Figure 2.

MFCA and MBCA effects

(A) After a person (e.g., woman) is selected, an outcome’s (e.g., monkey) reward (or lack thereof) reinforces the MF value of this chosen person alone. In contrast, MBCA will occur for that outcome and, consequently, it will affect equally the values of all people (e.g., woman and bearded man) who share that outcome, because the MB value of each of these people is the sum of the values of his (or her) preferred animal and vegetable.

(B) To probe MFCA, we analyzed trials that offered for choice (trial n + 1) the person chosen on the immediately preceding trial n (the repetition person) alongside a person who favored a common outcome, in this example the monkey. We tested the probability of repeating a choice as a function of the previous-trial common reward. The effect of the trial-n common outcome “cancels out” from MB trial n + 1 calculations. Hence, a common reward effect on choice repetition isolates MFCA. For example, when the monkey is rewarded versus not rewarded on trial n, the MF value of the woman will be higher and the probability of repeating the choice will increase.

(C) Trial n’s common reward effect on choice repetition (i.e., the difference in repetition probability between common reward and non-reward) is displayed as a function of a trial n’s format (S, standard; PI, prospective inference; RI, retrospective inference) and the common outcome’s serial position on trial n (first/second). The asterisks correspond to the significance of the common reward main effect in the mixed-effects model, corresponding to the average of all 6 data points (vertical black line).

(D) To test for an MBCA, we analyzed trials that excluded from choice the person chosen on the immediately preceding trial but offered another person (who we term the generalization person; e.g., the bearded man) who shares an outcome in common (e.g., monkey) with the person chosen previously. Note that the other person offered on trial n + 1 (e.g., the child) shares no outcome with the previously chosen person.

(E) Trial n’s common reward effect on choice generalization is displayed as a function of trial n’s format. The asterisks correspond to the significance of the triple interaction (see the X; common reward x format x position) and to the finding that for the second outcome, the common reward effect is stronger in standard versus retrospective inference format (red horizontal line).

Error bars correspond to SEM across participants calculated separately in each format. ^∗p < 0.05, ^∗∗∗p < 0.001. p values were calculated based on mixed-effects logistic regression models. Black diamonds in (C) and (E) designate concealed outcomes. See Figures S1 and S2 for supporting model simulations and Figure S3 for further analyses.