Fig. 6.
Reaction time analysis. Reaction times (RTs) plotted against the quadratic expansion of decision variable (a), relative chosen state beliefs (b, z-scored within participants) and relative chosen normalised utilities (c, z-scored within participants). Error bars are s.e.m. across participants (N = 22). These analyses of RTs, which constitute a dataset independent of both fMRI activations and choice data used for model fitting validate the approach used to investigate fMRI. Indeed, our first regression analysis of fMRI activations assumes that decision-making is more demanding when the decision variable becomes more ambiguous, thereby predicting RTs to vary as a quadratic function of decision variable (i.e., with −[(1−ω)(v1−v2)+ω(B1−B2)]2). Consistently, RTs decreased with quadratic regressor [(1−ω)(v1−v2)+ω(B1−B2)]2 (a: T(21)=6.6; p<0.0001). Our second regression analysis of fMRI activations further assumes that reaching a decision requires less processing resources, when the evidence supporting actual choices increases. Thus, model MIX predicts RTs to exhibit two additive decreasing effects: when given normalised utilities, relative chosen affective beliefs (Bchosen–Bunchosen) increase and, when given affective beliefs, relative chosen normalised utilities (vchosen–vunchosen) increase. We therefore entered RTs in a multiple linear regression analysis including these two regressors and their interaction as within-subject factors. As predicted, RTs decreased when (Bchosen – Bunchosen) or, to a lesser extent, (vchosen–vunchosen) increased, with no interactions between these factors (b, c; Bchosen–Bunchosen: T(21) = 4.9, p < 0.0001; vchosen–vunchosen: T(21) = 2.94, p = 0.0078; interaction: F < 1; difference in linear effects: T(21) = 2.5, p < 0.05,). As in fMRI analyses, we then included quadratic regressors (Bchosen–Bunchosen)2 and (vchosen–vunchosen)2 as within-subject factors in this analysis. The quadratic regressors presumably capture the encoding of decision-relevant information irrespective of choice computations and, are consequently predicted to have no influences on RTs. The results confirmed the additive decreasing effects of linear regressors (Bchosen–Bunchosen) and (vchosen–vunchosen) (T(21) = 5.2, p < 0.0001 and T(21) = 2.3, p = 0.03, respectively) with no significant effects of quadratic regressors (both Ts(21) < 1.16, ps > 0.26). In summary, RTs confirmed that affective state beliefs and normalised utilities contributed to decision-making independently with affective state beliefs dominating normalised utilities