And this is why we need error bars on all plots. Looking at these plots there is no way to know whether people guessed uniformly or whether the self assessment is clustered around the mean.