Abstract
Perseverative thinking (PT), such as rumination or worry, is a transdiagnostic process implicated in the onset and maintenance of emotional disorders. Existing measures of PT are limited by demand and expectancy effects, cognitive biases, and reflexivity, leading to calls for unobtrusive, behavioral measures. In response, we developed a behavioral measure of PT based on language. A mixed sample of 188 participants with major depressive disorder, generalized anxiety disorder, or no psychopathology completed self-report PT measures. Participants were also interviewed, providing a natural language sample. We examined language features associated with PT, then built a language-based PT model and examined its predictive power. PT was associated with multiple language features, most notably I-usage (e.g., "I", "me"; β = 0.25) and negative emotion language (e.g., "anxiety", "difficult"; β = 0.19). In machine learning analyses, language features accounted for 14% of the variance in self-reported PT. Language-based PT predicted the presence and severity of depression and anxiety, psychiatric comorbidity, and treatment seeking, with effects in the r = 0.15-0.41 range. PT has face-valid linguistic correlates and our language-based measure holds promise for assessing PT unobtrusively. With further development, this measure could be used to passively detect PT for deployment of "just-in-time" interventions.