Art films foster theory of mind

The paper proposes that characteristics of films do not only shape spectators’ inferences about on-screen characters’ mental states (as illustrated by the Kuleshov effect) but can also influence spectators’ broader accuracy in inferring mental states in real-world contexts. Building on distinctions in literary studies where literary (vs. popular) fiction enhances ToM, the author hypothesizes an analogous distinction in film: viewing art films (vs. Hollywood films) increases performance on ToM tasks. The study aims to test this causal hypothesis and explore mediators related to perceived character complexity and predictability.

Theory of Mind (ToM) refers to inferring others’ thoughts, intentions, emotions, beliefs, and desires, with affective and cognitive components supported by distinct neural systems. Multiple tasks assess ToM, including socio-perceptual (e.g., RMET) and socio-cognitive measures (e.g., moral judgment tasks). Prior research shows ToM in adults varies and can be enhanced by activities such as second-language learning, acting, meditation, and especially reading fiction. Exposure to literary fiction, as opposed to popular fiction, predicts better ToM, supported by multiple correlational and experimental studies; contrary null results have methodological issues. Literary fiction is characterized by foregrounding and discourse deviations that increase interpretive effort, rendering characters complex and less predictable, thereby eliciting mentalizing. In film theory, a parallel distinction is drawn between Hollywood films (plot-driven, schematic, typified characters, reliance on culturally shared schemas/Theory of Society) and art films (psychologically complex characters, ambiguity, suppressed explanatory cues), which are theorized to demand greater mind-reading. Scholars (e.g., Bordwell, Smith, Vaage) argue art films invite simulation and mentalizing more than Hollywood films that often provide redundant cues and stereotype-based inferences.

Design: Between-subjects experiment with random assignment to watch one of twelve 20-minute film excerpts: six art films vs six Hollywood films. Film served as a random factor in mixed models; film type (art vs. Hollywood) as fixed factor. Sequential and alternative mediation models tested via PROCESS (Model 6), 10,000 bootstrap samples, 95% CIs. Participants: Recruited 326 U.S.-based MTurk participants ($6 compensation). Exclusions: incomplete survey (17), playback problems (54), watched <19 minutes (12), at/below chance on RMET (4), aberrant MJS pattern (3), extreme time on film page >3.5 SD (4). Final N=232 (Art: n=112; Hollywood: n=120). Demographics: 55% female; age M=33.71 (SD=10.58); education: HS 10%, Some College 37%, College 44%, Graduate 9%; ethnicity recoded White (78.8%) vs non-White; majors: Business 16%, Humanities 21%, Natural Science 21%, Social Science 15%, Other 27%. Groups balanced across condition for gender, age, education, ethnicity. A priori power analysis (f=0.20, α=0.05, power=0.80) indicated N≈200; N=232 deemed adequate. IRB approval and informed consent obtained. Stimuli selection: Initial pool from BoxOfficeMojo top worldwide grossers (Hollywood) and Cannes Palme d’Or winners/nominees (art), years 2000–2010. Box office thresholds: art <$50M, Hollywood >$250M worldwide to ensure category separation. Final films selected via discussion with students/experts: Art: Spider (2002), Clean (2006), Certified Copy (2011), Bright Star (2009), All or Nothing (2002), Synecdoche, New York (2008). Hollywood: Monster-in-Law (2005), Fast & Furious (2009), Click (2006), Pirates of the Caribbean (2006), Quantum of Solace (2008), The Dark Knight (2008). First 20 minutes used. Procedure: After consent, participants watched one assigned clip (~20 min) in full-screen with headphones via Qualtrics/Vzaar, then completed: (1) film evaluation (enjoyment, desire to watch, ease of understanding; prior seen yes/no), (2) character assessment (perceived complexity/type/obscurity/consistency; semantic differentials; Big Five items; predictability and confidence in judgments), (3) ToM measures: Moral Judgment Task (abbreviated; 12 scenarios combining intention and outcome) producing Moral Mind and Moral Base indices combined into MJS (higher values indicate stronger intention-based ToM; sample M=-1.72, SD=1.49); RMET (36 items; sample M=26.15, SD=4.50). Playback issue check, demographics. Total time ~45 minutes. Measures and data reduction: Film evaluation composite: enjoyment and desire to watch (r=0.88; M=5.56, SD=1.73), plus ease-of-understanding and seen-before (binary). Character assessment factor analysis (oblique rotation) on items yielded two factors: Predictability (confidence in knowledge; future behavior prediction; loadings >0.80) and Complexity (type, complex, obscure; loadings >0.74). One item (behaves the same across situations) dropped (loadings <0.40). Composite scores: Complexity M=4.20 (SD=1.31); Predictability M=5.36 (SD=1.05). Analysis: Mixed models with Film Type fixed, Film random, for film evaluation, ToM outcomes (RMET, MJS), and mediators (Complexity, Predictability). Correlational checks among measures. PROCESS mediation tested sequential model (Type→Complexity→Predictability→ToM) and alternatives (reversed sequence; parallel mediation). Supplementary covariate analyses included film evaluation, ease-of-understanding, IMDB ratings (film-level), seen-before, gender, major, education.

