Engineering Fear
Protagonist Vulnerability and the Evolutionary Design of Horror
What is it about some stories and situations that make them more effective at evoking fear? One way to answer this is to reverse engineer the emotion of fear.
Our ancestors would have come across a number of hostile people and animals throughout their lives. Quickly computing the odds of survival in the event of a conflict would have been an extremely valuable cognitive ability. The score for each side’s ability to win an all out fight has been conceptualized as formidability and has been successfully applied to understand emotions like anger.
In a new paper, Edgar Dubourg and I argue that formidability may also help us understand the emotion of fear. Perceiving an adversary whose formidability far outmatches your own should trigger a computational appraisal of poor survival odds, perhaps felt as vulnerability. Fear should be one output of that appraisal, helping to coordinate defensive attention, arousal, and avoidance. Given this, stories that present a formidability asymmetry between the hero and villain such that the villain is powerful and the hero is weaker and more vulnerable should be better at evoking fear.
Horror fiction is the genre most closely associated with the emotion of fear — so much so, that the ability and intent to scare the audience is often found in definitions of the genre. If our hypothesis about formidability asymmetry and fear is correct, then the basic format of a horror story should be a powerful antagonist facing a vulnerable protagonist.
According to this hypothesis, it’s not only a weak protagonist or only a powerful antagonist that makes a story horrifying. Rather, it is the overall vulnerability of the protagonist. In Alien, (1979), a nearly perfect extraterrestrial predator stalking a commercial spaceship crew produces sustained fear because the crew is unprepared, poorly armed, and trapped, with limited information and no reliable escape.
Predator (1987) presents a similarly powerful and predatory antagonist stalking a group of people. However, this group of people are an elite paramilitary team led by a character who is portrayed by former Mr. Universe, Arnold Schwarzenegger. As a result of the powerful protagonists, Predator feels much more like an action movie than a horror movie. For the same reason, superhero movies that feature powerful antagonists and loads of violence typically feel more like action rather than horror. Likewise, many films with low-formidability protagonists (e.g., Kevin from Home Alone) can end up in the comedy realm when the antagonists are also low formidability.
To test this hypothesis at scale, we relied on an automated annotation method that leverages the latent cultural knowledge of large language models. Rather than manually coding films one by one, we used a structured prompting procedure to have a language model estimate key narrative features for each movie, such as the formidability of both the protagonist and antagonist, the persistence of the threat, and the hostility of the environment.
Analyzing 691 films spanning ten broad film genres including horror, action, thriller, fantasy, and romance, we found a clear structural partition. Horror films were uniquely characterized by weak protagonists, strong antagonists, persistent threats, and hostile environments. Action, fantasy and thriller films, by contrast, often featured formidable antagonists but paired them with capable or well-equipped protagonists, resulting in much lower protagonist vulnerability.
In other words, what separates horror from other genres is not the presence of danger, but the systematic imbalance between who threatens and who must endure the threat.
This raised a natural follow-up question: does protagonist vulnerability predict fear even outside the horror genre? To address this, we turned to viewer-generated data.
Among films not labeled as horror, some are nonetheless widely described as frightening. Using IMDb’s keyword system, we compared non-horror movies that were tagged with fear-related terms to those that were not. The result was strikingly consistent: non-horror films associated with fear keywords showed significantly higher vulnerability scores. This suggests that the same structural ingredients that define horror also predict fear at a finer narrative level, beyond genre labels themselves.
Finally, we asked whether protagonist vulnerability tracks not just perceived fear, but also its associated physiological responses. Using data from Science of Scare project that measured viewers’ heart rates while watching horror films, we found that horror movies with higher vulnerability scores reliably elicited stronger increases in heart rate.
In short, narratives that place weak protagonists in the path of overwhelming threats engage the body’s threat-management systems more intensely. All these tests held even when accounting for the film’s release year and remained robust across alternative ways of computing protagonist vulnerability.
These findings, reported in Evolution and Human Behavior, show that protagonist vulnerability captures a core input condition of fear in movies across genre boundaries, subjective judgments, and physiological responses. More generally, this perspective helps explain why certain narrative tropes recur so reliably in horror fiction.
Characters in horror films are isolated, cut off from communication, or trapped in hostile environments because isolation removes social and material resources that would otherwise reduce vulnerability. Settings such as remote cabins, deserted hotels, confined spaces, or unfamiliar worlds systematically limit escape and amplify exposure to harm. Even when protagonists are adults, horror often strips them of weapons, authority, or knowledge, placing them in situations where resistance is ineffective and outcomes feel uncontrollable. From this perspective, many classic ingredients of horror such as darkness, confinement, isolation, helplessness, are efficient ways of engineering vulnerability.
More broadly, these results shed light on how the fear system is activated. They suggest that fear is not triggered simply by the detection of threat-related cues, but by an assessment of whether harm is plausible given the situation and capacities of the individual facing the threat. Horror fiction is particularly well suited to revealing these input conditions because it systematically exaggerates and isolates them.
By holding constant the nature of the threat while varying the vulnerability of protagonists, fictional narratives create controlled contrasts that would be difficult or unethical to study in real life. More generally, cultural products that are explicitly designed to evoke specific emotions provide a powerful and efficient way to study the input conditions of evolved psychological mechanisms: when they succeed across audiences and contexts, they offer converging evidence about how those mechanisms are tuned to the structure of the world.
This is an adapted version of a blog post that Edgar Dubourg and I wrote for the Human Behavior & Evolution Society Blog.