Longitudinal Dyadic Data 1 Accounting for Statistical Dependency in Longitudinal Data on Dyads

Longitudinal data on dyads have statistical dependencies due to stable and time-varying characteristics of the dyad members and of their environment. This article discusses two ways of estimating these dependencies, one involving multilevel models and the other involving structural equation models. To illustrate each approach, we analyze data on daily reports of anger by males and females in couples where one partner was a law school graduate preparing to take the bar examination. Although social and developmental psychology define dyadic processes as an important part of their subject matter, there is still considerable uncertainty in these fields about how to analyze dyadic data. The reason for this uncertainty is that conventional statistical methods are designed for studying independently sampled persons, whereas the most interesting feature of dyadic data is their lack of independence. Moreover, in recent years this problem has been compounded as researchers have increasingly adopted intensive repeated-measures designs to study dyads in natural settings (Bolger, Davis & Rafaeli, 2003). When one collects repeated-measures data on dyads, one must not only contend with nonindependence of the members within the dyad but also