How social relationships shape moral wrongness judgments

The study investigates how relational context shapes moral wrongness judgments. Most moral psychology focuses on judgments involving strangers, but everyday moral evaluations typically concern familiar others embedded in specific relationships. The authors argue that different relationships are normatively expected to serve distinct cooperative functions (care, reciprocity, hierarchy, mating), and that neglecting the functions most prescribed for a given relationship will be judged as more morally wrong. They set three aims: (i) to map relational norms across a broad set of common dyadic relationships in a U.S. context; (ii) to test whether these relational norm profiles predict out-of-sample moral wrongness judgments for function-weakening actions within specific relationships; and (iii) to compare the relational norms model to alternative accounts based on genetic relatedness, social closeness, and interdependence. The authors predict that relational norms will robustly predict moral wrongness judgments and will explain more variance than the alternative models.

Prior work highlights the importance of relational context for moral judgment and behavior. Studies have variously characterized relationships by genetic relatedness (e.g., predicting helping obligations), by social closeness and interdependence (e.g., apathy vs harm judgments toward proximal vs distant others), and by single predominant cooperative functions (e.g., care for siblings, hierarchy for teacher-student). These approaches show that identical actions can be judged differently across relationships, but lack a systematic, multi-functional, data-driven account of the cooperative functions that characterize common dyads. Building on Bugental's framework, and related taxonomies of cooperation and moral motives, the present work focuses on four dyadic cooperative functions—care, reciprocity, hierarchy, and mating—proposing that relationships carry prescriptive norms about which functions they should serve and to what degree. The authors also position their model against traditional predictors such as genetic relatedness, social closeness, and interdependence drawn from relationship science.

The research comprised three pre-registered stages. Stage 1 (Sample 1) measured relational norms. Participants: 423 U.S. adults (nationally representative for age, race, gender) recruited via Prolific, after excluding 70 based on pre-registered criteria. Procedure: Participants received definitions of five cooperative functions (care, reciprocity, hierarchy, mating, coalition) and passed comprehension checks. They rated 20 common dyadic relationships on how much each ideally should serve each function on a -100 to +100 scale. Each participant provided ratings for all relationship-function combinations (100 data points), followed by demographics. Analyses: Coalition was excluded in main analyses. Mixed-effects models examined gender and demographic effects; representational analyses used Kolmogorov–Smirnov (K–S) distance to quantify relational norm dissimilarity across dyads and hierarchical clustering (farthest-point/Voorhees method) to identify clusters. Functional polarization and specificity were quantified; 10 relationships with distinctive norm profiles were selected for Stage 2 using an RSA-like K–S distance procedure to reduce redundancy (retaining both mother- and father-child under-18 relationships for theory-driven reasons in selection). Stage 2 comprised stimulus development and moral judgment testing. Stimuli development: Fifteen trained judges rated 86 action statements of the form “Person A does X to Person B” on characteristic weakening/strengthening of each function (-100 to +100), ignoring morality per se. Interrater reliability was high (ICC(3,k)=.97). An algorithm selected 12 function-weakening actions (3 per function) from among the most characteristic for weakening each function while balancing average extremity for comparability to future strengthening items. Target specificity (main effect minus mean side effects) was computed per action. Moral judgment study (Sample 2): 1,320 MTurk participants (after excluding 502 per pre-registered criteria) were randomly assigned to one of 10 selected relationships and rated moral wrongness (0–100) of the 12 actions tailored to that relationship; they also rated perceived action likelihood to control for conventionality/uncommonness. Analyses: For each participant, mean wrongness was computed for each function-weakening category. Linear mixed-effects regression predicted moral wrongness from Sample 1 relational norms, action likelihood, and target specificity, with participants as highest-level grouping and crossed random effects for relationship dyad and function-violation type. Additional analysis computed between-dyad dissimilarities in moral judgment space using K–S distances averaged across functions and correlated these with relational norm K–S distances (Spearman correlation). Stage 3 (Sample 3) collected alternative predictors. Participants: 85 MTurk participants (after exclusions from 149 recruited). Procedure: Rated the extent to which a well-functioning instance of each of the 10 relationships is characterized by social closeness (three items: deep understanding, acceptance/validation, caring for well-being) and interdependence (frequency, strength, multiplicity of influence), 0–100 scales. Genetic relatedness was coded objectively by dyad. Analyses: Linear mixed models compared predictive power of relational norms vs social closeness, interdependence, and genetic relatedness; model fit assessed via marginal R^2 and AIC. Robustness checks and full details are provided in Supplement; all materials and data are on OSF.

