Incorporating implicit knowledge into the Bayesian model of prior conviction evidence: some reality checks for the theory of comparative propensity