A structural neural autopilot analysis of social media use around the pandemic lockdown

This paper estimates a “neural autopilot” model of habit formation using individual-level data on posting behavior from a major Chinese social media platform before, during, and after the 2020 pandemic lockdown. The model yields interpretable parameter estimates for autopilot habit formation and demonstrates that, once habits are neuroscientifically formalized, changes in preferences are no longer required to explain observed behavioral changes. Moreover, the model provides a better fit to the data than both traditional state-dependent habit models and alternative reinforcement learning models. Our analysis further reveals that forced experimentation alone does not generate persistent habitual posting after the lockdown ends. Counterfactual experiments indicate that reducing the volatility of posting rewards, in conjunction with forced experimentation, can significantly strengthen habitual posting on social media. This finding highlights the importance of higher moments of the reward process in creating and sustaining habits.