• Film evaluation and familiarity: Hollywood films evaluated more positively (Art M=5.03 vs Hollywood M=6.01; F=9.55, p=0.01) and as easier to understand (Art M=5.39 vs Hollywood M=6.45; F=11.9, p=0.01). Previously seen: Hollywood 33% vs Art 3.57%.
• Theory of Mind outcomes: Art film condition outperformed Hollywood on both ToM measures: • MJS: Art M=-1.50 vs Hollywood M=-1.94; F=4.88, p=0.03. • RMET: Art M=26.58 vs Hollywood M=25.73; F=4.02, p=0.04. RMET and MJS correlated r=0.24, p<0.001.
• Character perception: • Predictability: Art lower (less predictable) M=5.12 vs Hollywood M=5.56; F=3.69, p=0.05. • Complexity: Art higher M=4.46 vs Hollywood M=3.94; ns (F=1.63, p=0.20). Predictability and Complexity correlated negatively r=-0.17, p=0.008.
• Mediation (sequential Type→Complexity→Predictability→ToM): Significant indirect effects: • For MJS: b=0.004, SE=0.003, 95% CI [0.001, 0.015]. Direct effect: b=0.1255, SE=0.0674, 95% CI [-0.007, 0.258]. • For RMET: b=0.006, SE=0.006, 95% CI [0.001, 0.018]. Direct effect: b=0.063, SE=0.067, 95% CI [-0.069, 0.196]. Reversed sequential model not supported (CIs included zero). Parallel mediation: Predictability mediated (MJS b=0.033, 95% CI [0.006, 0.076]; RMET b=0.042, 95% CI [0.0011, 0.095]); Complexity did not.
• Supplementary analyses: Adding film evaluation as covariate left main ToM effects and mediation largely unchanged; effect on Predictability reduced (F=2.47, p=0.15). Adding ease-of-understanding removed Predictability effect (F=0.38, p=0.53) and eliminated indirect effects in mediation for both ToM measures. IMDB ratings differed marginally by type (p=0.08); correlated with Complexity (r=0.30, p=0.01), not with ToM or Predictability; controlling for IMDB did not change results. Seen-before correlated with Predictability (r=0.16, p=0.01); controlling slightly reduced Predictability effect (to F=2.90, p=0.08) without altering mediation. Gender had no effect. Humanities majors scored highest on RMET; education showed a marginal RMET effect driven by graduate degree group; covarying these did not alter main findings.

Results support the hypothesis that viewing art films enhances immediate ToM performance compared to Hollywood films on both socio-perceptual (RMET) and socio-cognitive (MJS) tasks. Art-film characters were perceived as less predictable, and mediation analyses indicate that perceived character complexity and reduced predictability sequentially account for the ToM advantage, consistent with film theory positing that art cinema withholds explanatory cues and invites deeper mentalizing, whereas Hollywood films rely on schemas and typified characters. The findings were robust across multiple controls (evaluation, IMDB ratings, prior familiarity, demographics), though ease-of-understanding attenuated the mediation path. Methodologically, using multiple stimuli per condition and treating film as a random factor advances media-effects research. The author interprets the effect as a priming of ToM processes by art films, potentially translating into longer-term changes with repeated exposure, analogous to findings in fiction reading. The discussion situates results within broader debates on ToM specificity, the role of Theory of Society heuristics, and cinematic techniques (e.g., POV, shot scale) that may modulate the need for mentalizing versus embodied empathy.

The study demonstrates that brief exposure to art films, relative to Hollywood films, causally improves immediate performance on two established ToM measures. This advantage appears to operate through increased perceptions of character complexity leading to decreased predictability, which prompts greater mentalizing. The work extends parallels between literary vs. popular fiction to film, highlighting how narrative form influences social cognition. The author cautions against inferring long-term skill changes from single exposures and emphasizes that Hollywood films may better serve other social functions (e.g., reinforcing shared social schemas). Future research should: (1) investigate long-term and repeated exposure effects via longitudinal and cross-sectional designs; (2) refine film categorizations and examine structural cinematic features (average shot length, scene ordering, shot scale, POV) as mechanisms; (3) test robustness with tighter control of prior familiarity and alternative analytical strategies; and (4) explore broader social-cognitive outcomes (e.g., attributional complexity, social accuracy, egocentric bias).

• Short-term, single-session exposure; no assessment of durability of effects.
• Online MTurk sample; generalizability may be limited.
• Film selection restricted to 2000–2010 and constrained by box-office thresholds; categorical boundaries may be fuzzy.
• More participants had seen Hollywood films previously; although controlled, residual familiarity effects are possible.
• Art films rated as less understandable; when controlling for understandability, mediation effects were attenuated, suggesting potential confounding.
• Exclusion criteria and response filters, while based on prior work, may influence sample composition.
• Character complexity difference was in expected direction but not statistically significant in primary models.
• Use of 20-minute excerpts may not capture full narrative development.
• Mediation is correlational within the experiment’s structure; causal ordering among mediators, while theoretically motivated, remains debatable.