About this project
We aim to predict, in real time, when people are most likely to get “stuck” in unhelpful patterns of thought—specifically Repetitive Negative Thinking (RNT). RNT includes rumination (replaying past problems) and worry (imagining future threats). Both are linked to many mental health difficulties, from depression to anxiety, and can keep symptoms going even after therapy starts.
In this project, we collect short self-reports throughout the day, combined with information about daily experiences, emotions, and social interactions. We then build a personalized prediction model for each participant to forecast moments when RNT is likely to spike. In the next step, we test Just-In-Time Adaptive Interventions (JITAIs)—short, evidence-based exercises that are delivered exactly when a participant’s personal model predicts they’re at higher risk of getting caught in RNT.
How it works
Daily-life data collection
For three weeks, participants receive eight short surveys a day on their smartphone (each taking under a minute). They report on thoughts, emotions, stressors, and social experiences.
Building personal models
We use each person’s data to train a model that forecasts their RNT episodes. The model identifies which factors—such as certain emotions, types of events, or interpersonal experiences—tend to come before increases in RNT.
Targeted interventions
In the next two weeks, participants get short mental exercises delivered at the right moment. These exercises are chosen based on the predicted intensity of the episode and the participant’s availability. Examples include:
Self-compassion prompts (“How would you treat a friend in this situation?”)
Cognitive defusion (stepping back from thoughts instead of getting pulled in)
Flexible interpretation training (finding alternative explanations for stressful events)
Time-out strategies (brief distraction or calming activities for intense distress)
Half the time, these interventions are sent when the model predicts high RNT; half the time they are sent at random, to test their specific benefit.