,
8 tweets,
4 min read
Read on Twitter
Can social media engagement be explained by reward learning? Yes! In our new preprint (w. @david_m_amodio), we use RL modeling to explain the posting behavior of 4000+ individuals on several social media platforms: psyarxiv.com/78mh5 summary follows: (1/7) 
We modeled how people on social media space their posts conditional on their history of likes, based on theory of animal free-operant behavior (by e.g, @yael_niv ). If the reward rate is high, responses should be faster = shorter latency between social media posts (2/7)
We first tested this RL model of social media behavior on a dataset of 800,000 + posts from Instagram. Our RL model with reward learning fit behavior better than models without learning, and the latency between posts was shorter when the average reward(likes) rate was high! (3/7)
We then replicated these findings in 3 smaller (200,000 + posts) datasets from 3 topic based social media forums. These results show that theory for animal behavior in Skinner boxes can explain human behavior on social media, on completely different time scales! (4/7)
Are there clear individual differences in how people learn from rewards (likes) on social media? We used computational phenotyping, based on k-means clustering of model parameters, to answer this question (n = 4,000+ people). We find 4 distinct reward learning profiles (5/7)
Can these differences be explained, in part, by how people compare themselves to others? We modified the RL model to include social comparison, and found this to be the case! A large subgroup use upwards social comparison in reward learning, which changes prediction errors! (6/7)
Together,these findings support the social reinforcement learning view of social media engagement and offer key new insights into this emergent mode of modern human behavior on an unprecedented scale. Lot more in the paper! (7/7)
