Engaging the Dark Side: Exploring Identification with Morally Complex Antagonists

ABSTRACT

This work investigates what factors prompt audience members to engage with morally complex antagonists and examines potential real-world attitudinal outcomes of such engagement. A two-part experimental study (n = 140) examines whether the revelation timing of a character’s immoral behavior in a fictional narrative (late reveal versus early reveal) and a person’s morality salience (vice salience versus virtue salience) impact identification with a morally complex antagonist. Further, real-world attitudinal outcomes of antagonist identification are examined. Results demonstrated that identification with a fictional, morally complex antagonist can significantly impact real-world attitudes toward criminals. Additionally, revelation timing predicted identification with a fictional antagonist for male participants, but not for female participants, indicating a need for future research to investigate potential gender differences for identification with morally complex antagonists. A person’s own morality salience, however, did not significantly impact identification with a fictional antagonist. Overall, this work demonstrates that exposure to fictional narratives and identification with morally complex antagonists can affect real-world attitudes, and gives direction for future research to better understand the mechanisms that drive engagement with antagonists in narratives.

Introduction

The inimitable Mark Twain once said, “Everyone is a moon, and has a dark side which he never shows to anybody.” Twain’s words remind us of the complexity of human nature—possessing the capacity for darkness and light, evil and goodness. In the field of narrative communication, researchers have been captivated by the moral complexity of story characters and have sought to understand the ways in which story consumers engage with such complexity (e.g., Grizzard et al., 2018; Janicke & Raney, 2015; Krakowiak & Tsay-Vogel, 2015; Raney, 2004; Shafer & Raney, 2012; Tamborini et al., 2018). Yet this research has focused almost entirely on engagement with protagonists—main characters pursuing their goals at the center of the story arc. Antagonists—characters who oppose the goals of the protagonist(s) (Magliano et al., 2005)—are largely missing from this scholarship. To understand the nuances of human engagement with the breadth of characters featured in narrative, this gap in morally complex character research must be addressed. Thus, this work seeks to address conceptual ambiguity regarding morally complex character types and to extend the range of existing theory (Slater & Gleason, 2012) to examine engagement with morally complex antagonists.

Drawing upon current research, we first conceptualize and define the character type we refer to as morally complex antagonist. We next identify and test two key factors that may contribute to engagement with morally complex antagonists: (1) the order in which character details are presented and (2) the moral self-concept of audience members. We test the impact of these experimentally manipulated factors on identification with an antagonist and resulting persuasive outcomes. Further, in a post-hoc analysis, we examine the potential relevance of a third theoretically relevant predictive factor: audience gender. Finally, we provide recommendations for refining and expanding this important line of research. Overall, the present study demonstrates that identification with morally complex antagonists occurs, that it influences real-world moral judgments, and that it may feature unique predictors and outcomes as compared with engagement with protagonists.

Defining morally complex characters

The popularity of stories with less-than-moral main characters—such as Macbeth, Breaking Bad, and The Godfather—indicates that audiences can engage with immoral main characters and enjoy their triumphs. Researchers have studied this phenomenon and sought to identify what factors \enable people to enjoy such narratives and engage with such characters (e.g., Eden et al., 2017; Janicke & Raney, 2015; Krakowiak & Tsay-Vogel, 2015; Meier & Neubaum, 2019; Shafer & Raney, 2012). Though researchers often refer to these characters as anti-heroes, morally complex characters, or morally ambiguous characters, Tamborini et al. (2018) point out that conceptual clarity is lacking, noting that “a recent examination of literature in this area uncovers almost no formal definitions, but substantial discrepancies in how these types of characters are conceptualized” (p. 844). To add clarity, Tamborini and colleagues propose five distinct categories into which characters can be organized based on their morality: perfect heroes, imperfect heroes, morally equivocal characters, imperfect villains, and perfect villains. As indicated by their titles, these character types range from never violating common moral domains (perfect hero) to always violating common moral domains (perfect villain), with gradations in between.

Although Tamborini et al.’s (2018) typology provides important clarity to the description of characters based on the morality of their actions, this categorization scheme does not differentiate the various roles characters may play in the story arc, such as protagonist (main character driving the action) and antagonist (opposing character). For example, the authors provide the example of Hannibal Lecter (in Silence of the Lambs) as an imperfect villain due to Lecter’s overridingly immoral behavior, despite the fact that Lecter generally helps the protagonist (Clarice) catch the serial killer rather than opposing her efforts. Other efforts to categorize different character types have similarly relied on differences in morality and/or positive character traits to distinguish between categories (e.g., Eden et al., 2017; Kinsella et al., 2015). Yet theorists have noted that a character’s role in the story arc may have implications for character engagement, even apart from a character’s morality (Raney, 2004). In this vein, Magliano et al. (2005) conceptualize protagonist versus antagonist distinctions as separate from character morality, noting, “Protagonist–antagonist distinction hinges upon goal conflict; these characters desire mutually exclusive states of being” (p. 1358). The authors further note that in some genres (i.e., action-adventure), morality distinctions between protagonists and antagonists are more typical and clear-cut than in other genres. Fuzzier distinctions can be observed, for example, in epic dramas or tragedies, such as Victor Hugo’s classic Les Miserables. Though few would dispute the labeling of Jean Val Jean as the protagonist (main character) and Inspector Javert as the antagonist (character opposing the protagonist’s goals), morality distinctions between the two characters are far more complex. The two characters each display strong moral commitments to different moral priorities: Val Jean sacrifices all to protect others (mercy/care), while Javert devotes his life to the application of the law (justice/fairness). Thus, a clear protagonist versus antagonist distinction may be observed even where a clear moral versus immoral distinction cannot.

