Table 1. Model comparisons between the hybrid model and its special cases.

	Hybrid	Q-learning	Relative advantage
Negative log likelihood	5421	5506	5558
Aggregate LRT favouring hybrid	—	X240=170***	X220=274***
No. of favouring hybrids	—	13	8
Pseudo R2	0.608	0.602	0.597

Shown for each model: negative log likelihood; test statistic and P-value for a likelihood ratio test against the hybrid (full) model, aggregated across subjects; the number of subjects favoring the hybrid model on a likelihood ratio test (P<0.05); and the degree to which the model explained the choice data averaged over the individual fits (pseudo R²). ***P<1E-16.