Wouter van den Bos and I have published a commentary in Psychological Science showing that – contrary to the claim of an earlier paper by Ericson et al. (2012) – heuristic models of delay discounting do not outperform traditional utility-based models. Through reanalysis of their data, we further demonstrate that modeling delay discounting crucially depends on at least three aspects of the model comparison: The level of analysis (aggregate- or subject-level), the auxiliary assumptions used to map the models to behavior (exponential- vs. power choice rule), and the loss functions used to evaluate the models (e.g., MAD vs. LogLoss).
One Comment
Congratulations, nice work! And very fancy plots!