On how religions could accidentally incite lies and violence: folktales as a cultural transmitter

The study investigates whether and how deviant behaviors—lying and violence—interact with values from Vietnam’s Three Teachings (Confucianism, Buddhism, Taoism) to influence folktale outcomes. Motivated by apparent contradictions in folklore (e.g., violent or deceptive acts sometimes yielding positive ends despite virtuous teachings), the authors analyze Vietnamese folktales to uncover credible statistical patterns. The research question: What outcomes can be expected from interactions between negative behaviors (lie, violence) and the Three Teachings? The work contributes to computational folkloristics by applying rigorous statistical modeling to behavioral patterns in non-WEIRD contexts, addressing gaps in prior literature that often focuses on Western religions and qualitative analyses.

Two main strands are reviewed: (1) Religions and deviance (lying/violence): While religiosity is theorized to promote prosocial behavior and deter deviance (e.g., via cognitive dissonance; Durkheim’s social cohesion), empirical findings are mixed across contexts and traditions, with many studies centered on Western monotheisms and WEIRD countries. Some research finds no or context-contingent effects of religiosity on delinquency, cheating, or aggression. (2) Lies and violence in folklore: Violence is prevalent in global folklore for narrative and social functions; lying is also common though less studied. Studies explore how violent motifs relate to power, identity, and social crises, and how lies function semantically in narratives. A gap remains in understanding interactions between religious teachings and deviant acts in folklore. The paper also contextualizes Vietnam’s Three Teachings: Confucianism (social harmony and moral conduct), Taoism (naturalness, wuwei), and Buddhism (Four Noble Truths, Eightfold Path, karma). None explicitly ban lying; Buddhism’s first precept forbids killing. Vietnamese notions of “religion” (tôn giáo) encompass broad teachings and practices, justifying treatment as cultural-religious values rather than doctrinal systems.

Dataset and coding: 307 Vietnamese folktales were encoded into 345 binary observations. Variables: Lie (main character lies), Viol (main character commits violence), VB/VC/VT (behaviors express values of Buddhism/Confucianism/Taoism), Int1 (supernatural intervention), Int2 (human intervention), Out (positive outcome for main character). Transformed variables include interactions: B_and_Lie = VBLie; C_and_Lie = VCLie; T_and_Lie = VTLie; B_and_Viol = VBViol; C_and_Viol = VCViol; T_and_Viol = VTViol; and Int1_or_Int2 = (Int1+Int2 > 0 ? 1 : 0). Modeling: A Bayesian multilevel varying intercept model (Primary Model) estimates Out as a function of Lie, Viol, their interactions with VB/VC/VT, and an intercept varying by presence/absence of intervention. General formula (Eq. 1): O ~ b_B_and_Viol_OVBViol + b_C_and_Viol_OVCViol + b_T_and_Viol_OVTViol + b_Viol_OViol + b_B_and_Lie_OVBLie + b_C_and_Lie_OVCLie + b_T_and_Lie_OVTLie + b_Lie_OLie + a_Int1_or_Int2[(Int1+Int2>0?1:0)]. Comparators: Basic Model (Eq. 2) with Lie and Viol only; Violence Model (Eq. 3) with Viol and its interactions with VB/VC/VT; Lie Model (Eq. 4) with Lie and its interactions with VB/VC/VT. Estimation: Bayesian MCMC using 4 chains, 5000 iterations, 2000 warm-up. Model checking and comparison: PSIS-LOO cross-validation (Pareto k diagnostics), and model weights via WAIC weights, Pseudo-BMA (with/without Bayesian bootstrap), and Bayesian stacking. Convergence diagnostics: All models achieved Rhat = 1 and n_eff > 1000. Visual diagnostics (trace, autocorrelation, Gelman shrink) provided in supplementary materials. Data and code availability are provided via GitHub and CRAN (bayesvl package).

