The psychological, computational, and neural foundations of indebtedness

The paper addresses how and why receiving favors can carry hidden psychological costs by eliciting indebtedness, and how these feelings shape acceptance of help and reciprocity. Building on appraisal theory, the authors propose that beneficiaries infer benefactors' intentions—altruistic (care/communal concern) versus strategic (expectations of repayment)—and that these appraisals generate distinct emotional components: guilt (for burdening the benefactor) and obligation (pressure to repay). Using Psychological Game Theory to formalize belief-dependent preferences, the work seeks to conceptualize and computationally model indebtedness as a superordinate construct comprising guilt and obligation, to test how these components influence behavior, and to identify their neural substrates.

Prior research highlights gift-giving and favors as social signals that foster relationships, often eliciting gratitude and reciprocity. Theories of communal versus exchange relationships suggest that context and relationship norms shape responses to favors. Appraisal theories posit that cognitive evaluations of events generate emotions, which guide behavior. Psychological Game Theory models beliefs about others' intentions and expectations in social decisions, including fairness and reciprocity. Indebtedness has been variably framed as guilt for burdening a benefactor or obligation to repay; both may coexist in real interactions. Neuroimaging literature links vmPFC and default mode regions to positive social valuation and gratitude, insula and prefrontal areas to guilt, and dmPFC/TPJ to theory-of-mind and strategic inference. Together, these literatures motivate a model in which intention inferences yield distinct emotional components that shape reciprocity and help acceptance.

The research comprises three studies.

• Study 1 (online questionnaire, China): 1,619 participants recalled recent events involving accepting or rejecting help. They rated appraisals (perceived care, second-order belief about benefactor's expected repayment), emotions (gratitude, indebtedness, guilt, obligation), and behaviors (need to repay, acceptance/rejection). Participants also defined indebtedness (free-response) and endorsed sources (guilt vs obligation). Analyses included regressions, frequency summaries, and topic modeling (LDA) on emotion-related words (TF-IDF selected top 100; classified via independent raters; best topic number via cross-validated perplexity).
• Study 2 (interpersonal task): Two behavioral cohorts (Study 2a n=51; Study 2b n=57) were combined (n=108). In repeated single-shot rounds, a new anonymous same-gender benefactor supposedly spent from a 20 yuan endowment to reduce the participant's 20 s medium-intensity electrical pain (max reduction 16 s). Benefactor spending was computer-determined. Participants reported second-order belief (how much the benefactor expected in repayment), decided to accept or reject help when allowed, and decided how much to reciprocate from their 25 yuan endowment. Intention manipulation: in each trial, benefactors either knew participants could reciprocate (Repayment possible) or were told repayment was impossible (Repayment impossible); in fact, reciprocation was always available. Study 2b additionally manipulated help efficiency (exchange rate) but showed no interactions with other variables; thus combined analyses. After the task, participants rated trial-wise perceived care, indebtedness, guilt, obligation, and gratitude. Five trials were randomly selected for incentive realization. Analyses: linear mixed-effects for appraisals, emotions, reciprocity, and rejection; factor analysis (EFA) to extract Communal and Obligation factors; mediation analysis testing whether second-order beliefs and perceived care mediated effects of intention information on emotions.
• Computational modeling: Two models within a Psychological Game Theoretic framework predicted behavior by maximizing a utility function combining self-interest (greed parameter), communal concern (guilt and gratitude, via perceived care), and obligation (via second-order beliefs). Reciprocity model (Model 1.1) predicted continuous reciprocation amounts; help-acceptance model (Model 2.1) predicted accept/reject. U_obligation was defined from second-order belief normalized by endowment; U_communal from perceived care derived from benefactor cost and mitigated by inferred strategic expectation; parameter κ modulated how strategic inference reduces perceived altruism; parameter φ weighted communal versus obligation. Model fitting used per-participant SSE minimization for reciprocity and maximum likelihood for acceptance; model comparisons against alternative formulations; parameter recovery.
• Study 3 (fMRI, n=53): Participants completed the same task (no rejection option) during scanning; post-task emotion/appraisal ratings collected after scanning. Univariate model-based fMRI GLMs assessed voxel-wise tracking of trial-by-trial reciprocity, communal concern (from perceived care), and obligation (from second-order belief), using cluster-level FWE correction. Meta-analytic decoding (Neurosynth) interpreted psychological processes. Neural utility model: principal components regression with 5-fold CV trained whole-brain patterns to predict trial-wise communal concern and obligation model terms per participant. Cross-validated neural predictions replaced model inputs to forecast reciprocity; compared with a whole-brain model directly predicting reciprocity. Additional tests assessed alignment between individual brain patterns for reciprocity and communal/obligation and correspondence with behavioral φ.