We seek to address factors that may predict engagement with antagonists in narratives, as defined above by their role in the story as opposing the goals of the protagonist (Magliano et al., 2005). Thus, we specifically seek to better understand engagement with morally complex antagonists, which we define as characters opposing the protagonist(s)’s goals and displaying patterns of both moral and immoral behavior. Further, we seek to investigate the outcomes of engagement with morally complex antagonists to better understand both enjoyment of media entertainment and potential real-world attitudinal effects of engaging with such characters. We next review relevant theorizing on engagement, and then propose factors that may encourage engagement with morally complex antagonists and potential outcomes of such engagement.

Engagement with morally complex characters

Character categorizations and moral evaluation

Scholars investigating audience engagement with morally complex characters argue that initial impressions influence an audience’s dispositions toward a character as the story unfolds. Affective disposition theory (ADT; Zillmann, 2000; Zillmann & Cantor, 1976) predicts that audiences determine which characters we like based on our ongoing moral evaluations of characters’ actions. Generally, we like characters we judge to be moral and dislike characters we judge to be immoral. Yet some popular characters are known to commit objectively immoral actions. Accordingly, Raney (2004) argues in an extension of ADT that humans develop knowledge structures—called schemas—from our exposure to common story constructions and character archetypes. When we encounter a character, we form an initial interpretation of the type of character that is based on our preexisting schema. For example, if we initially categorize a character as the story hero, we are likely to apply a heroic schema to the character, based on our previous experience with story heroes. Because we view the character as a hero, we expect the character to generally behave morally, and thus we interpret the character’s behavior through this positive lens. Consistent with Raney’s predictions, experimental research has demonstrated that initially interpreting a character as a hero biases later moral judgments of the character’s actions, such that moral violations are judged to be more acceptable when a character perceived to be a hero commits them (Matthews, 2019; Shafer & Raney, 2012; see also Kleemans et al., 2017).

Positive moral judgments of characters are associated with increased positive character engagement, including liking (Grizzard, Huang et al., 2020; Zillmann, 2000) and identification (Tsay & Krakowiak, 2011). Cohen (2009) describes identification as involving “an intense, if temporary, merging of the self with a media character” (p. 229) and conceptualizes the phenomenon as including empathy, perspective-taking, and character understanding. Audiences are motivated to identify with characters to enjoy the narrative experience (Slater et al., 2014). Further, researchers have proposed that in the context of morally complex characters, identification may facilitate narrative enjoyment and attitudinal change (Meier & Neubaum, 2019; Oliver et al., 2019; Raney et al., 2009; Tsay & Krakowiak, 2011). Because identification involves a merging of the self with the character, audiences who identify with a morally complex character may be particularly motivated to interpret the character’s actions in a favorable light. Thus, Oliver et al. (2019) and Raney et al. (2009) argue that identification with a character is likely to encourage moral biases in favor of the character, promoting disengagement from moral standards that might typically condemn the character’s actions (i.e., moral disengagement; see Raney, 2004).¹

Though theory regarding engagement with morally complex characters has advanced considerably in recent years, research has focused nearly exclusively on the effects of schema-based moral bias formation toward engagement with morally complex protagonists, noting that people show a positive bias in their interpretations of the immoral actions of characters perceived as heroes/anti-heroes. Yet the observed popularity of morally complex antagonists who oppose the protagonist’s goals in entertainment narratives—to include The Avengers’ Loki and Hannibal’s Hannibal Lecter—suggests that positive engagement with antagonists does occur. Though such examples indicate that people can positively engage with antagonists, questions regarding how we are able to do so and the effects of such engagement remain unanswered. To explore these questions, we next propose two factors that are likely to promote positive engagement with morally complex antagonists.

Applying initial categorization logic to morally complex antagonists

Although negative schema-based biases toward perceived villains are discussed theoretically (Raney, 2004; Shafer & Raney, 2012), predictions regarding morally complex antagonists who do not fit the classic villain schema are largely missing. It is unclear from existing schema theorizing how a character who is introduced as opposing the hero’s goals, yet who is not initially presented as clearly immoral/evil, should be categorized and processed by an audience. However, Sanders’ (2010) character impression formation (CIF) model offers general predictions that could be meaningfully applied to morally complex antagonists.

Building on Raney’s protagonist-centered impression-formation work described above, Sanders (2010) indicates that the general logic of the effects of initial impression formation applies to all character judgments. Moving beyond the discussion of hero/anti-hero/villain schema alone, Sanders proposes a broader model (CIF) by which people form initial impressions of characters and then evaluate future character behavior through the lens of the that initial categorization. She notes, “the first piece(s) of information, such as a character label, presented about a character will greatly influence the resulting impression, prompting viewers either to reinterpret or to ignore information that is inconsistent with the initial label” (Sanders, 2010, p. 152).

