Social learning among opinion leaders

I study information aggregation among opinion leaders, who have access to private information, observe their predecessor's move, and care about both taking a correct action and maximizing the measure of followers. Potential followers are uninformed agents who care solely about the truth and choose to agree with either the current opinion leader or their predecessor. I find that the distinct incentive of an opinion leader to disagree with their predecessor, although encouraging every individual opinion leader to rely more on their private signal, actually exacerbates herding asymptotically in equilibrium. The learning patterns remain robust and lead to a number of practical implications in richer strategic environments. For instance, social learning becomes less precise when opinion leaders observe more predecessors; informational crowding-out could emerge between opinion leaders and agents if the latter became privately informed; etc.