LAUSR

Affect dynamics, as a field, has grown immensely in the past three decades—as have methods for measuring fast-changing affect and modeling complex processes. Still, the field as a whole is largely reliant on defining affect dynamics as simple summary statistics of timeseries, rather than modeling affect timeseries as a process. Process-oriented modeling (i.e., timeordered models, usually framed around change in the system based on past states) can account for multiple dynamic features at once, and parameters can be interpreted in terms of theoretical features of the processes which produce dynamics. For example, autoregressive (AR) models, one common choice for affect dynamic modeling, quantify affective baselines (the intercept), affect inertia (autoregressive parameter), and the degree of reactivity/variability (variance of the innovation/process noise) in one’s affect system. Another gap in the affect dynamics literature is understanding the relationship between affect dynamics and particular contexts which evoke them. Most studies of affect dynamics are conducted via ecological momentary assessment (EMA) or daily diaries, which observe how dynamics unfold over the timescale of days or hours, with limited ability to capture the contexts of daily life which evoked changes in affect. Lab-based paradigms, however, are particularly well-suited to examine individual differences in affect dynamics across contexts. In this study, we use data from a community-based sample (N1⁄4 73), wherein participants rated their “real-time” affective feelings using a joystick, following affectively-valanced prompts. The six stimulus categories (within which there were six trials each) were Major/ Minor Negative events, Major/Minor Positive Events, Neutral events, and Ambiguous events—where the affective valence was unclear, but the event was not inconsequential. The substantive motivation behind this study was to examine whether participants with “clinical status” (CS; a current diagnosis of either MDD or GAD) differed in affect dynamics either across or within contexts from those without current psychopathology. Methodologically, these data and the related research questions call for a hierarchical model: at the timeseries level, we modeled affect dynamics using an AR model to estimate affective baselines, inertia, and intraindividual variability. These three parameters were allowed to be personand categoryspecific, allowing examination of between-person differences in affect dynamics (i.e., differences associated with clinical status), and between-category differences. The Bayesian framework is particularly useful for fitting complex hierarchical models, as it provides flexibility to assign hierarchical distributions to nearly any parameter. Using custom-written JAGS code, we fit the hierarchical AR model, nesting trials within stimulus categories within persons. We found that participants with CS, on average, exhibited lower affective baselines and less variability in affect, but not credibly different affect inertia than participants without CS. There were, however, some categories where this average trend was not found to be credible (e.g., baselines between CS vs. non-CS participants in the Major Positive category were not found to be different). Additionally, contrasts of the posterior distributions between categories and clinical status revealed differences in magnitude of differences between categories for CS and non-CS participants (e.g., larger differences in affective baselines between