Under the CIF, initial information presented about a character causes the audience to categorize the character using salient and accessible categories. After categorization, audiences assess whether ongoing behavior of the character is consistent with their own assumptions/views regarding the category. For example, if a viewer initially perceives a character to be in the category “criminal,” and the viewer’s schema surrounding criminals assumes that criminals are immoral and uncaring, a later scene in which the character shows kindness to a stranger may strike the viewer as inconsistent with the initial categorization. When this occurs, the CIF predicts that the viewer could engage in one of several potential inconsistency resolution processes, including re-interpreting the inconsistent behavior to be consistent with the original category (e.g., “the criminal probably has a hidden, self-centered motive for helping the stranger”). Other resolution processes include ignoring the inconsistent information or disregarding the initial categorization and using the new information to build a new impression. Overall, the CIF predicts that initial information about a character guides the interpretation of future character behaviors, often biasing the viewer to interpret the character’s actions in a way that is consistent with the original categorization.

Consistent with Sanders’ model, studies by Kaufman and Libby (2012) tested whether revealing negatively perceived information about characters early in a story as opposed to late in a story affects audience’s willingness to identify with characters and their attitudes in response to characters. Their work used stimuli in which the target character was categorized as a member of a different group than the participants. It was understood that participants might be less willing to identify with an outgroup character. The authors found that audiences were more likely to identify with a character when outgroup membership was revealed late in the story after the reader had already formed an initial perception of the character that did not include the group categorization. Further, audiences in the late revelation condition reported less negative attitudes toward real-world individuals that matched the character’s category.

We propose that early revelation of negatively perceived behaviors of a character in a story discourages positive engagement with that character, whereas late revelation promotes more positive engagement. Importantly—unlike arguments regarding hero story schema—this logic is not dependent on a character’s role in the story arc; rather, it denotes that the timing of the revelation of morally-relevant information about a character biases engagement even apart from schema regarding story roles (protagonist or antagonist). We further note that although Kaufman and Libby (2012) tested these effects in the context of group membership, Sanders’ (2010) theorizing indicates that any type of label applied to a character early in a story contributes to initial categorization of a character and can bias future character perceptions. Thus, we offer the following hypothesis:

H1: People will more strongly identify with an antagonist whose immoral behavior is revealed late in the story as compared to early in the story.

Morality salience and morally complex antagonists

Whereas the revelation timing of negative behaviors constitutes a text element that may influence engagement with an antagonist, differences in the psychological conditions of the readers/viewers of a narrative may also prove impactful. Particularly, a person’s own moral self-concept may influence how they engage with morally complex characters. The concept of morality salience refers to a person’s moral self-perceptions (Krakowiak & Tsay-Vogel, 2015, 2018)—the extent to which one feels morally good or bad about oneself. Krakowiak and Tsay-Vogel (2015) argue that in the real world, morality salience is not a static trait; rather, people’s positive or negative views of their own morality likely change over time and are subject to the influences of both internal differences and external situations. As such, past research has experimentally manipulated morality salience, and findings indicate that the moral self-perceptions that a viewer holds at a given time while encountering a narrative text significantly influence how viewers respond to morally complex characters (Krakowiak & Tsay-Vogel, 2015, 2018).

It should be noted that morality salience as conceptualized by Krakowiak and Tsay-Vogel (2015, 2018) refers to a person’s overall moral self-perception and is a concept distinct from salience of moral intuitions. Salience of moral intuitions, a component of Tamborini’s model of intuitive morality and exemplars (MIME; Tamborini, 2012; Tamborini et al., 2013; see also Eden et al., 2014), refers to the relative weight of importance of a person’s key moral intuitions. Though these concepts are related, Krakowiak and Tsay-Vogel conceptualize morality salience as unconstrained to any specific moral domain, whereas Tamborini focuses on differences in the salience of five moral domains (care, fairness, loyalty, authority, and purity; see Haidt, 2007).

A person’s morality salience is thought to impact character engagement and narrative enjoyment through upward and downward social comparison, making certain characters more engaging than others depending upon whether one’s own vices or virtues have been made more noticeable. Consistent with social comparison theory (Festinger, 1954), individuals with negative self-perceptions may feel better about themselves when they compare themselves to characters they deem immoral. Research by Krakowiak and Tsay-Vogel (2015) demonstrated that people whose vices were made salient (inducing negative self-perceptions) enjoyed narratives about morally complex characters more than did individuals whose virtues were made salient. Further, because perceived similarity is typically highly correlated with identification, individuals whose vices are more salient may identify more strongly with a morally complex antagonist because they perceive themselves to be morally similar to the character. Krakowiak and Tsay-Vogel (2015) suggest this possibility, pointing out that further research “could examine whether morality salience affects feelings of similarity and identification with different character types” (p. 406). Such an impact suggests that naturally occurring differences in moral self-concept may influence character engagement, and may even influence which narratives people select.

This theoretical explanation should logically apply to all character roles, not simply protagonists. Thus, to test this potential relationship, we propose a second hypothesis:

H2: People whose vices are made salient will more strongly identify with a morally complex antagonist than will people whose virtues are made salient.

