Depressive symptom dynamics, including change trajectories and symptom variability, have been related to therapy outcomes. However, such dynamics have often been examined separately and related to outcomes of interest using two-step analyses, which are characterized by several limitations. Here, we show how to overcome these limitations using location-scale models in a dynamic structural equation modeling framework. We introduce location-scale modeling in an accessible manner to pave the way for its use in research integrating within-person dynamics and intervention-related change in psychopathology, and we illustrate this modeling approach in a large-scale internet-based intervention for depression ( = 1,656). Using eight data points sampled across about 8 weeks, we predicted improvement across the intervention (50% symptom reduction) as a function of early change and symptom variability. Early symptom change was associated with a more likely improvement across therapy. Variability of symptoms beyond change trajectories during the intervention was associated with less likely improvement. Location-scale models, and dynamic structural equation modeling more generally, are well suited to modeling how patterns of symptom change during psychotherapy are related to important (e.g., therapy) outcomes. Our illustrative application of location-scale modeling showed that symptom variability was associated with less overall improvement in depressive symptoms. However, this finding requires replication with more intensive sampling of symptoms before final conclusions can be drawn on when and how to distinguish maladaptive from adaptive variability during psychotherapy. (PsycInfo Database Record (c) 2024 APA, all rights reserved).