Hence, which RL model is most appropriate does matter for de-termining which kind of regression model is most suitable for ana-lyzing a given data set. Ito and Doya (2009) fitted various models to
rat choice behavior. Their results demonstrated that the F-Q model
(αL = αF ) better fit the rats’ choice behaviors compared with the
standard Q-learning model with αF = 0, and the F-Q model had
approximately the same prediction performance as the Q-learning
model with differential forgetting (DF-Q model; αL and αF were in-dependent free parameters). The F-Q model was ultimately favored
because the F-Q model has fewer parameters than the DF-Q model.
Together with our discussion in the present study, Ito and Doya’s
results may suggest that the simple regression model (with an actual trial count) would suffice to describe choice behavior in rats.