Real-world outcomes of antagonist identification

Although scholars have long argued that the potential societal effects of engagement with morally complex characters need to be investigated (e.g., Raney, 2004), this potential has not been fully explored. Yet theorizing from narrative persuasion and entertainment-education indicates that identification with morally complex characters could affect real world attitudes. Identification is considered a key mechanism of narrative persuasion (De Graaf et al., 2012; Slater & Rouner, 2002), even when a narrative is not explicitly persuasion-oriented in its content (Moyer-Gusé, 2008). Entertainment-education scholars note that identification with characters enables character-consistent attitudes to occur by reducing people’s tendencies to argue against messages that might typically be perceived as threatening to their way of thinking (Slater & Rouner, 2002). Further, the entertainment overcoming resistance model (EORM; Moyer-Gusé, 2008) posits that identifying with a narrative character can increase perceptions of vulnerability. If a character is encountering certain threats, feelings, or experiences, identification can create a sense of vulnerability in a person exposed to the narrative, opening the person to character-consistent changes in their real-world attitudes and judgments. Consistent with this theorizing, empathy with fictional characters has been shown to increase support for public policies that would benefit real-world individuals similar to such characters (Knobloch-Westerwick et al., 2020; see also Frazer et al., 2021).

Consistent with these theoretical arguments, Kaufman and Libby (2012) found powerful results of character experience-taking (conceptually the same as identification) with characters who are members of an outgroup. For White participants, experience-taking with a Black character led to significant decreases in covert racist attitudes. Similarly, for heterosexual participants, experience-taking with a gay character decreased real-world negative attitudes toward gay individuals. Though the context of outgroup characters is not the same as our proposed context of morally complex antagonists, we note that immoral behavior is similarly likely to lead to categorization of a character (see Sanders, 2010, discussed earlier). Thus, Kaufmann and Libby’s work, as well as the breadth of narrative persuasion theorizing discussed above, suggests the potential for effects of character identification on real-world moral attitudes/judgments. We thus propose our third hypothesis:

H3: Identification with a morally complex antagonist who commits clearly immoral acts will be positively associated with favorable attitudes toward real-world persons who commit similar immoral acts.

Though the connection between character identification and a variety of real-world attitudes is already well-established, we maintain the importance of testing this relationship in the novel context of morally complex antagonists. Even if it seems logical that identification with a morally complex antagonist in a narrative world would have similar effects on real-world judgments as engagement with other types of characters, this should not be assumed without further investigation. It is plausible, for example, that identification with a morally complex antagonist is a qualitatively different process than identification with a protagonist, such that narrative consumers may be willing to take the perspective of an antagonist temporarily to engage with a narrative or expand their experience (see Slater et al., 2014), but that this experience may lead to feelings of guilt or shame, not approval or leniency. Thus, the attitudinal outcomes of identification with antagonists are worthy of empirical investigation. To this end, we next introduce an experimental study designed to test the three hypotheses we have presented.

Method

Overview

Adult undergraduate students at a large Midwestern university took part in a two-part, 2 × 2 (early versus late reveal; vice versus virtue salience) experimental design.²

Participants

After exclusions for incomplete participation and failure of attention and manipulation checks, the final n = 140. The average age of participants was 20.46 years (SD = 2.70), with 70.0% (n = 98) identifying as female and 30.0% (n = 42) identifying as male. More details on exclusion criteria and demographics are available in our online supplement.

Procedure

Baseline measures (time 1)

Participants completed baseline measures of policy attitudes and general criminal identification in a short online survey at least one week prior to the main session.

Main session (time 2)

One week after completing the baseline measures, participants came to a lab to complete the main session. Participants sat at individual computers and were first randomly assigned to complete a vice/virtue salience manipulation. They then were randomly assigned to read a short story text that featured either the late reveal of immoral behavior of the antagonist or early reveal of the same information.³ After reading the text, participants completed identification measures for both the antagonist and the protagonist, as well as suspense measures. Participants then completed repeated policy attitude and general criminal identification measures from session one. Participants also completed manipulation checks, sentencing recommendation measures, an attention check, crime experience measures, and demographic measures. Finally, participants were thanked and debriefed.

Vice/virtue salience manipulation

Vice/virtue salience was manipulated using open writing prompts instructing participants to reflect on three of their past virtuous (virtue salience condition) or non-virtuous (vice salience condition) actions and write about them for ten minutes. This manipulation was adapted from Krakowiak and Tsay-Vogel (2015, 2018) who found it to be a successful method of vice/virtue salience manipulation.

Stimuli

Content

The stimuli consisted of two versions of a short story (length 1846 words for early reveal, 1850 words for late reveal) featuring a full story arc with a clear protagonist and clear antagonist. The story was developed specifically for this study to allow clear manipulation and to avoid any potential confounds of prior narrative exposure (see online supplement for full texts). The author was a graduate student with creative writing training and experience. In preparing the story, the chief goal was to create a compelling narrative that clearly communicates (1) that a certain character (Peter) is the antagonist and that (2) Peter displays patterns of both moral and immoral behaviors.

The story features two candidates for a congressional seat in an unspecified state. Though the story is written in third person, it follows the perspective and internal feelings of the protagonist, Jordan, as he prepares for and participates in a debate against the antagonist, Peter. Neither candidate’s political party is indicated. Though Peter is clearly opposing Jordan and is portrayed as mysterious and potentially shady, he is also a somewhat sympathetic character in that he is grieving the death of his daughter in a drunk driving accident and hopes to propose helpful legislation on the topic of drunk driving. However, near the time of the debate, Jordan’s campaign team discovers clear evidence that Peter has intentionally used insider trading to fund his campaign, a clear criminal offense which, in the words of Jordan’s campaign manager, “Could send [Peter] to jail for half a decade.” Peter and Jordan have a sharp discussion after the debate, in which Peter shares that his daughter’s death is a motivation for his run, Jordan confronts Peter, and Peter ultimately concedes to Jordan, asking him to take his drunk driving legislation to the House floor.

