Background: As phenotypes of depressive disorders (DD) are highly heterogenous, a growing number of studies investigate person-specific associations of depressive symptoms in time series data. Most available methods for estimating applicable models rely on the assumption that the associations between variables stay constant over time, which can be unrealistic in clinical contexts. To circumvent this limitation, we used a recently developed technique to estimate time-varying vector autoregressive models. Methods: In daily diary data of 20 participants with DD with a mean length of 274 days (SD = 82.4, range = 154-539), we modeled idiographic associations between core depressive symptoms, rumination, sleep, and quantity and quality of social contacts as idiographic time-varying dynamical networks. Results: Resulting models showed marked inter- as well as intraindividual differences. For some participants, associations between variables changed fast over time, whereas for others they showed more stability. Our results further indicated nonstationarity in all time series. Discussion: Idiographic symptom networks of depression can be of interest to clinicians and researchers as they can capture changes over time and provide detailed insights into the temporal course of mental disorders. Whilst the assumption of stationarity can hinder insights into important change processes, time-varying network models are a promising approach. We discuss limitations, their possible solutions, and recommendations for further use of the modeling technique.

Idiographic network models are estimated on time-series data of a single individual and allow researchers to investigate person-specific associations between multiple variables over time. The most common approach for fitting such graphical vector autoregressive (gVAR) models uses LASSO regularization to estimate a contemporaneous network and a temporal network. However, estimation of idiographic networks can be unstable in relatively small data sets typical for psychological research. This bears the risk of misinterpreting differences in estimated networks as spurious heterogeneity between individuals. As a remedy, we evaluate the performance of a Bayesian alternative for fitting gVAR models that allows for regularization of parameters while accounting for estimation uncertainty. We first compare Bayesian and LASSO approaches across a range of conditions and performance measures in a simulation study. Overall, LASSO estimation performed well, while Bayesian gVAR may perform better when the true network is dense. We also develop a novel test, implemented in the tsnet package in R, which assesses whether differences between estimated networks are reliable based on matrix norms. In a simulation study, the test was conservative and showed good false-positive rates. Finally, we apply Bayesian estimation and the novel testing approach in an empirical example using daily data on clinical symptoms for 40 individuals. Overall, Bayesian gVAR modeling facilitates the assessment of estimation uncertainty which is important for studying inter-individual differences of intra-individual dynamics.