More precise regression analyses found that all three predicted relationships by the model were present in the data: skill level, learning rate and their interaction all predicted participant deliberation in the predicted directions.

We asked whether their willingness to deliberate could be predicted by their levels of improvement before.
Initial fits to average behavior indicated that subjects who learned more were more willing to deliberate longer in a subsequent similar task.

We designed a 2 block dot motion experiment to test these theoretical predictions.

The model predicted that as opposed to a greedy model (current reward rate maximization, standard normative objective), a forward looking model would assign control based on skill level, learning rate and their interaction.

The model had a control component (with some objective, e.g. maximize total cumulative reward), a processing component (to make decisions, e.g. a DDM), and a learning component.