Revelation timing manipulation

The revelation of Peter’s clearly illegal behavior appears near the beginning of the story in the early reveal condition, but is moved near the end of the story in the late reveal condition. Otherwise, the story texts are identical, providing a clear manipulation of early reveal versus late reveal of the antagonist’s immoral behavior.

Measures

Manipulation checks

For this study, it was important that the stimulus clearly reflected a protagonist and an antagonist. Thus, two manipulation checks were included to determine whether participants generally understood the story characters in the way intended. The first manipulation check asked “Which character would you identify as the ‘main character’ (protagonist) in the story you read?” Four options were provided to participants with the order randomized: “Jordan (the young, new candidate),” “Peter (the older, experienced candidate),” “Brooke (Jordan’s campaign manager),” and “Other.” The second manipulation check asked “Which character would you identify as the ‘opposing character’ (antagonist) in the story you read, as compared to the main character?” and featured the same four options as the first manipulation check.

Identification

Identification was measured using Cohen’s (2001) ten-item scale, measured on a seven-point Likert scale. However, the two absorption measures of the scale were excluded from analysis because they are not character specific. The remaining eight items were measured both for the antagonist (Peter: α = .88, M = 3.98, SD = 1.09) and the protagonist (Jordan: α = .93, M = 5.01, SD = 1.16). The overall identification scores were calculated as an average across the eight measures.

General criminal identification

General criminal identification was assessed to determine how much participants generally understand and take the perspective of real criminals in society. To this end, three items were adapted from Cohen’s (2001) identification scale and applied to criminals in general. These items included, “I tend to understand the reasons why criminals in America do what they do,” “I feel like I know exactly what criminals go through,” and “I feel like I can really get inside criminals’ heads.” These items were then embedded in nine distractor items, which were the same items applied to three other groups of people (lawyers, educators, and politicians). The average of the three criminal items were used to calculate a general criminal identification score (for session one: α = .80, M = 2.70, SD = 1.27; session two: α = .90, M = 2.76, SD = 1.44). We also calculated a change score for general criminal identification using the session one and session two measures (session two – session one) (M = .06, SD = 1.17).

Policy attitudes

Policy attitudes regarding consequences for crime were measured using two relevant items (“More Rehabilitation for Prisoners” and “Higher Prison Sentences for Economic Crime”) embedded in five distractor items (e.g., “Increasing the Minimum Wage”). Participants recorded their attitudes using a slider scale of 1 to 100, with “strongly oppose” and “strongly support” as anchors. The two relevant items were analyzed separately, given that the rehabilitation item was only distally related to the experimental story while the economic crime item was closely related. The attitude measures were completed in both session one (rehabilitation: M = 70.53, SD = 18.31; economic crime: M = 52.65, SD = 17.48) and session two (rehabilitation: M = 70.27, SD = 19.88; economic crime: M = 59.02, SD = 18.88). We also calculated a change score for the economic crime item (“Higher Sentences for Economic Crime”), which was the measure most closely related to the stimulus. For this economic crime item, the change score (session two – session one) measured whether participants became more punitive or less punitive toward economic crime after stimulus exposure (M = 6.30, SD = 15.94), with higher values indicating more punitiveness. We reverse coded this measure in our reporting of the models in the results section, so that all outcome measures could be reported in a consistent direction (more lenient).

Sentencing scenarios

Three sentencing scenarios were developed for this study to give participants the opportunity to directly decide what appropriate consequences economic criminals in the real world should incur. Each participant read three short sentencing scenarios adapted from the United States Sentencing Commission’s practice scenarios (see the online supplement for additional details). Participants read each of the scenarios and were asked to assign an appropriate penalty between zero and 40 years in prison using a bar scale (bank robbery: M = 13.91, SD = 7.14; mail and wire fraud: M = 11.06, SD = 6.47; tax evasion M = 8.17, SD = 5.78). When combining the three scenarios, the overall average sentence recommended for a criminal was 11.05 years (α = .76, SD = 5.32). This average was used in all analyses throughout this study.

Results

Manipulation checks

Consistent with the study’s intention, of the 173 participants considered prior to exclusions for failing the antagonist manipulation check, 80.9% (n = 140) of participants identified Peter as the antagonist. Small minorities viewed Brooke (10.4%, n = 18) and Jordan (5.8%, n = 10) as the antagonist. Further, five participants (2.9%, n = 5) selected “Other.” Thus, overall, a strong majority of participants identified Peter as the antagonist, supporting the intended manipulation. However, since the antagonist manipulation was crucial to the theoretical basis of our study, performing analyses on participants who did not view Peter as the antagonist would cloud our ability to draw conclusions regarding identification with an antagonist as a unique process. Thus, we included only participants who identified Peter as the antagonist in our analysis (n = 140). Of these participants, 97.9% identified Jordan as the protagonist, which was consistent with the general intent of our stimuli, although the protagonist is not the focus of our research here.

