Over­com­ing Bias looks at two research papers on over­con­fi­dence, con­clud­ing that we “are more over­con­fi­dent on tasks we don’t actu­ally expect to per­form, and when we don’t expect to have to explain our eval­u­a­tion to others”.

On per­for­mance:

Par­tic­i­pants made pre­dic­tions about per­for­mance on tasks that they did or did not expect to com­plete.  In three exper­i­ments, par­tic­i­pants in task-unexpected con­di­tions were unre­al­is­ti­cally opti­mistic: They over­es­ti­mated how well they would per­form, often by a large mar­gin, and their pre­dic­tions were not cor­re­lated with their per­for­mance. By con­trast, par­tic­i­pants assigned to task-expected con­di­tions made pre­dic­tions that were not only less opti­mistic but strik­ingly accurate.