中文摘要
考虑一个社会，它面临关于某一与收益相关状态的不确定性，并希望培养学生做出正确决策。在教育机构中，学生向教师学习，但也会获得外部信息，并随后向同伴学习。我们表明，私人贝叶斯行动未必是社会学习的最优投入：当学生的行动反映了在同伴之间相关的教师侧信息时，观察许多此类行动可能会赋予该信息过高的社会权重。因此，社会规划者可能最优地降低教学的精确度，诱导学生在其行动成为他人信号之前更多依赖外部信息；尽管学生从教师那里学到的更少，最终却可能变得更为知情。当同伴互动具有同质性时，设置精确度上限的理由更强，因为同一教师的信息更有可能在聚合中留存下来；而当外部信息本身存在系统性扭曲时，该理由则更弱。这为将观点中立性作为一种制度政策提供了信息层面的依据：当学生主要向接触相似信息来源的同伴学习时，它限制了相关教师侧信息在社会层面的过度代表。
Abstract
Consider a society that faces uncertainty about a payoff-relevant state and wants to train students to make correct decisions. In educational institutions, students learn from their teachers, but they also get outside information, and later learn from peers. We show that privately Bayesian actions need not be optimal inputs into social learning: when students' actions reflect teacher-side information that is correlated across peers, observing many such actions can give this information excessive social weight. A social planner may therefore optimally reduce the precision of instruction, inducing students to rely more on outside information before their actions become signals for others, and students can end up better informed despite learning less from their teachers. The case for a precision cap is stronger when peer interaction is homophilous, because same-teacher information is more likely to survive aggregation, and weaker when outside information is itself systematically distorted. This provides an informational rationale for viewpoint neutrality as an institutional policy: it limits the social overrepresentation of correlated teacher-side information when students mostly learn from peers exposed to similar sources.