Testing the hypothesized model

H1 predicted that a late reveal of criminal behavior would increase identification with the antagonist, and H2 predicted that vice salience would increase identification with the antagonist. H3 predicted that identification with an antagonist would be positively associated with real world attitudes toward morally deviant persons. Taken together, the hypotheses were tested with a structural equation model, featuring observed variables and using maximum likelihood estimation (AMOS 26). The real-world attitude measures served as three distinct outcome variables: change in general criminal identification (T2-T1), change in policy attitudes (T2-T1), and sentencing scenarios with the three sentencing scenarios averaged to create a mean sentencing scenario variable.

We assessed model fit using three criteria: 1) nonsignificant χ² goodness-of-fit statistic, 2) a comparative fit index (CFI) of .90 or greater, and 3) a root mean square error of approximation (RMSEA) equal to or less than .06 (Hu & Bentler, 1999). Using these criteria, the hypothesized model (presented in Figure 1) was an acceptable fit to our data (χ² ₍₁₂₎ = 16.35, p = .18, CFI = .973, RMSEA = .05). However, a closer examination of correlations revealed that several predicted pathways were not correlated, contrary to our predictions. Most significantly, the predicted relationship between vice salience and antagonist identification (H2), as well as the predicted relationship between revelation timing and antagonist identification (H1), were not supported. Thus, although the model fit criteria showed good fit, our experimental manipulations failed to significantly impact identification with the antagonist in the manner predicted in H1 and H2. However, significant relationships between identification with the antagonist and two of our real-world outcome measures (policy attitudes and general criminal identification) show partial support for H3, although the change in policy attitudes was in the opposite direction than predicted.

Figure 1. Hypothesized model.

Post-hoc analysis

Due to the lack of significance of our two experimental manipulations in our hypothesized model, we considered whether further analyses might provide helpful insight regarding our theoretical predictions, albeit with the increased risk these results would reflect Type 1 error. Thus, we tested a second, revised model with two modifications.

First, we removed vice salience from our model due to the lack of evidence of a significant correlation between our vice/virtue salience manipulation and antagonist identification, as well as concerns regarding the applicability of our vice/virtue salience manipulation. Though our method followed the strategy of Krakowiak and Tsay-Vogel (2015, 2018) in manipulating vice salience generally, other theoretical frameworks have noted that the salience of particular domains of morality (e.g., care, fairness, loyalty, authority, and purity; see Haidt, 2007) may be more important than salience of morality/immorality in general when it comes to narrative engagement. Tamborini’s model of intuitive morality and exemplars (MIME; Tamborini, 2012; Tamborini et al., 2013), for example, predicts that moral judgments of characters are dependent upon persons’ differing levels of salience of the five different moral domains presented by moral foundation theory (Haidt, 2007). Our stimulus features a target character (the morally complex antagonist, Peter) who cheats the system and breaks the law (violating fairness) to try to win retribution for his daughter (family loyalty). Thus, in retrospect, the lack of specificity of a certain moral domain in our vice salience manipulation may have impeded the manipulation’s effects. Though no formal content analysis was performed, an informal examination of the participants’ responses to our vice/virtue salience prompt revealed that the vice salience manipulation caused participants to largely reflect on relatively minor “vices,” such as eating their roommates’ food or missing class, which do not necessarily match the domains violated by the antagonist. Given the lack of observed effect of the manipulation and these retrospective theoretical considerations, we removed vice/virtue salience from the model in our post-hoc analysis.

Second, we added gender to the model as a potential moderator of the relationship between revelation timing and antagonist identification. Considerations of the relevance of domain-specific salience on the interpretation of our narrative introduce the possibility of a gender difference in interpretation of the story. Research on moral foundations has uncovered robust gender difference in the relative importance of the five different moral foundations to a person’s moral judgment-making. In their moral foundations surveys of more than 118,000 participants, Graham et al. (2011) reported that women rate fairness as significantly more relevant in moral judgment than men, whereas men rate ingroup loyalty as slightly more relevant than women.

Our stimuli feature Peter (the antagonist) committing a clear fairness violation in order to win retribution for his daughter, upholding family loyalty. Thus, participants in the late revelation condition encounter a fairness violation and upholding of loyalty together right at the end of the story, priming participants to reflect on these moral domains as they evaluate the character. Because men and women are observed to value fairness and loyalty differently, we consider it plausible that—given the themes of our stimulus—men and women might react differently in their character judgments, particularly in the late revelation condition when moral domains have been very recently primed. We further considered a significant gender difference possible because the original Kaufman and Libby (2012) studies, which tested the effects of the timing of the revelation of outgroup status on experience-taking, featured gender limitations. Kaufmann and Libby tested two different character groups in their experiments: gay men and minority race. The experiment featuring gay characters used only male participants. Thus, after removing morality salience from the model and adding gender as a moderator, we assessed model fit using the same three criteria described above.

The revised, post-hoc model (presented in Figure 2) showed good fit to our data (χ² ₍₁₂₎ = 13.46, p = .34, CFI = .996, RMSEA = .03) and improved fit as compared to our hypothesized model. The interaction of gender and revelation timing significantly impacted antagonist identification such that, among male participants (n = 42), late revelation of the antagonist’s criminal behavior led to greater antagonist identification as compared to the early reveal condition (see plot in online supplement). Further, antagonist identification predicted increased identification with criminals from time 1 to time 2. A significant relationship was also observed between antagonist identification and changes in leniency of policy attitudes regarding punishment for economic crime from time 1 to time 2. Finally, no relationship emerged between antagonist identification and recommended sentencing in the scenario measures.