• General effects: Across models, lying and violence are associated with worse outcomes for main characters (negative posterior distributions for b_Lie_O and b_Viol_O).
• Interactions with religious teachings:
  • Confucianism × lying: Positive association with good outcomes; b_C_and_Lie_O concentrated in the positive range (Primary and Lie Models).
  • Buddhism × violence: Positive association with good outcomes; b_B_and_Viol_O concentrated in the positive range (Primary and Violence Models).
  • Taoism × lying or violence: Ambiguous/mixed effects (coefficients overlap positive and negative ranges).
• Interventions: Presence of intervention (human or supernatural) does not materially change outcomes; distributions of a_Int1_or_Int2[1] vs. a_Int1_or_Int2[2] are similar.
• Model comparison (PSIS-LOO):
  • Basic Model: elpd_loo = -197.5 (SE 8.9); p_loo = 3.1; looic = 394.9; all Pareto k < 0.5.
  • Lie Model: elpd_loo = -199.0 (SE 10.4); p_loo = 8.3; looic = 397.9; 99.7% good Pareto k, 0.3% very bad.
  • Violence Model: elpd_loo = -199.0 (SE 10.4); p_loo = 8.3; looic = 397.9; 99.7% good Pareto k, 0.3% very bad.
  • Primary Model: elpd_loo = -201.6 (SE 11.5); p_loo = 15.0; looic = 403.2; 99.1% good, 0.3% ok, 0.6% very bad Pareto k.
• Model weights (rankings):
  • WAIC weights: Basic 0.531; Lie 0.287; Violence 0.118; Primary 0.063.
  • Pseudo-BMA (no bootstrap): Basic 0.743; Lie 0.173; Violence 0.068; Primary 0.016.
  • Pseudo-BMA+ (bootstrap): Basic 0.384; Lie 0.307; Violence 0.181; Primary 0.128.
  • Bayesian stacking: Lie 0.445; Basic 0.217; Violence 0.216; Primary 0.122.
• Sample: 307 folktales encoded into 345 binary observations.

The findings address the research question by showing that while lying and violence generally lead to negative outcomes in folktales, their effects are context-dependent when embedded in the Three Teachings. Confucian-associated characters who lie are more likely to have positive endings, aligning with Confucian priorities of social harmony and loyalty that may tolerate deception to preserve order. In Buddhist-influenced stories, violent acts can co-occur with positive outcomes, potentially reflecting folk interpretations of karmic justice or the narrative use of violence to highlight Buddhist virtues. Taoist interactions show no clear directional effect. These patterns resonate with broader literature on the rationalization of violence in South Asian traditions and discrepancies between folkloric narratives and real-life moral norms. Methodologically, the study demonstrates how Bayesian multilevel modeling can detect nuanced interaction effects in cultural data. Substantively, it suggests a double standard within culturally transmitted teachings: ends-justify-the-means narratives may normalize or excuse deviant acts under certain valued frameworks. Interventions (human or supernatural) show negligible influence on outcomes, indicating that intrinsic behavior–value interactions dominate narrative resolutions.

Using Bayesian analysis of 307 Vietnamese folktales, the study finds: (1) Lying and violence generally predict negative outcomes; (2) Interactions with religious values alter these effects—Confucianism is associated with favorable outcomes for liars, and Buddhism with favorable outcomes for violent actors; (3) In some parameter ranges, interactions with any of the Three Teachings can yield positive endings. These results illuminate how folklore, as a cultural transmitter, can encode contradictory moral signals, potentially endorsing outcomes where deviant means are justified by valued ends. Future research should: expand datasets across cultures and regions; refine coding schemes (e.g., disaggregating violence types); and integrate additional social-context variables to probe mechanisms behind value–behavior interactions.

• Generalizability: Data are limited to Vietnamese folktales; cross-cultural replication is needed.
• Coding granularity: Behavioral categories (e.g., violence) could be further decomposed (murder, assault, etc.).
• Model complexity and overfitting: The Primary Model risks overfitting due to many parameters, as indicated by lower model weights and some non-ideal Pareto k values.
• Folklore representation: Oral traditions may reflect lay interpretations of teachings rather than doctrinal ideals.