Study 1 (n=1619):

• Both guilt and obligation independently predicted self-reported indebtedness: β_guilt = 0.70 ± 0.02, 95% CI [0.66, 0.73], t(1988)=40.08, p<0.001; β_obligation = 0.40 ± 0.02, 95% CI [0.36, 0.44], t(1988)=2.31, p=0.021 (FDR-corrected).
• Sources of indebtedness: 91.9% endorsed guilt for burdening benefactor; 39.2% endorsed obligation due to perceived ulterior motives.
• Topic modeling of definitions revealed two topics: Topic 1 (77% variance) with communal-concern-related words (e.g., guilt, gratitude, feel indebted), Topic 2 (23%) with burden/pressure/negative bodily states.
• Predicting need to repay after accepting help (n=1598 events): indebtedness β=0.20±0.04, t=5.60, p<0.001; guilt β=0.12±0.04, t=2.98, p=0.004; obligation β=0.09±0.04, t=2.27, p=0.024; gratitude β=0.38±0.04, t=9.86, p<0.001 (all FDR-corrected).
• Predicting rejecting vs accepting when counterfactually considering acceptance (n=1993 events): gratitude reduced rejection (β=-0.87±0.06, z=-13.65, p<0.001); indebtedness (β=0.23±0.10, z=2.40, p=0.017), guilt (β=0.46±0.09, z=5.06, p<0.001), and obligation (β=0.28±0.06, z=4.70, p<0.001) increased rejection.

Study 2 (n=108): Appraisals and emotions

• Intention manipulation effects (Repayment possible vs impossible): increased second-order beliefs (β=0.53±0.03, t=15.71, p<0.001); decreased perceived care (β=-0.31±0.02, t=-13.89, p<0.001). Effects scaled with benefactor cost: second-order belief β=0.22±0.02, t=13.13, p<0.001; perceived care β=-0.08±0.01, t=-6.64, p<0.001.
• Perceived care negatively associated with second-order beliefs controlling experimental variables: β=-0.44±0.04, t=-11.29, p<0.001.
• Emotions: indebtedness slightly higher in Repayment impossible (β=-0.09±0.03, t=-2.98, p=0.003); obligation higher in Repayment possible (β=0.30±0.03, t=9.28, p<0.001); guilt lower (β=-0.25±0.02, t=-10.30, p<0.001); gratitude lower (β=-0.27±0.02, t=-13.18, p<0.001). Effects increased with benefactor cost (obligation β=0.11±0.01, t=8.85; guilt β=-0.05±0.01, t=-4.28; gratitude β=-0.06±0.01, t=-4.20; all p<0.001).
• Factor analysis: two factors explained 66% variance—Communal (perceived care, guilt, gratitude) and Obligation (second-order belief, obligation); indebtedness loaded on both. Mediation: total indirect effect 0.59±0.04, Z=14.49, p<0.001; second-order belief mediated obligation (indirect 0.22±0.03, Z=7.18, p<0.001); perceived care mediated guilt (0.17±0.01, Z=13.23, p<0.001) and gratitude (0.19±0.01, Z=13.72, p<0.001).

Behavior:

• Reciprocity increased with benefactor cost: β=0.63±0.02, t=25.60, p<0.001; slightly greater in Repayment impossible vs possible (interaction β=-0.03±0.01, t=-2.99, p=0.003).
• Help rejection more likely in Repayment possible (rejection rate 0.37±0.10 vs 0.30±0.03): β=0.27±0.08, z=3.64, p<0.001; higher benefactor cost reduced rejection: β=-0.65±0.13, z=-5.16, p<0.001; no significant interaction with cost (β=0.07±0.07, p=0.279).

