LAUSR

Mechanistic hypotheses about psychiatric disorders are increasingly formalized as computational models. Model parameters, characterizing for example decision-making biases, are hypothesized to correlate with clinical constructs. This is promising (Moutoussis et al., 2016), but here we comment that techniques used in the literature tominimize noise in parameter estimationmay not be helpful. In addition, we point out related pitfalls whichmay lead to questionable research practices (Sijtsma, 2015). We advocate incorporating cross-domain, e.g., psychopathology-cognition relationships into the parameter inference itself. Maximum-likelihood techniques often provide noisy parameter estimates, in the sense of total error over an experimental group. In addition, in large studies brief tasks are often used, providing little data per participant. However, individual parameter estimation can be improved by using empirical priors (Efron, 2012). Here, parameter estimates are informed by, or conditioned upon, the population distribution that a case comes from. For an individual j with parameters θ j coming from a population distribution ppop: