuming no time trends will first be constructed. Then it will be integrated with a detrending method based on the conclusion of Study 1 to be applicable to more complex data with time trends. Study 3 turns to a more complex dyadic context, that is, round-robin data, in which every member of a group has judged all other group members and is also judged by every group member. This study aims to develop an appropriate statistical model for the intensive longitudinal round-robin data, as well as the corresponding parameter estimation algorithm, and then to extend the model to data with time trends. Study 4 illustrates the application of the constructed or extended models under three intensive longitudinal dyadic designs through empirical data.
Advances in data collection techniques and the deepening of research questions have posed new challenges to statistical analysis methods. The unique contribution of this research is to develop and extend a series of methodological tools for the newly emerging intensive longitudinal dyadic data. Considering the diversity of interpersonal interaction contexts, we consider three common dyadic designs. In addition, detrending is a non-negligible issue for the application of DSEM but has not yet been fully investigated. This research makes an exploration for detrending in dyadic contexts. Moreover, we plan to develop a user-friendly software package to include all the developed and extended models in this research, so as to provide a basis for the application of intensive longitudinal dyadic data analysis. This research will help applied researchers to gain a deeper and more scientific understanding of changes in individual behaviors and interpersonal effects in the context of social interactions, and to promote theoretical development in related fields.

Key words: intensive longitudinal data, dyadic designs, model construction, dynamic structural equation modeling