Multilevel Latent Profile Analysis for Daily Diary Data: Understanding Triadic Family Dynamics

Introduction: Recent exponential growth of intensive longitudinal data, such as daily diaries and ecological momentary assessments, calls for innovative ways of analyzing these data with an inherently nested structure. By matching theory to innovative methods, it is increasingly possible to characterize complex processes across and within contexts, relationships, and individuals. To understand adolescent development in the day-to-day dynamics of multiple family relationships, this study introduces multilevel latent profile analysis (MLPA) using daily dairy data from mother-father-adolescent (MFA) triadic relationships.

Methods: Daily dairy data across 21 days were collected from 145 adolescents (63.4% female) in two-parent families. Adolescents were 13-16 years old (M=14.61, SD=0.83). Adolescents reported on mother-father (MF), mother-adolescent (MA), and father-adolescent (FA) closeness each day. These 3 indicators were used in MLPA to identify and describe Level-1 latent profiles of family structures and Level-2 latent classes of dynamic patterns across the 21 days. MLPA extends the traditional latent profile model to accommodate data with a hierarchical structure by allowing Level-1 profile prevalences to vary across Level-2 classes.

Results: Six profiles were identified at Level-1 across all families and all days: Cohesive (high closeness in all three dyads), Mother Centered (high MF closeness, average MA closeness, low FA closeness), Adolescent Centered (high MA and FA closeness, low MF closeness), MA Coalition (high MA closeness, low MF and FA closeness), Disengaged (low closeness in all three dyads), and Average (average to low closeness in all three dyads). Building from this Level-1 solution, a Level-2 random effects model using a non-parametric approach was used to model relationship dynamics. Five classes were identified at Level-2: Stable Cohesive (35% of families, characterized by families with a Cohesive structure most days), Stable Disengaged (20%, families with a Disengaged structure most days), Stable MA Coalition (4%, families with a MA Coalition structure most days), Chaotic (17%, families with multiple different structures across days), and Stable Average (24%, families with an Average structure most days).

Discussion: This presentation will provide a step-by-step approach to applying MLPA as an innovative and feasible way to capture simultaneously family structure and dynamics at the daily level. Responding to the increasing availability of intensive longitudinal data in prevention research, this study illustrates how MLPA can characterize dynamic, complex, and multidimensional risk processes.