Fig. 5: Computational models of indebtedness.

a Participants’ reciprocity behavior in each trial plotted as a function of information about benefactor’s intention (Repayment impossible vs. Repayment possible) and benefactor’s cost. b Participants’ decisions to accept or reject help in each trial plotted as a logit function of information about benefactor’s intention and benefactor’s cost. For a and b, data are presented as mean values +/− SEM. SEMs were generated via bootstrapping that respected the repeated measurements within each participant (n = 12 trials for each condition in Study 2a/14 trials for each condition in Study 2b, 108 participants). c The observed amounts of reciprocity after receiving help and predictions generated by the reciprocity model at each level of the benefactor’s cost in Repayment impossible and Repayment possible conditions. d The observed rates of rejecting help and predictions generated by the help-acceptance model in Repayment impossible and Repayment possible conditions. For c and d, data are presented as mean values +/− SEM (n = 108 participants). e Model simulations for predicted reciprocity behavior in Repayment impossible and Repayment possible conditions at different parameterizations. The y-axis shows the average values of the predicted amount of reciprocity across all levels of benefactor’s cost. The model predicted reciprocity changes as a function of the tradeoff between communal and obligation feelings based on ϕ and interacts with the intention inference parameter κ. Increased emphasis on obligation corresponds to increased reciprocity to favors in the Repayment possible condition, but decreased reciprocity in the Repayment impossible condition; this effect is amplified as κ increases. f Best fitting parameter estimates of the computational model for reciprocity decisions for each participant (n = 108 and n = 53 participants for behavioral and fMRI studies, respectively).