Opinion Formation on Social Media Platform

In this paper, we formulate a hierarchical Bayesian learning model to investigate the dynamics of individual opinion formation on a social media platform where individuals are exposed to two sources of social influence, preceding peer posts in a microblog thread and comments from the general public, which are often endogenously generated and thus correlated. The focus of this study is to separate these two distinct sources and measure their respective effects. We take the advantage of a unique natural experiment contained in the data from a leading microblogging website in China to help model identification. Further, we adapt the Consensus Markov Chain Monte Carlo algorithm, a parallel computing approach, to our Bayesian estimation procedure to effectively handle the big data in this study. We find that preceding peer posts in a microblog thread influence an individual to converge to the opinions of her peers. It is observed that more comments from the general public make an individual less certain about the topic discussed in the thread, whereas preceding microblogs from closely connected peers reinforce her opinion. Simulation studies are conducted to show that restricting number of comments in the early stage of a social media campaign helps to build a consumer base that is more certain about their opinions, and identifying more influential users with favorable opinions to participate early positively shifts the average opinion.