Abstract
Drawing on psychological and literary theory, we investigated whether the warmth and competence of movie protagonists predict IMDb ratings, and whether these effects vary across genres. Using 2,858 films and series from the Movie Scripts Corpus, we identified protagonists via AI-assisted annotation and quantified their warmth and competence with the LLM_annotate package (human-LLM agreement: r = .83). Preregistered Bayesian regression analyses revealed theory-consistent but small associations between both warmth and competence and audience ratings, while genre-specific interactions did not meaningfully improve predictions. Male protagonists were slightly less warm than female protagonists, and movies with male leads received higher ratings on average (an association that was multiple times stronger than the relationships between movie ratings and warmth/competence). These findings suggest that, although audiences tend to favor warm, competent characters, the effects on movie evaluations are modest, indicating that character personality is only one of many factors shaping movie ratings. AI-assisted annotation with LLM_annotate and gpt-4.1-mini proved effective for large-scale analyses but occasionally fell short of manually generated annotations.
Introduction: The Dimensions of Social Perception
Screenwriters and psychologists have long debated the ingredients of a "likable" or "relatable" protagonist. While literary theory often emphasizes complexity and "three-dimensional" characterization, psychological models—specifically the Stereotype Content Model—suggest that human social perception is primarily driven by two dimensions: warmth (perceived intent, such as trustworthiness and friendliness) and competence (perceived capability and intelligence).
Historically, analyzing these traits at scale has been difficult. Manual coding of character personalities across thousands of films is prohibitively time-consuming. This study addresses this challenge by utilizing Large Language Models (LLMs) to quantify character traits across a massive dataset, testing whether these universal dimensions of social perception translate into higher audience evaluations as measured by IMDb ratings.
Datasets and Methodology
The study utilized the Movie Scripts Corpus, a dataset comprising 2,858 movie and series scripts. To process this volume of data, a "human-in-the-loop" AI pipeline was implemented to identify protagonists and annotate their traits.
Protagonists were identified using gpt-4.1-mini, which proved more effective than simple dialogue-counting methods. Character traits were then annotated using the LLM_annotate Python package. The LLM scanned scripts for specific behaviors and rated them on a 6-point scale for warmth and competence. Validation against a human-coded sample of 100 annotations showed high agreement ($r = .83$), confirming the LLM's reliability as a proxy for human judgment. Statistical analysis was conducted using Preregistered Bayesian regression to estimate effect sizes and account for uncertainty.
Results: The Personality Paradox
The analysis confirmed that both warmth and competence are positively associated with IMDb ratings, but the effect sizes were surprisingly small.
A 1 standard deviation (SD) increase in warmth correlated with a mere 0.042 point increase on IMDb’s 10-point scale. Similarly, a 1 SD increase in competence was associated with a 0.055 point increase. As illustrated in Figure 2, the dataset captures a wide range of character types, from the high-competence 'V' to the low-competence 'Beavis and Butt-Head,' and from the high-warmth 'Evelyn Abbott' to the low-warmth 'Jason Voorhees.' Despite these variations, character personality appears to be only a minor driver of overall film enjoyment.
Furthermore, these effects were remarkably stable across genres. Contrary to the expectation that Action fans might prioritize competence or Romance fans might prioritize warmth, the study found no significant genre-specific interactions. The preference for warm and competent protagonists appears to be a "genre-invariant" baseline.
The Gender Gap and Industry Factors
One of the most significant findings was the impact of a protagonist's gender. Movies with male leads received ratings 0.34 points higher on average than those with female leads—an effect size more than six times larger than that of character warmth.
The study also found subtle differences in how genders are portrayed: male protagonists were rated as slightly less warm than female protagonists ($\beta = -0.12$), though they were found to be similarly competent. The author notes that the higher ratings for male-led films may be confounded by industry-wide factors, such as higher production budgets typically allocated to male-led projects.
Discussion and Limitations
While LLMs offer a powerful tool for large-scale narrative analysis, the study highlights several limitations. Current models often process scripts linearly and may miss "narrative reveals"—moments where a character's late-movie actions fundamentally recontextualize their earlier behavior. Additionally, the study used an "average" trait score, which fails to capture "character arcs"—the dynamic growth or change that many theorists consider the hallmark of great storytelling.
Finally, the dataset consists of professionally produced films that likely already meet a baseline of character quality. Future research might find stronger effects by analyzing amateur scripts, where the variance in character competence and warmth is likely much higher.
Conclusion
This research demonstrates that while character personality matters, it is not the primary determinant of a film's success with audiences. The modest impact of warmth and competence suggests that movie ratings are influenced by a complex interplay of plot, production value, and systemic industry biases. As AI tools continue to evolve, they will provide even deeper insights into the mechanics of storytelling and the nuances of audience reception.
Summary updated