• Relational norms vary markedly across dyads. Reciprocity was generally prescribed (mean across dyads ≈ 54.23), while mating was broadly proscribed (mean ≈ -63.02) except for romantic partners (mating M≈95.12) and friends-with-benefits (M≈58.43). Agreement was highest for mating and care, lower for reciprocity and hierarchy. Some dyads were functionally polarized (e.g., parent–under-18 child: strong care, strong anti-mating), others less so (strangers). Some dyads were functionally specific (e.g., roommates strongly reciprocity; boss–employee strongly hierarchy), while romantic partners were pluralistic (high care, mating, reciprocity; low hierarchy).
• Gender differences: Women reported stronger average expectations for care than men (p<.001), especially for roommates/housemates, customer–seller, teacher–student, neighbor, colleague/classmate. Men reported stronger average expectations for mating than women (p<.001), notably for friends-with-benefits, roommates/housemates, acquaintances, close friends, colleagues/classmates, strangers, neighbors.
• Hierarchical clustering of dyads (using K–S dissimilarities in norm space) revealed four intuitive clusters: sexual (romantic partners, friends-with-benefits), hierarchical/authority-asymmetric (parents of minors, teacher–student, boss–employee), reciprocal/equal-status transactional (customer–seller, roommates, strangers, etc.), and familial/caring (siblings, extended family, parents–adult children, etc.).
• Predicting moral wrongness (Sample 2): Relational norms from Sample 1 significantly predicted moral wrongness judgments (p<.001), accounting for 63% of the variance in mean wrongness ratings. Target specificity positively correlated with wrongness (p<.001), and action likelihood negatively correlated (rarer actions judged more wrong; p<.001). A model with action likelihood alone explained far less variance (marginal R^2=.08; AIC=136,496.9) than a model with relational norms alone (marginal R^2=.30; AIC=130,804). When combined, the relational norms beta (16.26) was >80× the action-likelihood beta (-0.20).
• Dyad-by-dyad correspondence: Dissimilarity between dyads in relational norm space predicted dissimilarity in moral judgment space (Spearman r=.43, p=.003). Function-specific correlations were significant for care (r=.50, p<.001), mating (r=.69, p<.001), and hierarchy (r=.29, p=.05), but not for reciprocity (r=.10, p=.49).
• Alternative models: Social closeness (p=.11), interdependence (p=.14), and genetic relatedness (p=.78) did not significantly predict mean moral wrongness judgments, whereas relational norms remained significant (p<.001) controlling for these. Model fit favored relational norms (marginal R^2=.69, AIC=841.36) over social closeness (R^2=.44, AIC=908.04), interdependence (R^2=.44, AIC=908.30), and genetic relatedness (R^2=.44, AIC=910.33).

Findings support the central claim that moral wrongness judgments depend on relational context understood in terms of prescribed cooperative functions. People hold structured expectations about which functions different relationships should serve, and violations that neglect those functions are judged more wrong within the relationships where those functions are more prescribed. Dyads with similar relational norm profiles exhibit more similar patterns of moral wrongness judgments, indicating an underlying structure of relational obligations shaping moral cognition. The relational norms model substantially outperforms alternative accounts based on genetic relatedness, social closeness, or interdependence, suggesting that multi-dimensional functional expectations capture key normative content missing from one-dimensional relational metrics. An unexpected null emerged for reciprocity: relationship-specific reciprocity prescriptions did not significantly predict wrongness for reciprocity violations. The authors propose that collapsing across two forms of reciprocity may obscure effects: transactional, tit-for-tat reciprocity characteristic of exchange relationships versus mutual, need-based reciprocity within communal relationships. Distinguishing these forms may resolve the discrepancy. The results encourage integrating relationship science with moral psychology, moving beyond stranger-centric dilemmas to everyday relational contexts and norms.

This work systematically maps relational norms for common dyads and demonstrates that these norms predict moral wrongness judgments across relationships, outperforming models based on genetic relatedness, social closeness, and interdependence. The study reveals clustered structures of relational expectations and shows that moral judgment similarity tracks relational norm similarity. The contributions include a data-driven, multi-functional account of relational context in moral cognition and validated tools for predicting moral evaluations from relational norms. Future research should: (i) differentiate forms of reciprocity (transactional vs communal) in prediction; (ii) extend to judgments of rightness, praiseworthiness, and supererogation for function-strengthening actions; (iii) examine individual differences (e.g., impartial beneficence) and their relation to uniformity of relational prescriptions; and (iv) test cross-cultural generalizability and cultural tightness of relational norms.

The studies focus on U.S. samples; cultural variation in relational norms and moral judgments may limit generalizability. Only moral wrongness judgments for function-weakening actions were examined; praiseworthiness and function-strengthening actions were not tested here. The reciprocity construct may have been too coarse, conflating transactional and communal forms. Target characteristics beyond relational role and gender (e.g., race, religion, politics, age, social class) and observer individual differences (e.g., moral reasoning styles) were not comprehensively modeled and may interact with relational context. Sample 3 had a relatively high exclusion rate, though robustness checks suggested stable results. Actions were intentionally everyday and mild; effects for extreme actions were not assessed.