Figure 2. Revised model.

Summary of support for hypotheses

Overall, although both models meet the criteria for acceptable fit, the revised model features consistently predictive pathways and offers additional information about the moderating role of gender. Though the initial model did not show support for H1, our post-hoc model, which included gender as a moderator, revealed partial support for H1. Late revelation of immoral behavior predicted identification with an antagonist (β = .63; p = .05), and the interaction between late revelation and gender predicted identification with an antagonist (β = .69; p = .04), such that men who read the late revelation story were significantly more likely to identify with the antagonist than men who read the early revelation story. Revelation timing did not significantly impact antagonist identification for women (see online supplement for a plot of this relationship). H2, which hypothesized the effects of vice salience on antagonist identification, was not supported in our hypothesized model (β = .11, p = .37) and was eliminated from our post-hoc model. H3, which predicted that identification with a morally complex antagonist would be positively associated with favorable attitudes toward similar real-world persons, was tested identically in both models and was partially supported. Identification with a morally complex antagonist resulted in a significant increase in general criminal identification from time 1 to time 2 (β = .24, p = .003) as well as change in leniency of policy attitudes regarding punishment for economic crime (β = −.27, p = .001), but in the opposite direction of H3’s prediction (antagonist identification resulted in more punitive attitudes). Further, sentencing recommendations (in response to scenarios) were not significantly correlated with antagonist identification (β = −.07, p = .44).

Discussion

This experiment begins an overdue investigation into engagement with morally complex characters who play the role of the antagonist in a story. Through this work, we set out to clarify conceptual ambiguity regarding morally complex characters and to extend existing theory to examine the predictive factors and outcomes of engagement with morally complex antagonists. Though media entertainment researchers have focused on engagement with morally complex protagonists up to this point, our results provide evidence that identification with morally complex antagonists occurs and can lead to significant, real-world attitudinal outcomes. Further, although our manipulation of predictive factors provided mixed results, our post-hoc analysis clarifies directions for continued investigation into what narrative attributes and individual characteristics may encourage engagement with morally complex antagonists.

Antagonist identification and real-world outcomes

The observation of two types of significant real-world attitudinal outcomes of antagonist identification in this study (in partial support of H3) demonstrate the relevance of identification with morally complex antagonists. Importantly, our experimental design measured these significant attitudinal outcomes (general criminal identification and policy attitudes) at two time points: before narrative exposure and after narrative exposure. Antagonist identification in response to the narrative resulted in significant changes in these attitudinal outcomes between the first measurement and second measurement points, indicating a causal relationship between antagonist identification and attitude change. Though prior work (Grizzard et al., 2018; Matthews, 2019; Shafer & Raney, 2012) shows compelling evidence that heroic schema enhance positive character engagement and judgments, the results of this study indicate that positive character engagement (in this case, identification) with immoral characters can also occur absent heroic/protagonist schema. Participants were able to identify with a criminal character whom they identified as an antagonist, and this identification affected real-world attitudes, including increased identification with real-world criminals and support for harsher sentencing for economic crime.

Importantly, the fact that stronger identification with the antagonist led to increased identification with criminals in general demonstrates that antagonist identification can carry into the real world, affecting the way humans relate to real-world individuals. The fact that identifying with the story antagonist (a criminal) caused both (1) increased identification with criminals in general and (2) increased support for harsher penalties for economic criminals is somewhat counter-intuitive, suggesting that identification with an antagonist may involve unique processes. One possible explanation for this finding is that participants who identified with the criminal antagonist also felt guilt or shame on the antagonist’s behalf, and thus felt that they were deserving of punishment, leading them to support a more punitive policy attitude. However, further research is needed to investigate the role that moral emotions could be playing in this relationship. Overall, the fact that identification with a criminal antagonist in a brief, fictional narrative caused a significant shift in real-world attitudes regarding criminals points to the potential societal impact of morally complex character engagement and the need for more robust investigations of real-world attitudinal outcomes.

Predictive factors of antagonist identification

We originally addressed two factors that may encourage engagement with morally complex antagonists: revelation timing and morality salience. Importantly, we proposed these predictors because both have been supported in past literature as conducive to engagement with morally complex characters (e.g., Kaufman & Libby, 2012; Krakowiak & Tsay-Vogel, 2015, 2018; Sanders, 2010), but unlike the heroic schema explanation, neither factor relies on the target character being a protagonist. Results of our tests of these two factors were mixed. For revelation timing, our hypothesized model showed no significant effect of revelation timing on antagonist identification. However, the post-hoc addition of gender as a moderator of revelation timing (due to differences in moral domain salience between men and women observed by Graham et al., 2011) showed that late revelation of immoral behavior was a significant predictor of identification for men, but not for women. For morality salience, our hypothesized model showed no effect of vice/virtue salience on antagonist identification, despite the use of a manipulation that had previously proven successful in related research endeavors (Krakowiak & Tsay-Vogel, 2015, 2018).

