Reduced confidence in situations of increased observational learning

Humans can learn about the outcomes of available choice alternatives via their own choices, but also by observing others choosing. As agents learn, they should become more confident in knowing the best alternative. However, here we show that such confidence is lower in situations where observational learning typically happens, compared to contexts of predominant individual learning. Our first experiment let participants decide between learning by choosing between reward alternatives and by observing another agent’s choices. Introducing contexts with frequent reward loss made participants both observe more and report lower confidence in knowing which alternative delivered higher reward, despite more frequent choices of this better alternative. To disentangle whether reduced confidence was due to frequent losses or to more observation, we conducted two follow-up experiments where reward and choice agency were manipulated independently. Both factors were found to independently reduce confidence. Our results illustrate how confidence based on the alternatives’ value is reduced by the negative outcomes that often accompany observation, and by an agency component derived from fewer choices being made, thus offering a multifactorial account of the formation of confidence during observational learning.