Log-likelihood values of the fit self-consistent observer model for every subject (as well as the combined subject Sc), relative to the range defined by the likelihoods of the independent Bayesian observer and a hypothetical, omniscient model (’Data’). The latter can be thought of as the data explaining itself, that is, a model 'defined’ by the empirical probabilities of the data. The log-likelihoods of a random observer (’Chance’) are also given as additional reference. This observer can be thought of as 'being blind’, thus providing random answers in both the discrimination task and the estimation task (sampling from a uniform distribution). The self-consistent observer model is consistently outperforming the independent Bayesian model in explaining the data. Note, the self-consistent and the independent observer model have exactly the same model parameters. Also, a version of the self-consistent observer model that does not include noise in the memory recall of the sensory signal (Stocker and Simoncelli, 2007) generally does not fit the data as well.