Structure reader

The findings rest on two samples of healthy adults: Study 1 (behaviour only) with 40 participants and Study 2 (fMRI) with 44 participants — a sample size fixed a priori by a G*Power analysis on Study 1's effect size, with four Study 2 participants excluded from the fMRI analysis for excessive head motion.

claim_type assessmentrole scopeconfidence highevidence_verified trueevidence_verified_against slice

evidence “Forty healthy participants (14 male, mean age 26.1, range 22–31) participated in Study 1 (behaviour only study), and 44 healthy participants (19 male, mean (SD) age = 30.6 (6.5), range 23–50) participated in Study 2 (fMRI study).”

notes “Global scope condition bounding every empirical claim; the a-priori sample size and the four-participant head-motion exclusion are folded in as components of it rather than returned as separate claims.”

On each trial participants chose between a safe and a risky monetary option under three conditions: choosing for oneself (Solo), for oneself and the partner (Social), and having the partner choose for both (Partner).

claim_type assessmentrole scopeconfidence highevidence_verified trueevidence_verified_against slice

evidence “There were three kinds of trials: decisions by the participant only for themselves ( Solo condition), decisions by the participant for themselves and the partner ( Social condition), and decisions by the partner for both themselves and the participant ( Partner condition).”

notes “The within-subject responsibility manipulation on which the guilt and agency contrasts depend.”

To hold the partner's behaviour constant across participants, the partner's decisions were simulated by an algorithm that always selected the option with the highest expected value.

claim_type assessmentrole methodologicalconfidence highevidence_verified trueevidence_verified_against slice

evidence “In order to ascertain constant decisions by the partner, the partner’s decisions were simulated using a simple algorithm that always selected the option with the highest expected value”

notes “The partner was not a free agent; partner choices in the Partner condition were deterministic, which the responsibility/guilt contrasts rely on.”

A pre-task icebreaker succeeded in establishing a positive attitude toward the partner: participants rated their partners above 8 on a 1–10 scale for sympathy, cooperativity, honesty, openness and sociability in both studies.

claim_type empiricalrole controlconfidence highevidence_verified trueevidence_verified_against slice

evidence “participants’ average ratings of their partners in terms of sympathy, cooperativity, honesty, openness and sociability were all above 8 on a scale of 1–10, in both studies ( Appendix 1—table 11 ).”

notes “Manipulation check that the social relationship was positive and non-competitive; reported in Appendix 1—table 11.”

Study 2 reproduced the Study 1 design inside the fMRI scanner with identical parameters except for longer inter-stimulus intervals (3–11 s) and partners who were experimenters positioned outside the scanner.

claim_type assessmentrole scopeconfidence highevidence_verified trueevidence_verified_against slice

evidence “In Study 2, participants performed two sessions of the experiment described above inside the fMRI scanner. All parameters were identical except that ISIs varied from 3 to 11 s (drawn randomly from a gamma distribution).”

notes “In Study 2 the partner was experimenter MG or TW rather than another participant, so any replication of the Study 1 guilt effect holds under this changed social pairing.”

Momentary happiness was modelled with five computational models that shared exponentially decaying terms for certain rewards, expected value, and reward prediction errors.

claim_type assessmentrole methodologicalconfidence highevidence_verified trueevidence_verified_against slice

evidence “All models contained separate terms for certain rewards, expected value for lotteries and reward prediction errors, with influences that decayed exponentially over trials.”

notes “The five models are Basic, Inequality, Guilt-envy, Responsibility, and Responsibility Redux; the Basic, Inequality and Guilt-envy models are identical to those in Rutledge et al., 2016.”

Model selection among the happiness models used likelihood ratio tests comparing the Responsibility model pairwise against each other model, supplementing AIC, BIC, R2 and adjusted R2.

claim_type assessmentrole methodologicalconfidence highevidence_verified trueevidence_verified_against slice

evidence “we supplemented the AIC, BIC, R 2 and adjusted R 2 values reported in Table 1 with a series of likelihood ratio tests : we compared pair-wise the likelihoods of the Responsibility model given the data to the likelihoods of all the other models.”

notes “This model comparison licenses treating the best-fitting computational model's variables as the regressors entered into the model-based fMRI GLM (GLM2).”

A parameter-recovery analysis on synthetic data (each participant's estimated parameters, plus 1 SD of noise, refit over 10 repetitions) showed the happiness-model parameters could be reliably recovered.

panel fig3s1claim_type empiricalrole controlconfidence highevidence_verified trueevidence_verified_against slice

evidence “The results show that the estimated parameters could be reliably recovered from noisy synthetic data.”

notes “Validates the stability of the fitted happiness-model parameters; based on a simulation of synthetic data (Figure 3—figure supplement 1), not experimental data.”

The happiness regression with all three two-way interactions (Equation 10) fitted the data significantly better than simpler models without interactions (p < 2e-5) and no worse than the model with all interactions (p > 0.5).

claim_type assessmentrole methodologicalconfidence highevidence_verified trueevidence_verified_against slice

evidence “the model reported in Equation 10 fitted the data significantly better (p < 2e−5) than the simpler models without interactions (higher total and adjusted R 2 , see Appendix 1—table 2 ), but not significantly worse than the model with all interactions (p > 0.5; tested with the ANOVA function in R).”

notes “Warrants reporting the partnerHigh:participantDecided interaction (the guilt effect) from this specific model; details in Appendix 1—table 2.”

A model-based GLM (GLM2) entered the best-fitting computational model's variables—certain rewards (CR), expected value (EV), participant RPE (sRPE), and partner RPE from participant choices (social_pRPE) and from partner choices (partner_pRPE)—as regressors to locate brain regions reflecting them.

claim_type assessmentrole methodologicalconfidence highevidence_verified trueevidence_verified_against slice

evidence “In addition, we created another GLM (GLM2) with regressors designed to identify brain regions whose activation reflected the variables of the best-fitting computational model (see above).”

notes “The neural claims about tracking social_pRPE versus partner_pRPE depend on this model-based GLM being interpretable, which in turn depends on the model comparison above.”