The fact that neither revelation timing nor vice salience clearly predicted antagonist identification, combined with the post-hoc considerations discussed in our results section, raises important considerations for future research. Both the failure of vice salience to induce changes in antagonist identification and the gender differences observed in the relationship between late revelation and antagonist identification point to the potential influence of varying salience of humans’ moral intuitions on engagement with morally complex antagonists. Although past research (Krakowiak & Tsay-Vogel, 2015, 2018) has shown that manipulation of morality salience (and thus moral self-concept) in general can have significant effects on character engagement, this relationship did not emerge in our context of a morally complex antagonist. However, other research has indicated that the relative salience of different moral domains (i.e., care, fairness, loyalty, authority, and purity) for an individual audience member can profoundly affect how they engage with characters (Eden et al., 2014; Tamborini, 2012; Tamborini et al., 2013). We did not directly test such a relationship in this study.

Our post-hoc analysis revealed that gender and late revelation interacted in a manner consistent with observed differences in the salience of moral intuitions between men and women (Graham et al., 2011), with past research indicating that men value in-group loyalty more highly (a value consistent with the antagonist’s moral values) and women value fairness more highly (a value contrary to the antagonist’s moral values). This finding may also indicate that recency of exposure to a moral violation (for women, the violation of fairness) may negatively impact character evaluations given shortly afterward, raising questions as to whether the stigma-reducing nature of late revelation of negatively perceived information proposed by Kaufmann and Libby might in some cases be negated by the recency/priming effect in the context of moral violations.

Taken together, the findings regarding both of our manipulations (late revelation and vice salience) raise important questions that need to be investigated in future research in the context of morally complex antagonists. First, how might differences in audience members’ salience of moral intuitions and in the moral domain priorities displayed by the antagonist affect character engagement? Future researchers would be wise to measure both moral intuition salience for participants (see the Moral Foundations Questionnaire; Graham et al., 2011) and moral perceptions of characters for each of the five moral foundations (see the Extended Character Morality Questionnaire; Grizzard, Fitzgerald et al., 2020). Given the potential relevance of moral domain salience (Tamborini et al., 2013), we also note that future research should test these relationships using narratives with varying emphases on different moral domains.

Our results raise a second important question. Are effects of revelation timing a product of the power of an initial positive disposition toward a character that can withstand moral/norm violations once it is formed (see Raney, 2004; Sanders, 2010)? Or is revelation timing creating a recency/priming effect for moral violations revealed near the end of the story that can affect subsequent character evaluations? The fact that our post-hoc analysis showed that men responded more favorably to the antagonist when moral priorities known to be more salient to men (Graham et al., 2011) were displayed by the character late in the story may indicate a recency/priming effect of this information. Future research could work to separate potential effects of initial positive dispositions toward antagonists and effects of recency/priming of moral violations on subsequent character evaluations. Though our current findings provide some initial steps and exploratory evidence in pursuit of these questions, further precision in manipulation is needed.

Limitations

We recognize some design limitations of our current endeavor. First, although the longitudinal design and physical laboratory testing approach of our study enabled us to analyze change over time and to have higher confidence in our data quality than an online survey would have provided, these decisions did result in a lower number of participants than might have been ideal (n = 140). Though our models showed good fit and displayed several significant relationships despite limited power, future studies would benefit from larger recruitment numbers to potentially detect smaller yet significant effects. Second, the use of a single stimulus limits our ability to know whether our findings would be robust across a variety of story contexts. Further research is needed in more varied narratives contexts to examine the breadth and consistency of these findings.

Conclusion

Overall, this work conceptualizes morally complex antagonists as a unique character type and extends the reach of existing theorizing on character engagement, demonstrating for the first time that identification with morally complex antagonists both occurs and results in changes in real-world attitudes. Further, by investigating key predictive factors and demonstrating real-world attitudinal outcomes of antagonist identification, this study exhibits the importance of examining the real-world attitudinal outcomes of fictional narratives, particularly those featuring morally complex characters. We hope that this work is only the beginning of a continued investigation into humans’ engagement with a narrative’s dark side.

Supplementary material

Supplemental data for this article can be accessed on the publisher’s website

Additional information

Notes on contributors

Rebecca Frazer

Rebecca Frazer (M.P.A., M.A., The Ohio State University) is a Ph.D. candidate in the School of Communication at The Ohio State University. Her research focuses on media narrative processing and effects, with an emphasis on moral judgments and social issues.

Emily Moyer-Gusé

Emily Moyer-Gusé (Ph.D., Communication, University of California Santa Barbara) is an associate professor in the School of Communication at The Ohio State University. Her research examines how narratives in entertainment media influence viewers’ attitudes and behaviors.

Notes

¹ Notably, Tamborini et al. (2018) propose attribution theory as an alternative explanation to moral disengagement with regards to how audiences are able to maintain positive dispositions toward morally complex characters.

² The experiment underwent expedited review and was fully approved by The Ohio State University’s institutional review board (IRB) on October 10, 2018. Participants provided consent online prior to the baseline measures. Participants who participated in the main session provided consent a second time. Consent was provided by an alteration of consent process as approved by IRB, which involved clicking a button after reading the consent form in an online survey.

³ Participants spent an average of 7 minutes, 10 seconds reading the story (SD = 2 minutes, 45 seconds). This indicates an average reading speed of 258 words per minute.