Computational models:

• Reciprocity model predicted trial-wise reciprocity well (r=0.79; β=0.88±0.01, t=59.36, p<0.001) and outperformed alternatives; parameters identifiable (mean r for recovery ≈0.94).
• Model-derived appraisals mapped to self-reports: E* predicted second-order belief (β=0.68±0.03, p<0.001); ω predicted perceived care (β=0.72±0.03, p<0.001); ω predicted guilt (β=0.47±0.03, p<0.001) and Communal Factor (β=0.81±0.03, p<0.001); E* predicted obligation (β=0.38±0.03, p<0.001) and Obligation Factor (β=0.64±0.06, p<0.001). κ correlated with φ across participants (r=0.79±0.05, p<0.001).
• Help-acceptance model achieved 80.37% accuracy; better than some alternatives; moderate parameter identifiability.

Study 3 (fMRI, n=53):

• Behavior replicated Study 2; CFA confirmed two-factor model (CFI=0.986, TLI=0.970; RMSEA=0.079, SRMR=0.019).
• Univariate neural correlates: reciprocity tracked bilateral dIPFC, bilateral IPL, precuneus, bilateral ITG; communal concern tracked anterior insula, vmPFC, precuneus, bilateral dIPFC, bilateral ITG; obligation tracked dmPFC and left TPJ (cluster-corrected FWE p<0.05).
• Meta-analytic decoding: reciprocity linked to attention/calculation/memory; communal also to default mode; obligation to social/ToM/memory.

Neural utility model:

• Whole-brain patterns predicted communal concern (mean r=0.21±0.03, p_perm<0.001) and obligation (mean r=0.10±0.03, p_perm=0.004); patterns not spatially correlated (r≈0.03).
• Neural utility model predicted reciprocity using only brain-derived communal/obligation inputs (mean r=0.19±0.02, p_perm<0.001), numerically similar to direct reciprocity prediction model (mean r=0.18±0.03, p_perm<0.001).
• Neural φ strongly correlated with behavioral φ (r=0.88±0.04, p<0.001); subject-specific φ critical (permutation p_perm<0.001). Relative behavioral weighting (1−φ) correlated with greater spatial alignment of obligation vs communal patterns with reciprocity (r=0.68±0.09, p<0.001).

Findings support a conceptualization of indebtedness as a mixed emotion composed of guilt (from perceived altruistic intentions and communal concern) and obligation (from inferred strategic intentions and expectations to repay). Appraisals about benefactors' intentions causally shaped these feelings and, in turn, behaviors: all three feelings (gratitude, guilt, obligation) promoted reciprocity, but guilt and obligation increased help rejection, while gratitude facilitated acceptance. Computational models quantified a trade-off between communal and obligation motives and accurately predicted both reciprocity and acceptance decisions, linking belief-dependent appraisals to action. Neuroimaging dissociated communal concern (vmPFC, insula, dIPFC, precuneus) from obligation (dmPFC, TPJ), aligning with known roles in valuation/affect and mentalizing, respectively. A neural utility model demonstrated that brain-derived representations of communal and obligation utilities suffice to predict individual reciprocity decisions and capture person-specific weightings between these motives. Together, the results clarify how intention inference drives the emotional components of indebtedness and how these components shape social exchange behaviors.

The study advances a psychological and computational framework in which indebtedness comprises guilt and obligation arising from appraisals of altruistic versus strategic intentions. Across a large-scale survey, laboratory interaction task, formal models, and fMRI, the work demonstrates that intention inferences generate distinct emotional components with separable neural substrates and measurable impacts on accepting help and reciprocating. The neural utility approach shows that multivariate brain patterns encode the trade-off between communal concern and obligation and can predict reciprocity at the trial level. Future research should more finely dissociate guilt versus gratitude within communal concern, examine how relationship type and cultural context modulate these components, refine and generalize appraisal-based computational models to diverse help/gift contexts, and explore different help-receiving situations and help-seeking dynamics.

All samples were recruited in China, potentially limiting generalizability given cultural differences in gratitude, guilt, and indebtedness. The computational models simplify appraisal and emotion generation by operationalizing perceived care and second-order beliefs from task variables, which may require adaptation in other contexts. The help-acceptance model showed parameter instability and limited identifiability when disentangling guilt from gratitude. The paradigm did not explicitly separate guilt and gratitude during decision-making, constraining inference about their distinct contributions. Additional contexts (e.g., varied types of help, gift-receiving, help-seeking) remain to be tested.